In May I wrote “The Mississippi: Buzzi’s Structural Dominance in American Cement.” Most of the piece was about how expensive cement is to move, and why Buzzi has a very good position on the river.

I still own shares in Buzzi. I want to look at the things that could go wrong, not find more reasons to justify my thesis. There were still a few things I wasn’t sure about, so I went through them one by one.

I described the Continental terminal network as inland in May, but those terminals are on the Mississippi system too. By then, Quikrete had already bought Summit Materials and taken Continental private, so some of the information I was using was already the last public data from Continental. The barge share, customer mix and sales to outside customers aren’t public anymore.

Quikrete was Buzzi’s biggest cement customer as of last year. St. Louis was the place to look. Quikrete already has cement capacity on the river there, but that only matters if it has enough of its own business nearby to use the cement itself. And a terminal on the river isn’t the same thing as having a plant next to the market.

2. On the Map

Buzzi lists 36 distribution points across the U.S. About 19 are on navigable water. The rest are rail or truck terminals. Continental has nine terminals, all on the Mississippi system.

I added Amrize to the map too. Its cement network is centered on the Mississippi and Great Lakes, including Ste. Genevieve south of St. Louis.

I expected much more overlap when I first put Buzzi and Continental on the same map. St. Louis and the Quad Cities were the two places I looked at more closely. In St. Louis, Buzzi has the Festus plant about 35 miles south, Continental has a downtown terminal supplied from Hannibal, 129 river miles upstream, and Amrize has Ste. Genevieve farther south on the Mississippi. In the Quad Cities, around Davenport and Rock Island, it flips: Continental has the Davenport plant and Buzzi has the Rock Island terminal.

Buzzi and Continental also have terminals in Memphis, a few miles apart. I leave it aside because neither company produces cement there. New Orleans looks like another overlap on the map, but I leave it out too. Continental is on the Industrial Canal and Buzzi is farther east on the Michoud/Intracoastal system. I don’t have enough on their customer areas to say how directly they compete.

3. St. Louis

Buzzi serves St. Louis from Festus, about 35 miles south of the city, plus two river terminals around the metro. Continental has its Brooklyn Street terminal downtown, supplied from Hannibal. Amrize has Ste. Genevieve farther south on the Mississippi.

Festus is downstream from St. Louis, so cement moves upstream to reach the city, with no locks in between. Hannibal has five locks on the way down. The first three have 600-foot chambers, so a full tow can’t pass through in one piece. The barges have to be split, moved through in two stages and then put back together. Lock 25 is now getting a new 1,200-foot chamber so a full tow can pass through at once.

I still don’t have enough to put a per-ton cost on that difference, and the delay will vary from trip to trip, but Hannibal has five lock passages on the way to St. Louis, including three where a full tow has to be split. Festus has none.

That makes Festus look better than Hannibal as a route into St. Louis. It doesn’t give Buzzi that advantage over everyone. Amrize’s Ste. Genevieve plant is also south of St. Louis, with barge and rail access. Amrize completed a 660,000-ton expansion there in 2025, taking annual production to about 5.5 million tons.

BMC/Breckenridge and Kienstra are two of the bigger ready-mix operators around St. Louis. BMC says it is the largest aggregates and concrete producer in the region, with more than 70 facilities across Missouri, Illinois and Arkansas.

Buzzi also has RED-E-MIX around St. Louis. It lists eight plants in Illinois, six of them around the metro.

Quikrete has Valley Park and Labadie around St. Louis, but those are dry-mix and packaged-product operations rather than ready-mix. BMC and Kienstra are independent, and I don’t see the same captive ready-mix base here that Quikrete has in Dallas. That independent demand is also there for Amrize and the other cement suppliers to compete for.

4. The Quad Cities

Buzzi and Continental ended up about six river miles apart in the Quad Cities in 2015. Davenport went to Continental/Summit and Rock Island to Buzzi, both through the Lafarge-Holcim divestitures.

They weren’t comparable assets. Davenport came with 1.2 million short tons of annual cement capacity. Rock Island was a distribution terminal with 2,000 tons of storage, and Buzzi expected around 75,000 short tons of annual shipments through it.

In May I called Rock Island the northern anchor of Buzzi’s river network. The rail and river access still matter, but Davenport was sitting six river miles away and I wasn’t giving that enough weight.

Buzzi can bring cement into Rock Island by rail or barge. In 2016, Buzzi said it was supplied by barge from Festus and by rail from Greencastle, Cape Girardeau and Festus. The terminal still handles both barges and railcars today, but Buzzi no longer says where most of the cement comes from. By river, Festus is more than 330 miles away and there are eleven locks on the way, eight of them with 600-foot chambers.

Davenport is already producing cement inside the Quad Cities. Continental also added a 125,000-ton storage dome there in 2023, which management said would reduce barge demurrage and avoid having to curtail production over the winter. Davenport is a plant and Rock Island is a terminal, so the storage numbers aren’t directly comparable. Continental can store 125,000 tons next to production, while Rock Island was built with 2,000 tons and has to be replenished from somewhere else. Buzzi has access to the market through Rock Island, but it still has to bring the cement in.

Buzzi has expanded terminals before, so I wouldn’t treat the original Rock Island shipment estimate as a hard ceiling. Dallas is the obvious example. I just haven’t seen anything similar disclosed at Rock Island, and Buzzi still describes it as a small terminal.

In the 2022 10-K, Summit said around 70–80% of its cement moved by barge because it was cheaper than truck or rail. It had been around 70% in 2017 and 63% in 2019. Intercompany sales were also immaterial across West, East and Cement, and the segment numbers were based on external customers. It isn’t a clean Continental-only figure, but most of the cement business was still going to outside customers.

After Quikrete bought Summit in February 2025, most of that commercial disclosure disappeared. Hannibal and Davenport are still easy enough to track through permits and regulatory filings. What I can’t see anymore is how much cement moves by barge, how much stays inside Quikrete, or what the economics look like.

Share

Subscribe now

5. Quikrete

In March 2025, Pietro said Quikrete was Buzzi’s largest cement customer in the US. He called it “number one.” Later in the same Q&A someone tried to pin him down on whether it was 5% or 6% of US sales, but he wouldn’t give the number.

By then Quikrete had already bought Summit, which included the former Argos USA business. Argos had 140 ready-mix plants, and in 2022 it supplied 77% of the cement those plants used from its own production. But those internal sales were only 19% of Argos’s total cement volumes, so most of the cement business was still going to outside customers.

With Midlothian, Quikrete got the cement plant, the terminals and the North Texas ready-mix business in the same deal. Martin Marietta put the assets at around 2.1 million tons of cement and 3.6 million cubic yards of concrete a year, with more than 40 ready-mix operations across DFW and nearby markets. Martin Marietta already treated Midlothian as a vertically integrated cement asset. In 2024, its CEO described a strategic cement market as one with a sizeable ready-mix business downstream taking a significant share of the cement, and Midlothian was one of the examples he gave.

Argos’s disclosures give a rough number for the cement content. They work out to around 520 pounds of cement per cubic yard of concrete. Applied to the 3.6 million cubic yards that came with Midlothian, that is around 0.9 million short tons of cement, or around 40% of the assets’ 2.1 million tons of cement capacity. That doesn’t mean 0.9 million tons of capacity was simply free to replace outside purchases. The actual self-supply rate could be different because concrete mixes vary and we don’t know where every plant bought its cement. I’m only using the 0.9 million tons as a scale check.

After the Midlothian deal was announced in August 2025, Pietro said Quikrete would buy from itself where the logistics worked and keep buying from other producers where they didn’t. In March 2026 he said the same thing again: if a ready-mix plant was too far from one of Quikrete’s cement plants, it still had to buy elsewhere. By August 2026 he was talking specifically about Dallas, with Midlothian now another source of self-supply for Quikrete. South Texas was a different issue, with more pressure from imports through Houston, while Midwest pricing was improving.

There is another piece of this in Dallas. Amrize is expanding its own Midlothian cement plant and bought Rapid Redi-Mix in July 2026. Amrize says the ready-mix business has synergies with its cement and aggregates network and complements the Midlothian expansion. I don’t know where Rapid Redi-Mix was buying its cement before, so I wouldn’t count that as lost volume for Buzzi. But the more I look at Dallas, the less this looks like only a Quikrete self-supply issue. More of the market is being tied back to producers with their own cement.

Cement is a high fixed-cost business, so losing even some of that volume matters for Buzzi. Pietro has also said more ready-mix integration could make sense mainly to protect cement volumes, “more as a defensive move than something else.”

Around St. Louis, I don’t see a Quikrete ready-mix network comparable to Dallas. Buzzi has RED-E-MIX there, while BMC/Breckenridge and Kienstra are independent ready-mix operators. That isn’t enough to say anything precise about market share, but Hannibal still has those independent producers to compete for.

6. What I Think Now

I think the Buzzi thesis is simpler than I first made it. Buzzi has good assets and some strong local positions, but Amrize and Quikrete have very good assets too. Some of Buzzi’s advantages may be weaker than I thought, but the shares still look cheap to me.

I was too broad in May when I called this a Mississippi moat. The river still matters a lot, but I was giving it too much credit, especially in the Quad Cities where Davenport is producing inside the market and Buzzi is coming in through Rock Island.

Quikrete still matters, but I don’t think Dallas tells me much about St. Louis by itself.

A lot of what I like about Buzzi’s position in St. Louis comes from having Festus only about 35 miles from the city. In 2024 its single kiln ran for 6,115 hours, versus 7,463 at Hannibal and 7,576 at Davenport. That is like two months less, and I don’t know why. Planned shutdowns are normal, and the EPA data doesn’t tell me whether this was maintenance, weaker demand or something operational. One year is not enough to draw much from it, but I want to see whether the gap shows up again. Buzzi is also adding grinding capacity at Festus, although the plant still depends on that one kiln.

For St. Louis, I’d focus much more on what happens locally than on the US margin. The US numbers mix too many things together. I’d pay more attention to whether the extra capacity at Ste. Genevieve starts showing up in the market, Quikrete adding ready-mix around St. Louis, Buzzi talking about self-supply pressure in the Midwest, or Continental taking more of the independent business.

St. Louis still isn’t the reason I own Buzzi. It’s only one market, but it was one of the US markets I felt best about in May. I still like the position. I just don’t need as much from it as I thought I did in May.

Disclosure: I hold a long position in Buzzi SpA (BZUMI) as of August 2026. This is not investment advice.

Cement volumes were up 1.6%, but EBITDA fell 15.3% at constant currency. Most of the pricing pressure was in Texas, and that’s also where the Quikrete self-supply issue is showing up. I haven’t seen the same thing around Festus and Cape Girardeau. Buzzi doesn’t report the corridor separately, so I can’t say exactly what happened there.

I had Buzzi at €57 in March, so I went through the valuation again. I also found changed how I valued Mexico.

The group made €482.7m of EBITDA in H1. Here is the contry split:

2. The US business

US cement volumes were up 1.6%. Ready-mix was down 1.5%, mostly in Texas.

FX makes the decline look worse. The dollar took €49.6 million out of sales and €12.6 million out of EBITDA.

EBITDA fell 15.3% at constant currency while sales were down 0.6%. Group transport costs were up 4.1%, personnel 4.5% and fuel 9.8%. Electricity was down 15.6%, so total raw-material costs were basically flat.

Buzzi talked about fuel as a US headwind, while lower electricity costs are helping Europe.

On Dallas, Pietro said: “in Dallas, it’s not a new player, but in a sense it’s a new player because when Quikrete took over the Midlothian plant, it’s another potential point of self-supply versus maybe buying from other competitors. Which again, is somehow translating into some conflict, let’s say, between producer.”

Quikrete buying less from other producers had already come up on the last two calls. This time Pietro brought up Midlothian himself and linked it to self-supply in Dallas.

I know St. Louis and the Quad Cities much better than Dallas, so there’s still some work to do there and I don’t know yet how much merchant demand Quikrete can take out.

In Houston the issue has been imports for longer, already before Quikrete bought Midlothian.

Pietro put the pricing pressure “mainly in Texas,” while “in other parts, in other states, we had some improvements.” The Midwest, where the corridor actually is, isn’t where this pressure is coming from.

Festus and Cape Girardeau are around 40% of Buzzi’s US cement capacity. Texas is closer to a quarter, but more of the downstream business sits there. Buzzi doesn't give EBITDA by region, so I can't say how much each one actually contributes.

US CapEx in H1 was €117.5m, versus €127.4m last year. That number is for the whole US business though, so it doesn’t tell me how much Buzzi is spending in Texas.

Phase 1 is approved and Buzzi is budgeting around €70m this year. Pietro said it is only starting, so the impact this year should be relatively minor. San Antonio is very likely to start a new rail terminal, and the Beaumont import terminal is already underway. Beaumont isn’t necessarily just for cement; Pietro said it could handle other materials too.

The US price index is at 138, with FY2021 at 100. It was around 141 in 2023 and 2024. Italy is at 187, Czech Republic at 169 and Germany at 167. There isn't a regional split behind that number.

Festus and Cape Girardeau still have the barge terminals and the same distribution network I looked at in the original thesis. US pricing hasn’t done much over the last two years.

Cheap distribution was a big part of the corridor thesis. The transport increase I mentioned above is group-wide, so I can't isolate the US from it. Acquisitions added transport costs too, and the weaker dollar reduced the US number in euros. I still can't tell from that whether the economics of Festus and Cape Girardeau have changed.

3. Italy and Germany

Italy's margin was up 280bps in H1. Germany was down about 300bps. Cement and ready-mix volumes were both down around 6% in Italy. In Germany, cement was down 3.5% and ready-mix 6.9%.

Italy reported €93.3m of recurring EBITDA, from €84.3m last year, up 10.7%. Pietro said that without the power cost benefit from the Energy Release program “you come to a level which is very similar to last year.”

Half of this year’s allocation is already in H1, together with a non-recurring accrual related to the 2025 tranche.

Germany has low capacity utilization, which Pietro mentioned when discussing margins. He also talked about vertical integration: “the vertical integration of the producer is not as significant as we have in Italy. You have a number of customers in ready mix that can correctly play a game on prices by addressing offers from different producers.”

This is similar to what I wrote about RED-E-MIX in the US. Buzzi owns much more of the ready-mix network around St. Louis, so there are fewer independent customers choosing between different cement producers. Germany has much more of that.

SigmaRoc reported German infrastructure demand recovering since Q1, with rail, road and energy projects driving volumes. Buzzi isn't seeing the same thing in cement yet. Pietro was asked again about the federal infrastructure plan and when it could start showing up in demand. His answer was basically the same as two calls ago: “Will it become visible? Will it not?”

Cement may be later in the cycle than aggregates and lime, so maybe Buzzi hasn't seen those infrastructure volumes yet. Low utilization would still be there though, and German ready-mix customers can buy from different producers.

4. The rest of the portfolio

Take Brazil out of that group and the other four countries net to +€2.7m. That's basically flat.

Brazil reached a 27.5% margin in H1, 570bps higher than last year. Cement volumes were up 3.2% and prices increased much more. Pietro talked about high capacity utilisation, particularly in the Northeast, and said the Southeast is catching up.

Distribution in Brazil is all by truck. There is no river or rail network like in the US. Diesel costs were higher during the half and the margin still went from 21.8% to 27.5%.

Mexico isn’t on the screenshot because Buzzi accounts for the JV through equity earnings. That was €62.8m in H1, compared with €49.2m last year. For the full JV, cement volumes were up 12.7%, EBITDA reached €264.9m and margin was 46.3%.

Pietro also mentioned USMCA, the trade agreement between the US, Mexico and Canada, and called it a critical path for Mexico.

UAE reported €10.5m of EBITDA compared with €2.9m last year, but Buzzi only started consolidating the business in May 2025. On a pro-forma basis, last year’s EBITDA was €4.8m. This year it reached €10.5m, with cement volumes down 15.7%.

Buzzi reduced low-margin clinker exports and sold more into the domestic market, where prices are higher. Margin went from around 7% to 13.9%.

Luxembourg and the Netherlands lost €6.5m of EBITDA compared with last year. Cement volumes were broadly flat and ready-mix volumes in the Netherlands fell 8.2%.

Poland had cement volumes down 10.6% and ready-mix down 13.4%. EBITDA was down 6.3%. In Czech, binders grew 5.5%, ready-mix 10.9% and EBITDA 10%, with construction activity around Prague still strong.

I’m leaving margins out here because the regional margin Buzzi reports also includes Russia.

Read more

I continue to use the June thesis as a basis for certain aspects: approximately $120 million for the standard wood products division, the recovery of the real estate market between 2028 and 2030, real estate values per acre, and the difference compared to the values of private forest land.

The quarter

Adjusted EBITDA was $123.7 million, against $94.1 million in Q1 and $44.9 million a year ago. Q2 last year was still legacy Rayonier, so the year-on-year comparison doesn't tell you much.

Wood products reported adjusted EBITDA of $25.0 million, its best quarter since PotlatchDeltic’s third quarter of 2022. The Northwest division reported $26.3 million, compared to $8.6 million in the first quarter.

Southern also improved. The real estate segment fell short of the first quarter’s results but still exceeded the range management had indicated in May.

Guidance barely moved. Real Estate was unchanged, southern harvest kept the same midpoint, and Northwest was trimmed a little. Q2 is just a stronger seasonal quarter for timber, with lower volumes guided in H2.

The segment numbers were fine. I spent more time on the balance sheet, merger costs and purchase accounting.

The same acre can be worth very different amounts depending on what Rayonier does with it:

Rural HBU and the solar parcel were Q2 sales. RMS closed in August, and the $25,000+ figure is a company target net of development costs. Different things, but enough to show how much the use of the acre matters.

RMS

Both deals were with RMS and structured as a like-kind exchange. McHugh said that made it more tax-efficient and gave them more flexibility than just buying the new acreage.

Rayonier sold about 36,000 acres in southwest Washington for $145 million and bought about 57,000 acres in Alabama and Texas for $146 million.

On $/acre that’s roughly $4,000 sold and $2,600 bought. Those numbers look far apart, but the acres don’t produce the same timber economics.

I’m using 2025 here because the 2026 numbers already mix in PotlatchDeltic. Northwest has Idaho now, and the Southern numbers cover a much larger portfolio.

On 2025 numbers, I get about 31x current timber cash on what they sold and 34x on what they bought, around a 3.3% yield in Washington and 3.0% in Alabama and Texas.

On timber cash, the difference is much smaller. This only includes current stumpage cash, not timber growth or land appreciation.

My 31x and 34x come from segment averages. Rayonier has the actual parcels and puts the exchange at about $3 million a year of incremental timber Adjusted EBITDA over the next ten years, before HBU or land-based solutions.

There are some obvious differences my averages miss. Northwest has higher operating costs, the acquired acres are 69% plantable with a site index of 75, and McHugh said the Washington disposals have been reducing Hemlock and improving the age-class mix.

Share

Read more

I’ve been following FRPH for years, and I don’t remember any of its earnings calls generating this much reaction on X. Four shareholders spoke very clearly to management about capital allocation, including one who introduced himself as the fourth-largest outside shareholder and asked the CEO to take a 20% pay cut, among other things. By Thursday, there were already posts about the call.

Most of what I saw being discussed came from the call, at that time the 10-Q wasn’t out yet. It came the next morning, and some of the numbers in there matter for the discussion that followed.

Some of the criticisms may change after reviewing the documents. For now, there is one point (which I will discuss later) that is repeated in a way that I don’t think is correct, although I agree with the conclusion.

The other part is the part I always do, which is check my own work. In December I published three things I said I’d watch through 2026. We have the first half of the year now, so I went back to them.

Checkpoints

I had $15-$16 million for 2026 on SG&A. Q2 was $3.69 million without the one-time costs from the Altman closing, and McNulty said on the call that this is the level we should expect going forward. That would put 2026 at around $14.7 million, below my range.

I had Chelsea at 70% occupancy by the end of this year when I wrote about it in December. It’s been 15 months since was completed, and is still 100% vacant. If Chelsea still vacant at that date would break that part of the thesis.

I had DC occupancy at 88% in December and thought that was conservative. DeVilliers gave us the number on the call: about 8% of tenants aren’t paying. Against 93.2% physical occupancy, economic occupancy is closer to 85.7%.

DeVilliers said economic occupancy improved about 75 basis points this quarter. When he was asked when the environment changes, his answer was, “I don’t see it changing anytime soon.”

G&A came in better than I expected. DC is taking longer, and Chelsea hasn’t moved.

Industrial

Revenue in the segment was $983,000 down 28.5%. Cost of operations was $986,000, so the segment was already losing money before G&A. Last year it made $443,000 profit.

Levy said users looking for more than 250,000 square feet are choosing Pennsylvania and New Jersey:“a lot of those sites in New Jersey have economic incentives associated with them, i.e. PILOTs and other job creation incentives, which further create a delta between Chelsea and those opportunities.”

Some of the difference is simply economic. New Jersey can offer PILOTs and other job creation incentives that Chelsea doesn’t have. That gap may be harder to close than the current supply problem.

Levy described Cranberry differently. “The local tenant pool is not very deep,” and he said Cranberry isn’t going to attract a tenant from outside the market. Some of the smaller tenants there depended on larger tenants that have since left. He also mentioned lost government tenants and one bankruptcy that ended in an eviction.

Of the 408,000 sf available, FRP signed 20,700 square feet and had another 97,500 under active negotiation. About 15,000 sf was leased but not occupied, almost all at Cranberry. From around 800,000 sf at the end of 2025, is expected to reach approximately 2.1 million sf in Q1 2027. Much of that new space will likely be added before the current vacancy is filled, so occupancy will probably come down.

Mining

Revenue after the add-backs, depletion and unrealized revenue we get $4.12 million NOI. Without this segment, the guidance would have been $1.8 million cut not $900,000.

Growth eased a little from Q1, both volume and price, but this line can move around quite a bit. There have been years where tons fell and pricing still went up double digits. Q3 will look worse because last year had a one-time $1.85 million minimum royalty catch-up.

Depletion, the cost assigned to the tons being mined, was up 53% in the quarter while volume was up 6.8%, which puts depletion per ton up about 30%. I couldn’t find anything in the filings explaining the increase. It gets added back to NOI, so normally I wouldn’t care much about it. Here I do, because depletion is based on estimated deposits and we don’t get much else in the filings on reserves.

One of the analysts asked about the bulkhead lease at Square 664 and mentioned that he thought the lease was expiring around this time. DeVilliers confirmed they are discussing it with the tenant, who has been there a long time, and said he expects them to stay “until we’re ready to develop, or until it becomes a nuisance to our adjacent properties.” Then asked whether a renewal would improve the economics, but deVilliers didn’t give any detail on that.

DC Multifamily

NOI was $4.32 million versus $4.74 million last year, down 8.9%, and occupancy was 93.2% versus 94.1%. Compared with Q1 though, occupancy improved.

Dock 79 occupancy was up 400 basis points and Maren 290, while renewals also improved, from 63.6% to 71.4% at Dock and from 55.6% to 71.1% at Maren. Q2 is a stronger leasing quarter, so some of that is seasonal, but it was still a better quarter for both buildings. Bryant Street and The Verge, barely moved.

NOI was down $421,000, with $277,000 of that coming from operations. Unrealized rents went from negative $31,000 to negative $149,000, and the 10-Q ties that to “lower occupancy, higher bad debts, and increased costs at Bryant Street and The Verge.” Those are also the two properties where occupancy didn’t improve this quarter, while the improvement we saw above came from Dock 79 and Maren.

DeVilliers put a number on this for the whole portfolio. Renewals are going out above 1.5%. New lease is about 10% below than the previous. Retention is above 50%, but there’s still a important amount being re-leased at lower rents.

Same with the 8% of DC tenants that aren’t paying, DeVilliers said the courts are backed up and it can take twelve to eighteen months to get a non-paying tenant out. That 8% is for the whole portfolio. Baker also mentioned cases of people signing leases who never planned to pay in the first place.

One analyst questioned whether DC was the problem. He owns Camden, which has around 60,000 units and gets 13% of NOI from DC and Northern Virginia, he said that in years of meetings with the company he had never heard them mention delinquency there. Baker agreed that Camden is doing something they aren’t and said tenant screening needs to improve. Levy said they’ve been working on that with the property manager for about a year and a half.

Another analyst asked about a recent deal across from Bryant Street. DeVilliers said it came up in Bryant Street’s refinancing. They’re in the middle of the refinancing now and the appraiser used the Rowan as a comp, a Trammell Crow building right across from Bryant Street that sold at a 6% cap rate.

In my valuation I had Bryant Street at a 5.50% cap rate, using FY2025 NOI of $6.098 million. That gave me $31.56 million for FRP’s share of the equity after debt. The Rowan sold 50 basis points above the cap rate I used. If I use 6%, it takes about $0.48 per share off my valuation.

I wouldn’t change the valuation based on one transaction, but is a useful comp. It’s the appraiser working on Bryant Street’s refinancing is using. We’ll know more once the refinancing is done.

Read more

Revenue came in at $584.3 million, compared with guidance of $570-582 million. Adjusted EBITDA (with one-off costs like acquisition and integration expenses out) was $94.8 million, compared with guidance of $90-96 million. Adjusted gross margin rose 135 basis points year over year, and adjusted EBITDA margin rose 40 basis points.

Some headlines described the results as “below expectations,” using a consensus EPS of $0.36. TIC guided on revenue and EBITDA, not EPS, and I can’t quite reconcile the $0.36 with what TIC reported.

In the Q1 update I wrote that net debt moving the wrong way for one quarter is a data point, two consecutive quarters is a trend. Net debt was $1,250.3 million at June 30, from $1,182 million in March and $1,173.7 million in December.

1. Checklist

TIC completed three acquisitions this quarter at a multiple of between 5x and 7x EBITDA, within the range management indicated in July. The leverage ratio, however, increased during the quarter. Integration costs fell from $14.1 million in the first quarter to $8.7 million in the second, but remained above the $4 million to $5 million range projected in May. The conference call provided little insight into this discrepancy.

Net debt and the minimum organic growth rate will be discussed later in this article. The Gulf Coast, data centers, and synergies have not changed significantly since May. GEO Agent remains difficult to value, as the company has not yet provided any financial figures regarding it.

2. Margin

I&M was down a little, C&E barely moved and GEO jumped. GEO is only 14% of revenue. I&M went from about 55% of revenue to 51%, while C&E went from 31% to 35%. C&E runs close to 47% gross margin, so the change in mix helped even though the margins inside I&M and C&E barely moved.

Adjusted gross profit was up $14.8 million. Only $5.3 million of that made it through to adjusted EBITDA. The difference is $9.5 million. Adjusted SG&A was $129 million, about $9 million higher than a year ago, and rose from 21.2% to 22.1% of revenue. Schultes attributed the increase to incentive compensation, indirect labor, legal reserves, benefit costs and overhead from the acquired businesses, partly offset by synergy savings.

And this isn’t the $8.7 million in integration costs, which gets added back before adjusted EBITDA. Some of the costs associated with the acquired companies still exist, and synergies have not yet offset them. As for incentive pay, results are better this year, so bonuses are higher.

Schultes said H2 already has a little over 100 basis points of margin coming from mix and execution. Q2 had 135 basis points at gross margin, and less than a third made it through to EBITDA. H2 assumes more of that makes it through. Some of this year’s higher bonus expense is already in the numbers.

3. Half the floor

Looking at the table below, we start with revenue up 86% year-over-year. If we exclude NV5, growth drops to 3.3%, and if we subtract this year's acquisitions, it comes to 2.5%.

In May, management set a target of around 5% organic growth, excluding synergies. Organic growth was 2.3% in the first half of the year and 2.5% in the second quarter.

I&M revenue fell by 5.5%. Heraud said that, excluding scheduled outages and facility losses that had already occurred in 2025, I&M would have grown by 4%. CFO Schultes quantified that adjustment: $30 million, comprising those 2025 losses and plant shutdown work that was shifted out of the quarter. C&E grew by 16.8% and GEO by 7.9%.

June was a positive month for I&M compared to Q2 2025, the first time in a while, Heraud said. New project awards were concentrated on the Gulf Coast, the same region that Sullivan noted in May as underperforming relative to its share. Technical rates have increased, and rope access work grew by nearly 10% during the quarter. TIC was also was awarded a multi-year LNG contract worth $30 million that Heraud described as “fresh off the press,” meaning it came too late to have an impact on the second quarter.

Whether that turns into revenue fast enough is a different question, and the backlog is where I'd look for the answer.

Read more

NAV that comes out of this quarter is lower than what I published in July, so before getting into the numbers I think it’s worth explaining why.

The March revision was straightforward: I used an old balance sheet and overstated cash by about $40 a share. In July I understood the Bloomberg mechanism, but I was looking at the wrong line when I applied it.

Nothing in the filing changed the framework from July. What changed is that I had three modelling decisions I'd left open, and I don't think I can leave them open anymore. I'd rather explain those than just publish another NAV.

731 Lexington retail goes in at zero. The C-Note sits between Alexander's and the asset: the first $171.52M of value belongs to the lender, only after that does any value come back to Alexander's.

This time I took total liabilities straight off the audited balance sheet: $979.44M. That already includes the full Lease incentive liability, $93.72M at June 30, which is where the Bloomberg tenant fund is recorded. I spent a lot of time trying to separate the tenant fund from the rest of that liability. Eventually I stopped rebuilding it and used the audited number instead.

The cover page isn't the denominator I should have been using. Alexander's weighted average share count already includes the 32,201 deferred stock units because they receive dividends. That takes the total to 5,139,491.

None of these changes altered the business itself. At $262, the stock trades at a discount. At $285, the same model would have called it fully valued.

Quarter, Line by Line

Rental revenue for the quarter was $54.7M, up from $51.6M a year ago. NOI rose 14.5% to $28.0M. FFO came in at $3.02 a share versus $2.88. Looking at those numbers, you wouldn't think Alexander's was halfway through the Bloomberg abatement.

The Lease incentive liability on Bloomberg’s tenant fund fell from $113.62M to $93.72M this quarter, a drop of exactly $19.896M. Divide that by the three months of abatement and you get $6.63M a month. That’s the same number, to the dollar, as the fixed rent in the 2024 lease amendment. Bloomberg isn’t paying rent right now. Alexander’s is settling a tenant fund obligation it already owed with rent it isn’t collecting. That's why the hit doesn't show up in rental revenue. It reduces a liability instead.

Back in July, before the 10-Q existed, I wrote that this was how the mechanism worked. Where I got it wrong was the accounting line. I expected reported NOI to fall 81%. It rose 14.5% instead because I was tracking one side of the entry and not the other.

The abatement does show up somewhere else: operating cash flow. Cash from operations for the first half was $13.05M, down from $59.28M a year ago, against $46.2M of dividends paid over the same period. That’s 28% coverage. FFO was $28.90M, or 63% of the dividend. Neither covers the dividend. I’ll come back to whether that gap matters for the dividend later, because it needs more than one paragraph. It's not just this quarter, the dividend has been running ahead of recurring cash generation for a while.

One thing changes next quarter without the business doing anything differently: interest income. It fell 60% in H1, $7.87M down to $3.13M, and that looks worse than it is. The $303.31M sitting in the bank today didn't arrive until May 28, when Rego Park I closed. For most of H1, Alexander's was holding closer to $92M. A big cash pile that only exists for the last month of a six-month period barely has time to earn anything. Hold that same $303.31M for a full quarter instead of one month, and interest income should come in around $3.0M, around 80% more than Q2 produced. That closes about 17% of the gap. Nothing changes operationally. The cash just has a full quarter to earn interest.

The 10-Q never names the outgoing tenant. Matching the square footage and lease expiry back to the 2025 10-K leaves one candidate: Kohl's. That wasn’t in any model I’d published before.

Read more

The first slide reports both GAAP and adjusted EPS. GAAP diluted EPS was unchanged from last year at $5.17, while adjusted EPS increased 4% to $5.38 because of CEO transition costs.

GAAP is the standardized figure, but adjusted is the cleaner comparison here because last year’s quarter didn’t include those costs.

Almost all of the adjustment comes from $6.3m of accelerated share-based compensation. The $6.3m would normally have been recognized over the next few years, but it is all booked this quarter even though only $2.0m is cash.

Gross Margin

Gross profit came in at $540.8m, up $5.6m, a 1% increase.

The release gives you the 30bps and the $540.8m, but not what those 30bps mean in dollars or operating income.

Adjusted operating income increased by $3.2m, from $272.7m to $275.9m. Both years are comparable because 2025 didn't include CEO transition costs. Opex also grew 1% while sales increased 2%.

Adjusted opex was 14.5% of sales versus 14.7% last year. Higher Q2 sales usually improve opex leverage by spreading fixed costs over a larger revenue base.

Last year's gross margin was 30.0%. On this quarter's sales that would've been about $546.9m of gross profit instead of $540.8m, $6.1m more. Without the margin drop, adjusted operating income would've been around $282m instead of $275.9m. That would've been roughly $9.3m above last year, or 3.4% growth instead of 1%.

The bridge attributes the change to supply chain, inbound freight and customer mix, but provides no numbers.

On the Q4 call, Melanie Hart said supply chain, pricing and private label would offset the shift towards larger customers.

Adjusted operating margin in that quarter was already down 80bps, from 6.1% to 5.3%. Q1 improved slightly to 7.3% from 7.2%. By Q2, customer mix was a negative on the bridge, supply chain was the only positive contributor and inbound freight had become a headwind.

Wells Fargo asked Melanie Hart to quantify the split between freight, customer mix and the shift toward larger customers. The only number they got was a ranking: "inbound freight is by far the most significant component."

Stifel asked about Pentair and whether customer consolidation was starting to pressure margins. Melanie said: "No change in our guidance that is reflective of anything that Pentair has reported." Bigger customers usually mean tougher pricing.

The disclosures don't quantify the split between freight and customer mix.

Freight

Watwood described freight as something that “jumped on us pretty hot and heavy.” By Q2, the issue wasn't outbound freight anymore. That had already been recovered through surcharges. What remained was inbound freight, which sits inside cost of goods sold.

Recovering that means higher product prices. Those price increases would mainly fall on the same larger accounts already driving the mix shift. Freight and customer mix appear as separate drivers on the bridge, but they're connected through the same customer base.

Price recovery starts in Q3, but Melanie cautioned that "we may not be able to recoup all of it immediately." BofA asked whether any recovery was already reflected in guidance. "We don't have anything material included in the guidance at this time."

My first read was that the guidance was mostly a Q2 effect. Q2 is the biggest quarter of the year, so one weak quarter drags down the full-year average. Listening to Melanie again, I don't think that's what she's saying.

I went back to last year's margins. Q3 2025 was 29.6%. Q4 was 30.1%. Take another 30bps off both and you’re around 29.3% and 29.8%. That’s not company guidance, it’s my way of translating what Melanie said into something I can check once Q3 comes out.

Price contributed 3% in Q1 and another 3% in Q2. In H2 it's closer to 2%, maybe slightly below, because they're lapping last year's price increases. That's exactly when they also need pricing to do more work recovering freight. The problem is pricing is doing less work than it was six months ago.

Stifel saw the same thing and asked it, whether the H2 margin guide still holds up once they start lapping last year's price increases. His words: “That seems difficult.” He'd followed it with a question about Pentair. The answer covered Pentair and inventory, but never really came back to the first question. He didn’t press. Nobody else picked it up either.

Last thing. Melanie’s on what’s driving the increases: “heavier density items such as the building materials.” Building materials is one of the better-performing parts of the business right now. It's also one of the most expensive categories to transport.

Building materials carries more freight than the average sale. So as that category becomes a bigger part of the mix, freight becomes a bigger drag too. I’d been treating that share gain as clean upside.

Customer Mix

Heritage was the competitive risk I kept coming back to in December, especially in Florida, Texas, California and Arizona. Three of the four turned negative this quarter. Florida stayed close to flat, but Horizon and soft permits still explain that better than competition does. I don't see much pointing to Heritage yet.

Heritage still showed up, just not there. Consolidation doesn't pull volume out of a state. The pools still get serviced every week no matter who owns the route. It shows up in pricing instead. That's where I should've been looking.

I went back through the 10-Ks to see how long this had been on their radar. The risk factors talk about regional distributors and mass retailers, but I couldn't find any discussion of customer consolidation. That's why I went back to the roll-up discussion on the call.

No way to know how much of this is Melanie's cyclical explanation and how much is customer consolidation.

The maintenance floor is a revenue floor, not a margin floor. Maintenance was 64% of pool product sales in 2025, so the revenue is still there regardless of what new construction does. Whether that revenue comes through 500 small accounts or 50 large ones changes the economics. That's the part I wasn't thinking enough about. POOL's disclosures never really answered it. I ended up spending more time looking at the software those larger contractors were using.

What Worked

Building materials grew 4%, slightly below Q1's 5%, while permits are still down low single digits YTD.

Europe looked weaker to me at first. Headline growth slowed from +16% to +11%, so that was my first read. Then I looked at local currency and went from +5% to +9%. The FX tailwind was just much smaller this quarter.

Watwood blamed the heat and people staying home instead of taking holidays. He also made a point of saying "mini-COVID" wasn't his phrase. It was something he'd been hearing from people he'd spoken to.

POOL360 reached another record at 18% of sales.

Q3 is usually the strongest digital quarter of the year. That’s when the platform gets used the most. Q2 this year already came in above last year’s Q3 peak.

Technology and POOL360 were one of Pete Arvan’s three priorities for 2026 back in February: “fully leveraging our technology and network investments... driving greater efficiency, reach, and agility.” They don’t appear in Watwood’s four priorities anymore. The topic didn’t come up until someone asked about it during the Q&A, and most of the answer focused on returns from investments already made.

The value of the platform comes from getting deeper into customer workflows before someone else does. SPS PoolCare is already trying to do exactly that.

POOL360 accounted for 15% of annual sales. I couldn't tell whether pros were actually running their businesses on it or simply using it to place orders. Has risen to 18%, but that question is still open.

On March 17th, SPS PoolCare rolled out ServiceTitan across more than 30 branches in five states for routing, chemical tracking and project management. It’s the same company that bought Pool Troopers in January.

From there I went to Skimmer. More than 35,000 pool service professionals run their business through it. Heritage was the only distributor built into the platform. Plenty of those users probably still buy from POOL. They just have to step outside the software to do it.

Skimmer is also much bigger than just the roll-ups. Plenty of smaller operators use it too. Even a small amount of friction adds up differently across 35,000 users than it does across one acquisitive platform.

None of this changes a state’s sales overnight. Over time it can change where the next order gets placed, and that’s a margin question before it becomes a volume question.

Watsco publishes one number I wish POOL did. Their digital customers churn about 60% less than everyone else. POOL tells you 18% of sales go through POOL360. It doesn't tell you what share of that comes from the small operators this platform was clearly built for, versus the larger accounts increasingly running their business somewhere else.

Cash and Capital Allocation

Back in April I kept coming back to the $1.7bn early-buy inventory. It had to clear through peak season at the expected prices. Otherwise the margin would tell you it hadn't worked.

Inventory went from $1,660.8m at the end of Q1 to $1,378.7m at the end of Q2, a $282.1m drawdown in three months. Last year over the same period it fell $130.5m. That's a much bigger unwind than last year, which is what I'd expect if the early-buy inventory was selling through.

Inventory turns were running at 2.6x over the trailing four quarters at the end of Q1, down from 2.8x a year earlier. That's not a new low though. They were already at 2.6x back in 2022. I don't have the Q2 number yet because the 10-Q isn't out.

Back in February they bought ahead of supplier price increases. Management's explanation was simple: it "positions us to protect and expand our gross margins."

Sales don't answer that. The reserve gets you closer. It stood at $24.1m at June 30, down from $27.7m a year earlier, even though inventory itself was 4% higher (they only publish the dollars). The percentage is mine, 1.75% now against 2.08% last year. The reserve came down even though inventory was higher. That's the direction I'd expect if inventory was moving normally. It's a small number and management controls it, so I don't read too much into it. Still, it’s probably the best evidence I’ve found that they sold through the early-buy inventory without much discounting. Gross margin still fell 30bps, so they didn’t deliver the expansion they talked about in February. But the bridge points to freight and mix, not discounting. The early buy doesn't look like the problem. It wasn't enough to offset freight and mix.

Pentair's comments in April also make more sense now. POOL took $282. 1m out of inventory this quarter, meaning they bought a lot less from suppliers than they sold to customers. Pentair represents about a fifth of POOL's purchases, so part of what Pentair described as channel destocking was simply POOL placing fewer orders while it worked through inventory.

POOL's sales still grew 2%. Purchases from suppliers fell much more sharply. That's what inventory drawdowns look like. Customers still buy the product, POOL just doesn't need to replace everything it sells.

Back then I wrote about how cheap that early buy was. Payables had risen $119m alongside the inventory build, so vendors were funding most of it. Net cash was around $28m. I called it the purchasing power of a distributor nobody can afford to lose. I still think that's right.

I hadn't looked at payables again since then. They've mostly reversed. Payables are down to $474.5m from $529.3m, and cash is down to $28.8m from $83.7m.

Operating cash flow barely moved, minus $0.7m for the half against minus $1.5m last year. Inventory released $72.5m of cash. Payables used more than double that going the other way. Working capital was still a net use of cash, even with inventory doing its part.

Q2 is usually the peak debt quarter anyway. Cash usually comes back through H2. Melanie said as much, and I don't think that's changed. Seasonality explains why Q2 is above Q1. It doesn’t explain why leverage has gone 1.47x, then 1.67x, then 1.73x, now 1.78x. Or why even the year-end low keeps moving higher. In April I was watching inventory. Wrong line to watch, turns out, more on that below.

From here I get into the buyback, the balance sheet, the 2028 math, and whether any of it changes my thesis. Thanks for reading this far.

Read more

Eurofins published H1 2026. Before the numbers, there's a line from April I want to check against what actually happened. I wrote: “Q2 2026 is the first quarter where that explanation is no longer available.”

I went back and reread that before opening the Q2 print. No hedge built into it back in April. Just the sentence, saying the excuse runs out in July.

Q2 organic growth: 3.0%. H1 as a whole: 2.7%.

That’s the number the rest of this piece has to work around. A full point above the 2.6% in Q1. Still short of what the annual target needs. Nowhere near the 6.5% long-term number the company keeps repeating, including in this release.

There’s something else in the guidance line, easy to miss if you only read the growth number. In January it was flat, mid-single-digit for FY2026, and it hadn’t moved by April, not even after the weather quarter. This time there’s a sentence attached to it that wasn’t there before:

“Organic growth will depend on the speed of pick-up of business in certain ancillary activities, and the end of contraction in others.”

They didn’t touch the number, they added a condition to it.

I think that condition has a name, BioPharma.

BioPharma

The explanation the release gives for part of the softness in Bioanalytical Laboratories is that a lab moved buildings in North America sometime in H1. “disruption around a site move in H1.” The release doesn’t say how much of the quarter that accounts for on its own. It’s a small detail. It’s sitting inside a much bigger number.

Organic growth, Q2 2026: -1.1%. First negative quarter in a series I’ve tracked since Q1 2025: +1.5%, +0.4%, +3.7%, +1.4%, +1.1%, and now this.

H1 2026 came in at -0.1% organic. Reported revenue rose 3.0%, acquisitions doing most of that work.

The release does mention the weak parts. Europe, “slightly down on the prior year,” a “mixed picture,” CDMO “declining.” North America, BPT “stayed solid” but ancillary activities “remained soft.” It’s all there if you read closely.

What it doesn’t really convey is how fast this moved. BioPharma went from +1.1% organic in Q1 to -1.1% in Q2. Two points, one quarter. Both regional write-ups stay pretty calm about it, and most of that calm is because they’re describing BPT, the core piece. The rest of the area is where the -1.1% actually lives.

An analyst from Jefferies asked about the same thing. He wanted an actual number, how much of the BioPharma deterioration came from contract exits and how much came from something else, but he didn't get it. Martin gave a few examples instead. Phase I clinics in Europe, down 16%. European CDMO, down 20%. North American bioanalysis, down 20% against last year's comp. Agroscience, down 20% in both regions too.

CDMO dragged the segment down too. The release says “Contract ends that have not yet been replaced.” I wrote about that phrase back in Q1. Back then they didn’t say where. Now they do, at least partly. Martin's explanation: "sometimes pharma has certain programs... it makes it a little bit lumpy." Not lost to a competitor, in his view.

In April I said Q3 would be the quarter that tells us whether the Central Lab pipeline is a timing problem or something worse. That checkpoint hasn’t arrived yet. What arrived instead, one quarter early, is a worse number than the one I was waiting to check.

Margin

There’s a footnote in the SDI table I want to start with.

H1 2026 included "two one-time legal settlement gains totalling €19m," a competitor settlement and an insurance recovery, both in North America. The release says even without that, total SDI still improved YoY. That’s true, technically. I did the calculation myself: total SDI went from -€37m to -€15m, a €22m improvement. Take out the €19m gain and what's left is €3m.

Laurent, the CFO, confirmed something on the call I'd only estimated myself: "including a 50 basis points exceptional gain from legal settlements," out of the 190bps improvement in reported margin. Nice to have that from the source instead of just my own math.

Worth separating where that €3m actually came from. Non-mature scope, the start-up and integration losses, went from -€20m to -€8m, a real €12m improvement, nothing to do with the settlement. Mature scope: -€17m last year, -€7m this year. Strip the €19m gain out of that -€7m and mature scope one-offs got worse, not better, year on year.

One thing about that €12m improvement, Non-mature scope revenue is down, €251m to €240m. Part of that is Genomics Europe, the release says is getting reclassified into mature scope this year. So some of the improvement in non-mature losses isn't the same businesses performing better, it's smaller businesses.

The regional breakdown on the call was more useful than the consolidated number. Europe, +210bps. North America, +230bps, but +90bps once you strip the legal settlement out, Laurent gave that number directly too. Rest of World, +130bps.

None of this touches adjusted EBITDA margin, which excludes SDI. 23.7% on total revenues, up 130bps. The reported margin, 23.3%, +190bps, has a piece of a legal settlement in it, and that piece doesn’t happen again next year.

An analyst from Citi asked the version of the margin question I wanted answered too. How much headcount came out. Where it came from. Whether any of it was forced. Martin never gave a number back. He talked about SYNLAB duplication instead, excess clinical capacity sitting around since COVID, a few businesses closed or folded into others, "the margin is mostly stopping things that cost money." Fine, I buy that as part of it. Still no headcount number though.

The number I actually went looking for is the one from my own May thesis. I’d written a test: watch the non-mature scope EBITDA headwind as a percentage of group EBITDA, below -1.5% in H1 2026 means the integration’s progressing. The release doesn’t publish that percentage anywhere, I had to compute it myself. -€8m against €862m group reported EBITDA. -0.9%. Down from -2.6% in H1 2025.

One question got a flat answer, flatter than everything else on this call. Barclays asked what could push margin past the 24% 2027 target, mature scope basically runs at that level already. Martin gave nothing. No pricing, no scale, no direction at all.

I think he just hasn’t built that model yet. Every other hard question tonight, capex, BioPharma, headcount, got some kind of answer, even the evasive ones had content in them. This one didn’t, and I don’t think that’s strategy. Could be he’s holding back on purpose too, not wanting to anchor expectations above a target he hasn’t hit. But if I had to bet, nobody’s run the 26% scenario internally yet, they’re still finishing the 24% one.

Capital Allocation

Martin bought back shares in H1 2026 at an average price of €63.04. My own entry was €57.

3,653,476 shares. €230m gross, €206m net of LTI proceeds. FY2025 full year was €541m, annualize H1 2026 and you land close to that same pace.

Capex is the number I actually want to check, because in April I was more cautious here. FY2025 net operating capex was €328m, FY2026 guidance (confirmed in this release) is around €400m, 22% higher. Back in April I changed it to 4.5-5% of revenue. The original model was using 3.5%, and I didn't think that made sense anymore with €400m of capex.

H1 2026 net operating capex: €145m. 3.9% of revenue. Annualized, that's closer to €290m for the year. Well under guidance.

Goldman asked why capex came in light. "Some of our businesses that are more challenged on growth, they spend less, that may be why we're a bit below," Martin said. Frugality too, "we don't spend when we don't need to." €110m under guidance. I don't know the split between the two.

On owned sites specifically, separate line from operating capex, Martin said it isn’t linear and Lancaster, the last big build in the programme, might slip past 2027. “Maybe some will shift to 2028.” One site. First time I’ve seen any piece of that deadline move though.

On the three-year picture: margin up 500bps, capex down 240bps, working capital intensity down 190bps. FCFF times five over that stretch.

Whether that holds through H2 or spending is just back-loaded this year I don’t know yet, it sometimes is.

65 is close to where this settles. Eurofins buys reagents and instruments for 950 labs, suppliers aren’t pushing terms on a client that size, and 65 days isn’t unusual in industrial supply anyway, but could be wrong.

DSO is where I’d look instead. It came down again this half, and Martin said in January collections still aren’t where he wants them, maybe more room there than on payables.

Net debt €3,854m, leverage 2.2x, unchanged from December, inside the 1.5-2.5x range. Same leverage carrying more cash out the door, buybacks, and, this half, a deal.

That deal's next. Few things from the call too. And the model, went back into that one after. The rest is for paying subscribers, thanks for reading this far.

Read more

Back in June, I finished a piece looking at TIC's data center business region by region. I said the cleanest binary signal would be a platform acquisition above roughly $200 million of EV that brought TIC a European or Middle East entry. That was the thing to watch. If it happened, the optionality was real. If the silence continued, the $68 billion of EU and Middle East demand on the TAM slide was there to fill a slide.

What I didn’t know when I published is that the signal was already two months old. Bureau Veritas had made the acquisition in April. I was watching for a door to open or close, and it had already closed while I was writing.

1. Who bought what

Bureau Veritas agreed to buy Lotusworks on April 6. Irish firm, commissioning and QA/QC for mission-critical facilities, hyperscaler clients, around 750 specialists across the US and Europe, €131 million of 2025 revenue (Disclosed: BV, April 2026). Bureau Veritas called it a one-of-a-kind platform in the TIC industry and said it would represent about 15% of its Buildings and Infrastructure division.

Read that against TIC's own acquisition criteria and the overlap is obvious: commissioning as a capability, data centers as an end market, an established European base built on hyperscaler relationships. It is the target TIC’s strategy describes, bought by a competitor.

And it wasn’t a one-off. Lotusworks was the third mission-critical specialist to change hands in about four months. In December, Accenture, a consultancy, signed for 65% of DLB Associates. In January, Jabil, a contract electronics manufacturer, closed Hanley Energy for roughly $725 million. That same month SGS, one of the big three in testing and inspection, closed Applied Technical Services for $1,325 million. ATS doesn't quite belong with the other three, and I'll come back to why it may matter most of all.

This was the loud wave. The quiet one ran from 2021 to 2024, when the big engineering houses absorbed the independent mission-critical shops one at a time, almost always at undisclosed prices. kW Mission Critical was one of them. None of those came with a disclosed price, which makes it difficult to know where the market was clearing at the time.

The second wave came with nine-figure numbers, and that’s what I keep sitting with.

2. What TIC says it will pay

The chronology matters more than I first gave it credit for. Bureau Veritas announced Lotusworks on April 6. TIC’s investor day came a month later, in May, where the company published the framework it uses to buy companies. I quoted part of it in June, the “selective international beachheads” line, so I'll only add the part that matters here: what they're willing to pay. Planned deployment of $100 to $150 million a year, at about five to eight times EBITDA, or one to one-and-a-half times revenue.

The market had already paid more than that.

Hanley went to Jabil at around 1.9x revenue, my own estimate (Estimated: $725M against Hanley’s projected $350-400M revenue, both Disclosed by Jabil). And Lotusworks came with a full price tag, €375 million of EV against €131 million of 2025 revenue, about 2.9x, at a disclosed 15 times 2026 EV/EBITDA, with an earn-out on top if the business beats its plan (Disclosed: BV, April 2026).

TIC’s own valuation lens caps this asset at €197 million, one-and-a-half times Lotusworks’ €131 million of revenue. Bureau Veritas paid about 90% above that ceiling, and that’s before the earn-out.

So, in January the category cleared at around 2x revenue. In April it cleared at close to 3x. In May, TIC published a deck saying it pays up to 1.5x. A company wrote down, in its own materials, a maximum for this category. The market had already spent months repricing it to around double that level, and the slide is dated a month after the evidence it would have had to price in.

3. Why the M&A engine carries the plan

Independent commissioning is the last verification step before a hyperscaler energizes hundreds of megawatts. The value is entirely in certified people and standing framework agreements, which is why these firms sell on revenue multiples. What I want to add to the regions piece is how much of TIC’s plan depends on being able to keep buying them, because it’s more than the 4% of revenue data centers represent today.

You can see it in the growth plan. The 2029 framework targets +$3 billion against a 2025 base near $2.1 billion, and about $400 million of that bridge is meant to come from acquisitions (Disclosed: framework; the 44% share is my calculation). Nearly half the plan is bought, not grown. That isn’t unusual and I don’t hold it against them, but it means the M&A engine is load-bearing. Cut it and the plan doesn't reach $3 billion.

That dependency traces back to how this piece of the business got built. On the Q3 2025 call, Ben Heraud described the data center business starting as MEP and commissioning that entered “as we bought” in Asia Pacific a couple of years back, then brought to the States and layered with more scope. The capability was built from bolt-ons made before data center commissioning was an investment theme, bought at tuck-in prices in a category that has since repriced to platform levels.

Then there's the timing. That's the part I keep coming back to. In the March release, Robbie Franklin tied the pace of acquisitions to the balance sheet: as leverage comes down toward the below-3x goal, they expect to step up tuck-in and strategic deals. Net leverage is around 3.8x now. Take them at their word and the window for strategic-sized purchases opens in 2027 or 2028. The category consolidated between late 2025 and April 2026, while TIC’s leverage was still coming down toward the room it would have needed to compete for it. That delay is a sequence they chose, as I read it, and it's worth seeing why.

How it comes down matters. The Investor Day puts capital allocation in order: organic growth first, accretive M&A second, share repurchases third, opportunistic debt repayment fourth.

Nothing forces a faster pace, either. The only financial covenant sits on the revolver, and it tests only if drawings exceed 35% of the commitment, which they haven't. So the ratio falls mainly as EBITDA grows, and that takes years.

One detail changed the way I looked at it. Buying EBITDA only improves a leverage ratio if the multiple paid sits below the ratio itself. TIC’s is around 3.8x; the program average paid is 5.4x. Run $150 million of FCF through both uses and the arithmetic is close: left on the balance sheet it takes net leverage to 3.3x, spent on tuck-ins at the program multiple it lands nearer 3.5x. (Estimated from the Q1 2026 balance sheet: $1,608.9 million of debt against $426.6 million of cash, measured against combined 2025 adjusted EBITDA of $311.8 million, all disclosed.)

Leverage here comes down through organic EBITDA growth, presumably why it sits first on the slide and why repayment sits fourth. And it doesn't move in a straight line either: net debt was slightly higher at the end of March than at year end, on working capital timing. The deals get bought for the business, not for the balance sheet.

Read more

One thing to keep in mind: the model reflects what I know when the thesis is published. If the facts change, the numbers change too, and the update becomes the version that matters. Whenever I publish one, I’ll link it here.

A quick note before you start: this one is about a 45-minute read, longer than I usually aim for. Four metro markets, a debt maturity wall, and a NAV bridge made it difficult to make it any shorter. No need to read it in one sitting.

1. Investment Case

You’ve probably never heard of BSR REIT, and there’s no real reason you should have. It’s a small, Toronto-listed trust that owns apartments, 26 garden-style communities, 7,170 units, spread across Texas, Oklahoma and Arkansas. Pools, gyms, covered parking. Nothing exotic, and nothing most investors ever bother to look at.

The Toronto listing is the first thing that looks strange about it. The trust is Ontario-registered, the main ticker is HOM.U on the TSX, but BSR doesn’t own a single property in Canada. Every dollar of rent is American, priced in US dollars. There’s also a US OTC line, BSRTF. None of this is a tax structure someone engineered. Two Arkansas families built the platform over decades and just happened to take it public in Toronto, in 2018.

It’s internally managed, close to 90% of its income comes from three Texas metros (Houston, Dallas, Austin) and it’s still run by people who were operating apartments in those exact submarkets before the company existed.

Right now it also looks cheap. In early 2025, BSR sold a third of its portfolio to AvalonBay at a price that landed almost exactly on BSR’s own book value. Since then the units have dropped to around $11.90, near their 52-week low, about 29% below BSR’s reported NAV of $16.72. A big buyer validated that book value in cash, and the market has spent over a year pricing it like that book value is wrong by a third. I keep coming back to that gap.

Run the same arithmetic in the other direction and the gap gets stranger. Back a cap rate out of $11.90 against income from BSR’s stabilized properties, the ones no longer leasing up, and the market is charging something close to 6.5%. I went through every Houston, Dallas, and Austin transaction with a disclosed price. None of them printed that wide, distressed or not. Somebody is pricing in a version of these three markets that hasn’t shown up in an actual deed yet.

Some of that discount is earned. Per-unit FFO and AFFO both dropped by about a fifth over the last year as BSR rotated the AvalonBay proceeds into newer, still-leasing Houston and Dallas properties, and leverage went up, not down. The AFFO payout ratio climbed from 63.8% to 82.8% over the same year, touching 125.7% for one quarter in between. Texas overbuilt apartments between 2022 and 2025, and BSR’s own new-lease rates are still negative. Management has also said it wants to carry more floating-rate risk heading into the biggest single-year maturity wall the company has, almost half its principal due in 2029, though so far its hedging moves have gone the opposite way, adding fixed-rate cover instead of shedding it. New deliveries are projected to fall 43% in Austin and close to 30% in Dallas in 2026, while population growth in both metros is still positive, decelerating in Austin, steady in Dallas.

Whether that’s enough to close a 29% gap is what the rest of this document tries to answer. I don’t know yet.

2. The Portfolio and How It Generates NOI

My past theses on Veris and FRP spent real time on the physical character of the assets themselves, and I'm carrying that habit into this one. Garden-style multifamily isn't Veris's kind of building, no dramatic views, maybe no five-star amenity stack, but the BSR properties have their own version of that appeal to me: pools, parks, gyms, clubhouses, the ordinary texture of where people live. It's possible I only find this kind of asset nice because none of it exists where I am.

BSR splits properties into two buckets. Same Community properties were stabilized and held through both this period and the same period a year ago. Non-Same Community is everything else, mostly recent buys or builds. A property can sit at 95% occupied and still count as Non-Same Community, because the classification is about time held, not how full it is today. Without a full trailing year of stabilized income, there's no clean number to run a cap rate against, which is why some strong performers below get valued at cost instead of by income.

HOUSTON: 40% of NOI, the single largest market

Houston is where I start because it's where the most money sits: ten properties, 2,874 units, 40% of portfolio NOI as of Q1 2026.

BSR discloses occupancy by property, but rent and NOI only at the MSA level, where average rent came in at $1,521/unit as of March 31, 2026, down from $1,525 a year earlier.

Forayna Vintage Park and Botanic Luxury matter most here, and neither is the biggest or oldest. Forayna sits in northwest Houston's affluent Vintage Park submarket; Botanic Luxury, in Spring, is smaller and slightly less affluent.

BSR bought both in May 2025 for $141.0 million combined, $221,000/unit, funded by AvalonBay proceeds. About the lease-up, a buyer without a Texas leasing team doesn’t get there this fast. Colliers’ Q1 2026 data puts Houston Class A occupancy metro-wide at just 85.6%, flat for two straight quarters, against BSR’s own 94.8%.

At the other end, Lakes at Westview I (2005, 128 units) has the lowest occupancy in Houston at 91.4%; Falls at Borough Park (2003) ties for second-lowest at 92.5%. Both are small and old enough to watch as candidates for a future sale or renovation, my read of the pattern rather than anything management has said.

Houston has the smallest supply problem of the three Texas metros, imbalance concentrates in the Northwest, which delivered 3,575 units and absorbed 1,668 in Q1 2026 alone, more than double any other submarket on both counts, with 5,256 more still under construction. Oberste points to prior Houston supply peaks, in 2007 and 2014, absorbed within two to three quarters. If that holds again, Houston is probably the first market to stabilize.

Employment shows no flattening near the chart's right edge, so whatever pressure Houston real estate is under isn't coming from the job market. Census Bureau Vintage 2025 data sharpens the point: Harris County added more residents than any county in the nation and just became the first Texas county past 5 million people, though growth is decelerating even here.

Houston, Building an independent value

BSR discloses NOI at the MSA level, not per property, so I can't value each of the ten properties separately. What I can do is value Houston the way an appraiser would: divide annual income by the return rate buyers expect, the cap rate, and what's left is what the properties are worth today.

I used a trailing twelve-month figure rather than annualizing Q1 2026, which would understate seasonal swings, from Same Community NOI on the 2,236 stabilized units. The two lease-up properties are handled separately below.

LTM Houston Same Community NOI: $22.84 million on 2,236 units.

The cap rate. Cushman & Wakefield's 4.75%-5.25% institutional range, plus six Houston transactions I could find with pricing attached, in a state that doesn't require sale prices to be recorded, only two of which disclosed a going-in cap rate, both above 5.1%. The distressed context on Copper Springs is a fair reason to treat it as a ceiling, but with only two real prints, I'd rather sit with that tension than resolve it in BSR's favor, so I ran three scenarios instead of picking one. The gap between conservative and upside, $48 million, about 10% of value.

I'm not giving Forayna and Botanic a cap rate yet. BSR only reports their income mixed together with three other properties outside Houston, and if I split that number up by unit count, I'd just be pretending to know something the data doesn't actually tell me. So instead, I'm using what BSR paid for them in May 2025: $141.0 million, $221,000 a unit. That's probably a conservative number too, since occupancy has already jumped from about 80% to over 95%.

My Base Case per unit, $204,899, sits below what AvalonBay paid for BSR’s Austin and Dallas assets ($218K-$234K/unit), consistent with Houston's lower rents, which is what I'd expect to see. It also matches BSR's own portfolio-wide weighted average cap rate exactly, which could mean it's right for Houston, or that I've anchored to a number I already knew. Every piece of real evidence still sits above 5.1%. At some point I have to stop calling Conservative the cautious scenario and just call it the base case.

Camden, MAA, and NexPoint all reported flat-to-negative insurance renewals after years of increases, and BSR's own Q1 2026 MD&A cites a $0.2 million drop in insurance expense helping the quarter's NOI. One less reason for cap rates to drift wider, if it holds, though this is still gross asset value, not NAV.

Dallas: 32% of NOI

Dallas is the second-largest market, 32% of NOI on 2,026 units across six properties. Dallas absorbed more oversupply than Houston did.

At the MSA level, Dallas Same Community occupancy is the only one of the three Texas metros that went backward last quarter, excluding The Ownsby and Venue Craig Ranch. The Ownsby’s own occupancy is disclosed separately: 73.1% as of March 31, 2026, up from 70.4% at year-end.

The property that matters here is The Ownsby, and the story isn't good yet. BSR bought it in August 2025 for $87.5 million, 368 units, $238,000/unit, in Celina, in the McKinney/Frisco growth corridor, a 3-mile radius that's grown about 10% in population since 2020 and carries $177,100 in average household income, the richest catchment of anything BSR bought in its entire 2025 acquisition round. On paper it's the best-located property BSR bought that year, which makes the occupancy gap harder to explain away.

Three possibilities occurred to me before I had the data to settle it. Submarket-specific oversupply, if new competing product landed disproportionately in Celina rather than spread across North Texas. A timing lag: three months separate The Ownsby’s purchase from Forayna and Botanic’s, and lease-ups don’t all move on the same clock. Or, less comfortably, BSR just isn't as fast at lease-ups outside Houston, its most established market, as it is inside it.

The answer, once I had it, follows below.

Employment kept climbing straight through 2024 into 2026, past 4,300,000. Whatever pressure Dallas real estate is under right now, it isn't coming from the job market.

Population growth here looks the least interrupted of the three Texas Triangle metros, barely a pause through 2008 to 2010 and a steady climb since. The Census Bureau’s Vintage 2025 estimates sharpen the picture: Collin County, home to Frisco, McKinney, and Celina, added close to 43,000 residents, second only to Harris County nationally. Dallas County, the urban core, actually lost population over the same period. Celina itself, where The Ownsby sits, was the single fastest-growing city in the entire United States that year, up 24.6%.

Venue Craig Ranch, acquired back in January 2025 for $61.0 million (277 units, McKinney), has had over a year to season and is still at 89.2% occupancy as of December, below the Dallas Same Community average. Slower than I’d want this far past acquisition, though nowhere near as concerning as The Ownsby.

On the other side, Riverhill is worth naming: built in 1995, the second-oldest property in the Dallas book, and it carries the highest occupancy of the group at 97.3%. Age alone doesn’t predict performance.

Dallas NOI: the clean number

Same Community here means the 1,381 units excluding The Ownsby and Venue Craig Ranch, same trailing-twelve-month method as Houston.

LTM Dallas Same Community NOI: $14.39 million on 1,381 units.

Why The Ownsby is leasing slower

Celina took on more new supply than the rest of North Texas. CoStar's Q1 2026 data shows Prosper/Celina posting the worst rent decline in the entire DFW MSA, down 4.7% year over year. Three projects explain why: the Depot at Celina Station, 336 units, opened February 2026; Merritt Crossing, 335 units total, 226 in phase one, starts move-ins this summer; and Gates of Prosper is building toward 600 units. Matthews reaches the same conclusion on its own, flagging Frisco/Prosper and Allen/McKinney as the most pressured submarkets in DFW. Three sources land on the same few square miles, my own research, MAA's Section 6 commentary, and Matthews, none citing each other. The submarket absorbed more competing supply than almost anywhere else in the metro, right as BSR was trying to lease up. That timing explains the gap. The platform's execution doesn't need defending here.

For Houston, carrying Forayna and Botanic at acquisition cost was conservative because occupancy had already jumped to 95%+. The Ownsby hasn’t made that jump, and now I know why: it’s leasing against at least 1,100 new competing units in its immediate corridor. Carrying it at cost isn’t obviously conservative anymore; it might be closer to fair, or eventual NOI might land lower than what BSR underwrote.

That $218,213 sits below the $230,233 BSR paid for Ownsby and Venue Craig Ranch because the older stabilized book prices lower, not because the lease-ups lost value; both are still carried at cost.

Austin: 18% of NOI

Austin is the smallest of the three Texas Triangle markets in BSR's book: four properties, 1,317 units, and the one that took the worst of the supply wave. New-lease rates fell 8.1% in Q1 2026 alone, more than double Houston's 3.8% decline and worse than Dallas's 7.1%.

Aura 35Fifty is the one to watch here, and unlike The Ownsby in Dallas, this lease-up is working. Built in 2024 in the Round Rock submarket, it opened straight into the thick of Austin's supply glut, leasing into what Oberste himself called "all the headline deliveries in that Round Rock submarket," and still posted a 57-point occupancy swing in nine months. If Forayna and Botanic in Houston were the first evidence for the platform's lease-up execution, this is the second, arguably the more impressive one given what it opened into.

The stabilized side tells a different story: Markham at Lakeline, the oldest of the four Austin properties (2014), carries the lowest occupancy among the three Same Community assets at 92.0%, below Aura 36Hundred (95.5%) and Ariza Plum Creek (94.6%), the opposite of what age alone would predict. All three still sit in the low-to-mid 90s, nowhere near distressed. The 8.1% new-lease rate decline is where the damage shows: Austin's hit this cycle landed on rate, not occupancy, so a property can hold its units full and still take the loss the moment a lease renews.

This is the one employment chart that looks different from Houston and Dallas: the recovery off the 2020 dip was the steepest of the three metros through 2023, but 2024 through 2026 flattens, still growing but visibly decelerating, confirming the Austin softness already showing up in the new-lease numbers above.

Population growth shows similar late-period flattening. This FRED series runs through 2022, but Austin crossed 1 million residents in 2024-2025 per Census Bureau Vintage 2025 estimates, and Williamson County, where Round Rock and Aura 35Fifty sit, saw 81% of its recent growth come from migration. The deceleration is real, but a slower positive, not a reversal.

Austin NOI: the clean number

Same Community here means the 1,079 units excluding Aura 35Fifty, still in its lease-up period. Same trailing-twelve-month method as Houston and Dallas:

LTM Austin Same Community NOI: $11.08 million on 1,079 units.

CBRE’s H2 2025 Cap Rate Survey breaks Austin out by exactly the segment BSR owns: suburban Class A, stabilized, at 4.25% to 5.0% in H2 2025, unchanged from H1, with CBRE’s own view that yields have hit their cyclical peak.

Northmarq's Q1 2026 report puts Austin at 5.5% to 6.5%, a full point wider than CBRE, but that reflects a sales mix weighted toward Class C product in North Austin, a lower-quality pool than what BSR owns. Applying it here would understate the portfolio.

The anchor I trust most is still BSR’s own transaction: the March 2025 sale of 857 Austin units to AvalonBay, including Retreat at Wolf Ranch in Georgetown, disclosed as a “high 4% range,” with a third-party estimate of 4.7% to 5.1%. CoStar corroborates: Class A product in Round Rock and Georgetown priced near 5.0% in mid-2025, drifting toward 5.5% to 5.7% by early 2026 as asking rents stayed negative.

Not every property here fits that anchor cleanly: Ariza Plum Creek sits in Kyle, an outer submarket rather than the Round Rock/Georgetown core, and BSR's book skews partly B+ rather than pure institutional Class A.

Aura 35Fifty never had a purchase price, so the value has to come from comparables

BSR developed Aura 35Fifty through a joint venture, finished construction in December 2024, then bought out the developer’s stake in February 2025 with Class B Units, taking full ownership of all 238 units.

Comparable new-construction Class A product in Round Rock and Georgetown has traded in the $190,000 to $220,000 per unit range through the same window, bracketing Aura 35Fifty at $45-$52 million. Given the lease-up has gone 35% to 92% occupied in under a year, that range looks fair rather than optimistic.

Colliers’ Q1 2026 report adds confirmation: the metro-wide median cap rate sits near 5.0%, up from a 4.3%-4.4% trough in Q3 2025, still widening rather than flattening, matching CoStar’s read and running against CBRE’s H2 2025 “cyclical peak” call.

Of the three stabilized properties, only Aura 36Hundred sits inside the Round Rock/Georgetown submarket that CoStar’s 5.5%-5.7% print describes, on the same street as its lease-up neighbor Aura 35Fifty, valued separately above. Markham at Lakeline and Ariza Plum Creek sit outside it, in Cedar Park and Kyle, so that print shouldn’t apply to two-thirds of the stabilized bucket by units, pulling the blended number closer to CBRE’s range.

Colliers’ submarket data backs this up: 36Hundred and Ariza Plum Creek both outperform their submarket’s Class A occupancy by three to four points. Markham at Lakeline is the outlier, running behind its own submarket, 92.0% to 93.2%, evidence its weakness is asset-specific, not neighborhood-wide.

Read more

Last month, in #03, I wrote about GATX and its railcars: a business that has to lay out enormous cash years before it earns anything back. Every new tank car gets paid for upfront and then spends decades slowly earning its keep.

This time I went looking for the opposite kind of company, one that pays for almost nothing upfront, because it’s own customers and suppliers end up funding the growth for it. Same question as last month, how does a business pay for it’s own growth, just answered the other way around.

There's a company in this issue that owed $2.67 billion in bills due within the year, but only had $1.99 billion on hand, in cash, inventory, and money customers still owed it, to cover them. Taken on its own, a number like that usually means a business is one rough quarter away from missing a payment. Keep reading anyway. What's actually going on here comes down to a habit almost nobody thinks to check for, and it's worth learning to spot.

I want to show you how to tell this apart from the version that really is a warning sign, because from a distance they can look the same.

The funding gap

Every business that sells physical things has a timing problem. It pays a supplier today, puts the goods on a shelf, waits for a customer, and only gets its cash back once that customer pays. The money is tied up the whole time in between. That stretch, from cash going out to cash coming back, is the cash conversion cycle.

For most businesses, that cycle is positive. They’re out of pocket for weeks or months, and they have to fund that gap with their own cash or a loan. The longer the gap, the more money just sits there, in inventory and unpaid invoices, instead of working for the owner. It’s a real cost, and it grows as the business grows.

Three numbers build the cycle: how long inventory sits before it sells, how long customers take to pay, and how long the company itself takes to pay its suppliers. Add the first two, subtract the third. If a company sells its inventory fast, gets paid by customers right away, and takes its time paying its own suppliers, the cycle can flip. It goes negative.

A negative cash conversion cycle means the company gets paid before its own bills come due. The customer’s money arrives first. The supplier gets paid later, with that same money. The business runs on cash it hasn’t had to spend yet, courtesy of the people on both ends of the transaction.

Why it looks like a problem

Here is where many investors get it wrong, and it’s an understandable mistake to make.

When a company collects cash before it pays its suppliers, that unpaid supplier money sits on the balance sheet as a liability, accounts payable. If the company is good at this, those payables can grow large enough to outweigh the inventory and cash it holds on the short-term side. The standard measure for this is the current ratio, short-term assets divided by short-term liabilities. When it drops below 1.0, the company owes more in the short term than it owns in the short term.

For most of the history of financial analysis, a current ratio below 1.0 gets taught as a red flag. Supposedly it means the company might not be able to cover its bills. Sometimes that’s exactly what it means.

But in a business with a negative cash conversion cycle, a current ratio below 1.0 isn’t telling you the company is in trouble. It’s what the funding gap looks like when it’s working in the company’s favor. The supplier money piles up because the model is good, more than because the company can’t pay. Two businesses can land on the same ratio for completely different reasons.

That’s why the consequences for the investor go in different directions depending on which one you’re looking at. A strong business with a negative cash conversion cycle generates cash as it grows, because every new sale gets funded by someone else before the company has to spend a cent. A weak business with negative working capital is quietly stretching its suppliers because it can’t afford to pay them on time, and the day a supplier loses patience, things get difficult fast. The balance sheet can look almost the same either way. What’s happening underneath it isn’t.

Change in working capital

Think of the current ratio as a photograph. It shows you where the company stands on one particular day, once a year. The cash flow statement is closer to a short video: how that position moved over the year, and whether the movement brought cash in or pulled it out.

A 10-K shows the pieces separately. Most financial websites combine them into a single line: change in working capital. Same idea, different presentation.

A positive number sounds like good news, and often it is, real proof the company collected faster or leaned on suppliers a little longer than before. But it can also be something less flattering. A company can simply delay paying suppliers until a few days after year-end. Cash stays in the business a little longer, operating cash flow looks stronger, and the bill becomes next year's problem. The easiest check is the following year. Real improvements usually stick. Timing effects usually reverse.

A negative number isn’t automatically bad either. Growth costs working capital cash even in a business that’s run well. More stores just means more inventory sitting on more shelves, in absolute dollars, even if the business is exactly as efficient as it was the year before. A negative change next to a company that’s opening new locations is often just what expansion costs, nothing more sinister than that.

The question is how much of operating cash flow came from working capital. Compare the two. If working capital is only a small part of the total, there's usually nothing interesting going on. If it accounts for a large share, it's worth understanding why before giving the company credit for it.

— BJ’s Wholesale Club Holdings ($BJ) —

Let’s use BJ’s as a lab. BJ’s operates 263 large-format warehouse clubs and 199 gas stations across 21 states, mostly on the eastern half of the United States. Founded in New England in 1984. Over 8 million members pay an annual fee for access, which brought in $499.8 million in membership income in fiscal 2025 alone, on top of everything they spend inside the clubs. If you know Costco, you already get the shape of this business. Buy in bulk, sell in bulk, keep the assortment narrow so inventory turns fast. BJ’s follows the same model, just at a smaller scale and with a much more concentrated footprint.

Look at the balance sheet. At the end of fiscal year 2025, we get a current ratio of 0.75. The year before, 0.74. Two years before that, 0.67. The year before that, 0.76. BJ’s has sat below 1.0 for years running, in roughly the same band each time.

BJ's actually explains why in its own annual report. High sales volumes and fast inventory turnover mean the company sells most of what it buys before its bills to vendors are even due. Vendor payment terms end up financing the inventory on the shelves, not the company's own cash. That's a negative cash conversion cycle, even though the filing never uses the term.

This is where it shows up in the cash flow statement. Every year you'll find the same three lines: receivables, inventory, and payables. Together they explain how much cash working capital added or consumed during the year.

In fiscal 2025, the amount customers owed BJ's dropped from $277.3 million to $252.8 million, freeing up $24.8 million in cash. Inventory grew as the business grew, using $46.5 million. Accounts payable grew faster, adding back $53.9 million. Change in working capital on those three lines: about plus $32.2 million. (You'll sometimes see a higher working capital figure on financial websites, around $99.8 million for BJ's. That's because they also include items like prepaid expenses, accrued compensation, and other operating balances. I'm focusing on receivables, inventory, and payables because those are the lines that explain how the retail business itself generates cash.)

The bigger number, the one worth sitting with, is what happens to the total across several years. Operating cash flow was $718.9 million in fiscal 2023, $900.9 million in fiscal 2024, and $1,030.1 million in fiscal 2025. Free cash flow follows the same climb once you subtract what the company spent on new clubs and equipment: $251.8 million, then $312.9 million, then $328.0 million across the same three years. If the cash flow improvement were mostly coming from supplier financing, you'd expect the underlying picture to look more fragile than this. It doesn't.

A fair objection: a current ratio below 1.0 genuinely is dangerous for some companies, and calling it “just the business model” is exactly how investors get into trouble. The ratio by itself isn't enough. What settles it here is that BJ's has operated in roughly the same range for years while operating cash flow kept climbing. A business running out of room doesn't usually manage both at once.

Two companies can land on the same current ratio for completely different reasons. One is efficiently financed by its suppliers. The other is paying them late because it doesn't have a choice. The balance sheet alone won't tell you which is which. That's why I never stop at the ratio.

Three tests worth running on any company with a current ratio below 1.0.

Test 1. Is operating cash flow growing, and how much is working capital contributing? A healthy negative working capital business makes most of its cash from operations, not by squeezing suppliers a little harder each year. BJ's change in working capital came to about 3% of last year's total operating cash flow, small enough that the growth clearly wasn't riding on that alone.

Test 2. How fast does inventory move, and who is financing it? Read what management says about how it funds inventory. Warehouse clubs like BJ's lean on supplier payment terms because they sell through inventory before those bills are even due. A distributor or a manufacturer might be funding that inventory with its own cash instead, and that difference tells you more than the ratio does on its own.

Test 3. Does the current ratio stay consistent over time? Don't stop at the two years sitting side by side in the latest filing, pull four or five. BJ's has stayed in roughly the same band, 0.67 to 0.76, across several fiscal years, with cash flow climbing the whole time. That kind of consistency is a business model showing up in the numbers.

A current ratio below 1.0 only describes a shape on the balance sheet. Whether that shape means the business is healthy or in real trouble is something only the cash flow statement, read across a few years, can tell you.

BJ’s has owed more in the short term than it owns in the short term for years now, and its own filings explain exactly why: vendors are financing the inventory, not the company. Glance at the balance sheet on its own and it looks like a business one bad quarter from trouble. Read three years of the cash flow statement next to it and the story flips: a company generating more cash every year, mostly because its suppliers are carrying a cost a less efficient retailer would have to carry on its own.

Share

If you made it here, here's one thing I'd take into the next 10-K you read. Whenever you see a current ratio below 1, don't stop there. Pull three years of operating cash flow and see which direction it's moving. It's one of the quickest ways I know to tell whether you're looking at a business model or a warning sign.

See you next month.

Disclosure: I do not hold a position in BJ’s Wholesale Club Holdings ($BJ) as of June 2026. This is an educational piece, not a recommendation.

I. Target at Rego Park II

Alexander’s leased 135,000 square feet to Target at Rego Park Shopping Center on June 29. The press release says the center is now 99% leased, with a weighted average lease term of about 9.3 years. But the story here isn’t what happened to the occupancy rate.

That space wasn’t empty before. Kohl’s closed its store at Rego Park II some time ago, but the company kept paying rent on it. As of December 31, 2025, the date of the last audited annual filing, Kohl’s was still listed as a tenant, still obligated under its lease through 2031, with a footnote in the 10-K noting the store was “currently closed.” A closed store paying rent doesn’t show up as a vacancy in the occupancy table. It creates something worse: a dark anchor pulling foot traffic down while the numbers look fine on paper.

Target replacing a closed Kohl’s with an operating store is a credit upgrade, not a fill. The lineup at Rego Park II was already Costco, Burlington, Best Buy, Marshalls, T.J. Maxx, Aldi, and Petco. Adding an operating Target on a 15-year contract with renewal options changes the quality of that center, not just the count. What it means for the valuation, I’ll get to in section III. The short version is: the valuation impact is limited. The risk it closes isn’t.

II. The Q2 Filing

Rego Park I closed May 28. Gross price $235.5 million. Alexander’s received $224 million of cash at closing ($21 million of transaction costs had already been paid in prior quarters) and expects to recognize a $148 million financial statement gain in Q2. It’s confirmed, it’s already in the bank, and it’s the largest single income event this company has produced in years.

The tax gain from the transaction is $145 million. $48 million was already recognized in 2025. The remaining $97 million hits in 2026, and under REIT distribution rules, that triggers a required distribution to shareholders sometime in 2026 or early 2027. At 90% distributed across 5,107,290 shares, that works out to about $17 per share on top of the regular $18-a-year dividend. The exact amount and timing depend on the final tax calculation, which hasn't been published. The direction is not in doubt. I’ve written about this before. I still think it’s the most likely near-term catalyst.

The Bloomberg abatement

Before April 2026, the flow was direct. Bloomberg paid $6,632,051 per month in Fixed Rent straight to Alexander's. From there, ALX covered the $1,680,000 monthly debt service on the $400 million office mortgage and kept the remaining $4,952,051 for operations and dividends. The asterisk on the diagram matters: the lender held a Sweep Period and Trigger Period in place during this time, giving them the ability to trap cash under certain conditions. Both ended as part of the March 31, 2026 arrangement.

Paid subscribers get the diagram showing what happens to that $4,952,051 from April to December, the $40 data error in the March NAV and the corrected figure, and the full return case from now.

Read more

A quick reminder of where the thesis left things, for anyone who hasn’t gone back to it recently. The balance sheet at the time was sitting on €1.13 billion in net cash, generating over a billion euros of EBITDA a year, with very little of either being returned to shareholders. I called it “Lazy Capital”, a bunker that survives every cycle but rewards the people who own it badly while it does. The thesis listed four triggers that could close what I called the Italian Discount, and one of them, word for word, was “special dividend / buyback → governance signal.”

That trigger fired on February 26. I want to spend this piece on what’s happened since, because the headline number (Buzzi now holds just over 8% of its own share capital in treasury) undersells how interesting the story actually is once you look at it week by week.

One leg of the discount

I framed the Italian Discount as resting on three legs: Russia exposure, ESG exclusion, and governance. The buyback only ever had a shot at fixing one of them.

Russia is still on the books, still generating cash, still impossible to repatriate cleanly, still the thing Pietro Buzzi himself called the biggest risk the company carries. ESG-screened funds are still barred from touching the stock regardless of what management does with the cash pile. Neither of those moved an inch because the family started buying shares.

What did move is the governance leg, and it moved further and faster than I expected when I wrote that sentence in February. Here’s the simplest way to see it. Buzzi held €1,131 million in net cash when I published. By the end of Q1 2026 that was €1,019 million. In between, they spent €102.3 million buying their own shares and still paid for €116.5 million of capex, and the cash pile barely moved because the business kept generating enough to cover both. That’s the part worth holding onto: the company isn’t draining the bunker to fund this. It’s just stopped letting the bunker sit there doing nothing.

What reports actually show

Buzzi publishes a report every week with the exact number of shares bought and the average price paid. I went through all eighteen filings, from the first one in February to the one that came out this morning. One thing worth flagging before the table: the "treasury stake" column is a running total, not this program's contribution on its own. Buzzi already held 11,601,276 shares in treasury before any of this started, 6.023% of share capital, left over from older buyback programs. So February 27's 59,944 shares bought is a tiny fraction of the float on its own, but added to what was already sitting in treasury it pushes the cumulative stake to 6.053%. Here's the whole program rolled up by month:

That 3,857,014 shares is the actual size of this program so far, the part that’s new since February. The cumulative stake of 8.060% mixes that in with the 6.023% Buzzi was already sitting on, which is why the headline percentage jumps around less dramatically than the monthly buying does.

The number that matters more than the total is what happened in the middle of that table. Buyback execution in March averaged a little over 119,000 shares a day. By the third week of April, with the stock trading above €47-48, daily purchases had collapsed to somewhere around 5,000-7,000 shares. Look at April 13-17: 24,362 shares across an entire week, at an average of €48.02. Compare that to March 2-6: 597,434 shares in five days, averaging €46.22.

I keep coming back to this gap because it answers a question I didn’t fully resolve in the thesis. Is this capital allocation, or is it just a mechanical program running on autopilot to flatter EPS regardless of price? A program on autopilot doesn’t slow down by 95% the moment the stock gets expensive and speed back up the moment it cheapens again, which is exactly what happened in May once the price drifted back toward €43-45. This is a family with the full books deciding, week by week, whether their own stock is cheap enough to buy.

There's a less comfortable way to read that same gap, though, and I don't want to skip past it. My own probability-weighted target on this stock is €57. If the family is genuinely slowing down hard at €47-48, that's a level where my model still says there's about 19% left on the table. The most consistent explanation, given everything we already know about how this family operates, is that their bar for "cheap enough to spend cash on" sits well below where my own valuation work puts fair value, this is the same family that sat on €1.1 billion doing nothing for years rather than return a euro of it, so a conservative trigger for spending it isn't exactly out of character. That's different from the family actively disagreeing with €57 as a number. But I can't fully rule out the less comfortable version either, that they're sitting on something in Russia or the European numbers that makes them less confident in the upside than I am. I can't tell which from the outside, and I think it's fair to say the buying pattern doesn't fully back my own target the way I'd like it to.

Three questions the buyback answers

Is this actually creating value for shareholders, or just optics?

Start with the mechanism, because it’s simpler than it sounds. Buzzi buys its own shares on the market and parks them in what’s called “treasury”, a drawer where the company holds shares of itself, separate from what regular investors hold. A share in that drawer earns no dividend and has no vote. At some point, shareholders can approve cancelling those shares for good, and that’s the part that actually creates value: picture the company’s annual profit as a cake, and the share count as the number of people you’re cutting it for. Fewer slices, same cake, bigger pieces. Earnings per share go up without the company selling one extra bag of cement. It only works in your favor if the shares were bought cheaply rather than overpaid for.

Buzzi is mostly still on the buying side of that, not the cancelling side. None of the treasury shares have been cancelled yet, the board got the green light at the May 13 shareholder meeting, but a separate, more formal vote still has to happen before the share count used to calculate EPS actually shrinks. Once it does, the math gets concrete: the original valuation work modeled shares outstanding falling from roughly 181 million to somewhere near 171 million if Buzzi finishes the full €300 million buyback around €43 and cancels everything in the drawer. That’s about 5-6% more earnings per share, from arithmetic alone. And the price matters as much as the volume, at an average of €45.13 against what Buzzi earned per share in FY2025, every share retired is being bought at something like 8-9x earnings. That's the number I keep landing on whenever the "window dressing" question comes up. Dressing isn't usually this cheap.

Does this mean they’ve simply run out of better ideas for the cash?

Maybe partly. This is being funded out of operating cash flow, not by draining the balance sheet, which tells its own story: they decided buying back stock at 8-9x earnings beats sitting on a money-market yield, and they’re right about that arithmetic. Whether they’d say the same about a transformational acquisition is a different question, and not one this data answers.

Does this mean Buzzi isn’t finding M&A opportunities?

This is the one I want to sit with, because it’s the one the rest of the sector actually answers for me.

What the rest of the cement sector is doing

Heidelberg Materials launched a buyback of up to €448 million in May 2026, on the back of strong 2025 results and a credit profile management considers more than solid enough to fund it. In the same window, Heidelberg announced a strategic investment in AmeriTex Pipe & Products, expanding its US infrastructure-products footprint. Same balance sheet, same quarter, both checks written. Buyback and M&A aren’t substitutes for each other in this sector right now. They’re running in parallel at the largest cement company in Europe.

Holcim took the more radical route. Instead of choosing between dividends, buybacks, and acquisitions, it spun off its entire North American business as Amrize in June 2025, freeing both halves to pursue “growth-focused capital allocation and value-accretive M&A” independently, in the company’s own words. It’s a structural answer to the same question Buzzi is working through with a buyback: where does the capital actually belong, and who decides.

CRH closed a $300 million buyback this spring while mid-way through its own corporate simplification. Martin Marietta solved its capital allocation puzzle differently again, an asset exchange with Quikrete Holdings completed in February, plus the Minnesota aggregates assets picked up from CRH in December, neither of which touched the share count at all.

None of this reads like a sector that’s run dry on M&A and is recycling cash into buybacks for lack of anything better to do. It reads like a sector where almost everyone with a clean balance sheet is doing more than one thing at once.

Buzzi has its own version of this sitting on the table right now, and it hasn’t picked an answer yet. CSN (a Brazilian competitor, not a Buzzi asset) started a process last year to sell off its cement division. Pietro Buzzi confirmed on the FY2025 earnings call that Buzzi is looking at it: “it could represent an opportunity. It has to be investigated. The process is at the beginning.” If Buzzi ends up buying those assets, it roughly doubles its footprint in Brazil overnight. That’s the M&A option, and it comes with everything M&A usually comes with, competing bidders, a seller who might prefer someone else’s offer, no guarantee Buzzi wins or even formally bids.

The buyback doesn’t require any of that. No auction, no rival bidder, no negotiation. Buzzi just shows up in the market every week and buys its own stock at whatever price is on screen. That’s the sense in which the two options aren’t really competing with each other, one is available on demand, the other depends on someone else’s process playing out in Buzzi’s favor. The CSN sale hasn’t closed as I write this. I don’t know whether the weekly buyback pace changes the moment it does, or whether Buzzi would simply run both at once, the way Heidelberg has been doing.

A note for anyone who’s tired of waiting

I know some of you have held this stock for a while and watched it go nowhere, and the thought creeps in that maybe this is just dead money. I don’t think waving that away is fair to anyone who has a real decision to make, so I’ll engage with it directly instead.

Two separate questions get bundled together whenever this comes up, and they shouldn’t be. One is whether the thesis is still true. The other is whether you, personally, still want to be the one holding it while it plays out. Those have different answers, and conflating them is how people end up making the decision for the wrong reason.

On the first question, nothing about the underlying business changed for the worse while the stock sat still. The discount was never about the assets. It was about a family that wouldn’t return cash, a Russia exposure that’s still there, and an ESG exclusion that’s still there. One leg out of three just started moving. That’s not the same as the thesis breaking, and it’s not the same as the thesis fully playing out either. It’s one piece of progress, accurately sized.

On the second question, I’d separate two different reasons for selling, because only one of them is really about Buzzi. Opportunity cost is real: if you’ve found something else with a clearer catalyst and more conviction, walking away from a stock that might stay cheap for another two years is a defensible, adult decision, and I’d rather say that plainly than pretend every sale is a mistake. But selling out of pure fatigue is a different thing from selling on the merits, and it’s worth knowing which one you’re actually doing before you do it.

I hold it. I’m not selling. I also built the model, so weight that accordingly.

Where this goes from here

Three paths, and the one that actually matters is the one where nothing visibly changes.

If the weekly reports keep coming at something like the current pace and the treasury cancellation gets formalized at the next extraordinary session, the governance leg of the discount closes almost completely. The float keeps shrinking with management’s own conviction behind every purchase, and the floor I described in the original thesis (bond-like downside protection even in the bear case) moves up a notch.

I get pushback on that floor argument sometimes, usually some version of “Martin Marietta and Vulcan got crushed after the 2008 crisis, multiple compression in this sector is brutal, don’t assume Buzzi is different.” That’s a fair point and I don’t want to wave it away. But the thing that made MLM and VMC ugly post-GFC wasn’t just the multiple. It was leverage. A levered balance sheet going into a downturn has to worry about refinancing at the worst possible moment, and that fear alone compresses the multiple further, on top of whatever the earnings are doing. Buzzi enters any slowdown from net cash, not net debt. That doesn’t make the stock immune to a weaker cycle or a derating, earnings can still fall and sentiment can still sour regardless of the balance sheet. It does mean nobody’s forcing Buzzi to refinance into a closed market at the worst possible time, which is the specific mechanism that turned a bad cycle into a brutal one for those two names. This is the boring outcome and also the one I’d bet on.

If the purchases stop, that’s the signal worth paying attention to, more than any press release framing around it. A family that’s been buying every single week for four months, through a stock move from €43 to above €48 and back, doesn’t quietly stop unless something on the books changed their read on value. I don’t have a clean way to distinguish “program technically exhausted, new authorization pending paperwork” from “they saw something in May’s numbers they didn’t like” from the outside. The weekly filing cadence is the only early warning system a retail holder actually has, which is part of why I’m going to keep reading every one of them.

If the next move is a large acquisition instead of continued buybacks (CSN’s Brazilian assets or something nobody’s mentioned yet) the thesis shifts shape entirely, from a floor-protection story to a growth-and-execution-risk story, and the multiple the market assigns probably moves with it before the ink is dry on any deal.

I don’t think the buyback alone moves my price target. What it does is make the bear case for Buzzi materially harder to argue with a straight face, which is a different kind of progress than the number itself suggests.

Disclosure: I hold a long position in Buzzi SpA (BZU.MI). This is not investment advice. DYOR

What got me digging was realizing that TIC's data center story isn't one story. It's four (one per region) and each answers a slightly different question: how much of a head start TIC actually has, what the real competitive set looks like, and which gaps are structural versus temporary.

Side by side, the four regions changed more of my read-through on the segment than treating "data center growth +20%" as a single input ever did, and more than any single quarter's results probably will.

Here’s the short version before I walk through each region: the US is weaker than the relationship history suggests, and that’s fine, it’s a vendor-category problem with a visible solution. India and Indonesia are real but small, with a kind of “established” status I can’t translate into market share. China is the risk almost nobody should be pricing, because it isn’t there to price. Europe and the Middle East are the gap that actually matters going forward, because that’s where $68 billion of demand sits with zero TIC footprint and no acquisition target on the table yet.

The number everything else gets measured against

Q1 2026 call, May 6. CEO Ben Heraud: data center revenue around $80 million trailing twelve months, backlog of similar size, against roughly $2.1 billion of consolidated revenue. About 5% of the total. Three months earlier, on the Q4 2025 call, the full-year 2025 figure had been close to $70 million, with management already pointing to “line of sight to nearly $100 million.” That’s a business that doubled in a year, and I don’t want that buried under everything that follows, because it’s the one piece of this that’s unambiguously good.

The Investor Day deck also puts data centers at about 9% of the $714 million Consulting Engineering segment, which works out to roughly $64 million, lower than the $80M consolidated number. That’s not an inconsistency, it’s a denominator difference: the 9%-of-CE figure leaves out data center-adjacent work sitting in Inspection & Mitigation, like the weld inspection and NDT contracts the deck cites separately. Once you line up the denominators, data centers are 3.5%-4.5% of consolidated revenue. $80 million is the figure management itself uses when asked directly, and it’s the one everything below needs to be measured against before it gets measured against a TAM slide.

$80 million against $120 billion of projected 2030 market size in North America alone. That ratio is the lens for the next section.

The thing I keep noticing about this number is how small it is relative to how much airtime it gets on every call. Management mentions data centers more than any other CE sub-segment, and I get why: it’s the line that doubled, it’s the line with the AI story attached, it’s the line that’s easy to talk about with energy. But $80 million inside a $2.1 billion company is the kind of number that makes me want to mentally file it under “watch, don’t weight” until it’s at least double what it is today. I don’t think management is doing anything dishonest by talking about it this much. I think they’re doing what any management team would do with the one growth line that’s genuinely exciting, and my job is just to not let their enthusiasm migrate into how much I’m underwriting.

UNITED STATES

87% percent of Consulting Engineering’s revenue is American. Q1 2026 CE growth, 9.5% to $187 million, was explicitly “led by data centers.” NV5’s international data center expansion into Korea, Thailand, Japan, and Taiwan happened, per the division’s COO Gary Hui, as “a direct result of requests from U.S. technology sector clients to support their Korean hyperscale data center infrastructure investments.” The American hyperscaler is the one asking NV5 to follow it abroad. I want that on the record before saying anything skeptical about the US position, because the easy bear case, that TIC built its data center identity overseas instead of at home, doesn’t survive contact with that fact.

What does survive contact is a scale gap that has nothing to do with relationships and everything to do with vendor category. Building a data center runs through three distinct layers of provider, and where TIC sits in that stack matters more than how big the company is overall.

The first layer is the physical build: the general contractors and EPC firms who pour the concrete, run the cable, and stand up the building. AECOM, Holder Construction, Jacobs, Turner, DPR, Whiting-Turner, Fluor, Mortenson, EQUANS, and Collen top that list globally by capacity delivered. TIC doesn’t show up here, and it never will, because TIC isn’t trying to be a general contractor.

The second layer is design and engineering, the firms that spec out the power, cooling, and structural systems before anyone breaks ground. This is the layer TIC's commissioning and owner's-engineering work actually sits next to, and this is the one where the absence should bother a long-term shareholder. TIC fields 700+ data center-focused professionals across 10 countries, a number that sounds substantial until you put it against Jacobs, whose total backlog hit $22.7 billion at the close of fiscal Q3 2025, with CEO Bob Pragada describing the data center pipeline as multiplying 5x in a single quarter. AECOM's own marketing materials advertise "11+ GW built across 45 countries over 25 years." WSP bought a 175-person specialist, kW Mission Critical Engineering, in 2021, and said at the time that the deal put it "among the top five engineering firms serving data center clients." TIC doesn't show up on Building Design+Construction's 2025 Giants 400 report either, which ranks Burns & McDonnell, Jacobs, WSP, HPE, Woolpert, Kimley-Horn, Olsson, and EXP as the largest US data center engineering firms by 2024 revenue, the most recent year the report covers.

McKinsey's 2025 report on the global data center build-out gets at why TIC isn't on that second list yet. Hyperscale procurement "no longer starts with an RFP," the formal bid process where a project gets opened up for competing proposals, "it starts with data center builders soliciting suppliers that can engage at the semiconductor-engineering level," and "those that wait for the RFP may find the opportunity has already been allocated, without them." In plain terms: the vendors already inside the relationship get the work, and everyone else finds out too late that there was never an open competition to begin with. That rewards incumbents like Jacobs, or the private specialist Salute, which advertises commissioning work for Digital Realty going back to 2014.

Salute sits in a third layer of this stack that neither table above is built to capture: independent commissioning and verification, checking the work after the EPC firm built it and the design firm specified it. You won't find Salute on a builders list or an engineering-firms ranking, because it doesn't build and it doesn't design. It signs off on whether what got built actually works. This is TIC's actual category. And even here, TIC is still early. Its American data center book is built almost entirely from recent bolt-ons, Kisebach and Senergy BCS in October 2024, Herman Cx in March 2025. The multi-year track record that gets a firm pre-qualified before the RFP even shows up still has to be built, one project at a time.

From the $80M base, at the 20%+ organic growth NV5's data center business has historically run, $120-200 million within 2-3 years is a defensible path. That's not nothing for CE's margin profile, since data center work scales at gross margins close to the segment's 47%, but it's also nowhere near a transformation of the $2.1 billion company. Getting the US position from "early" to inside the top five against Jacobs or AECOM is a different question entirely. The comps point to one or two platform acquisitions, 300-750 specialists each, somewhere around $300-500 million combined. That's most of what TIC says it deploys on M&A in a single year, $100-150 million. I don't think that happens inside the current re-rating window. I think it happens, if it happens, after.

INDIA and INDONESIA

“Mature,” in the company’s own framing, feels more confident than what I could actually verify.

The story is entirely acquisition-driven. Industrial Design Associates International, bought in January 2021 (Hong Kong-headquartered with India offices). AT Advanced Technologies, December 2021 (Singapore/Hong Kong/India). Red Technologies, August 2023, just 37 employees at the time, bringing NV5’s regional headcount to “over 250 employees in Asia Pacific.”

NV5 has been commissioning data centers in the region since 2014, per its May 2025 release. That’s a real decade of presence, with projects in both India and Indonesia. The India commissioning team posted on LinkedIn that it completed integrated systems testing of “more than 62 MW” in a single month for “multiple global hyperscalers.” If that’s accurate, it’s not a rounding error.

I couldn’t find much outside the company confirming any of it. No trade press, no industry awards, no operator-side mentions from names like NTT, Princeton Digital Group, or AdaniConneX. Everything external stays at “confidential hyperscaler” level.

The addressable engineering and commissioning fee pool across both countries, stripped out from broader construction capex, runs $120–370 million a year. Against that, TIC’s exposure looks like low-single to low-double-digit millions, competing with AECOM, Jacobs, WSP, Arcadis, Aurecon, and Arup, all with established regional practices.

What this implies for the thesis is narrower than the company framing suggests. It’s a real, recurring, high-margin niche built on independent commissioning relationships that matter precisely because they sit outside design and construction. It is not evidence of dominant share, and I can’t size its contribution to CE growth precisely, since TIC doesn’t disclose India or Indonesia revenue separately.

My best estimate is low-to-mid single-digit millions annually, growing with the region (India 9 GW by 2030, Indonesia 2 GW), but not large enough to move consolidated growth in a meaningful way.

What stands out is that this is probably the closest thing TIC has to a moat in data centers anywhere in the world, yet it barely features in how the company talks about itself. Independent commissioning only works because the firm doing it has no stake in whether the design or the construction was any good, and that kind of trust takes years to build on both sides. TIC clearly has that history here.

The question is whether management sees it as strategic, or just incidental.

CHINA: the geopolitical risk that turned out not to be one

I went into this workstream expecting to find a geopolitical overhang somewhere in TIC’s APAC numbers. I didn’t find one, and the evidence against it is unusually clean for something that’s normally ambiguous.

NV5 founder Dickerson Wright, Q3 2024 call: “the client base is all of the Western Company hyperscalers. We specifically can’t mention them by name, but they are household names.” Every concrete expansion since then sits in APAC markets outside mainland China: South Korea via SA Bricks (May 2025), Thailand’s $6 million contract for a US tech client, Japan and Taiwan (April 2024).

The only projects that explicitly mention “China” in company materials are in Hong Kong: China Mobile Data Centre Hong Kong, iAdvantage, Global Switch. Regional offices are in Kwun Tong, Kowloon, alongside Dubai and Kuala Lumpur. No presence in Beijing, Shanghai, or Shenzhen.

TIC’s FY2025 10-K lists 23 employees in China, but only in the context of the CEO pay-ratio disclosure, alongside a wide mix of other countries. No functional breakdown. It’s plausible this is legacy support tied to earlier acquisitions, but I can’t confirm that.

More importantly, neither NV5’s FY2024 10-K nor TIC’s FY2025 filing includes a dedicated China risk factor. “China” appears only twice in TIC’s filing, both in relation to US tariffs, not operations. For a meaningful China exposure, that omission would be unusual. For marginal exposure, it’s exactly what you’d expect.

China’s regulatory environment makes the alternative implausible on its own terms. Critical Information Infrastructure rules require government approval before any vulnerability assessment, and foreign hyperscalers operate through minority structures with local partners holding certifications they don’t have. Capital has been exiting the sector, not entering it: Bain Capital’s sale of its Chinese data center operation, Chindata, for roughly $4 billion, in a deal that closed in January 2026 and was reported at the time as the largest M&A transaction in the history of China’s data center industry.

The implication is that I should stop discounting APAC growth for China-specific geopolitical risk, because there isn’t meaningful China exposure to begin with. It also means TIC’s “$80B+” APAC market figure likely embeds a large China component that’s structurally inaccessible.

The addressable market is therefore closer to APAC ex-China: Singapore, Japan, Korea, Australia, India, Malaysia, Thailand, the Philippines, Indonesia. I don’t have a clean split, and I doubt TIC will ever disclose it at the country level.

What would change this view: a named Chinese hyperscale client, or a dedicated China risk factor in a future filing. Neither exists today.

I went in expecting China to be the key geopolitical discount. I came out writing the opposite conclusion. That reversal is worth sitting with, because it says something about how these narratives form. A clean bear case often dissolves once you check where the actual operations are.

I’m not carrying China risk in the model. That feels correct, though clean conclusions like this are exactly the ones I’ll want to revisit in six months.

EUROPE and the MIDDLE EAST

This is the one I keep coming back to, because it’s the only region where the absence isn’t explained by either a vendor-category problem (like the US) or a confirmation problem (like India/Indonesia). It’s just absent.

$60 billion projected for Europe, $8 billion for the Middle East, both by 2030. Combined, that’s bigger than all of APAC. TIC’s footprint there is close to nothing: a residual trace from the 2021 IDA acquisition, which brought commissioning work “across Asia and Europe” in 18 countries but never converted into a dedicated European office or team, and literally nothing identifiable in the Middle East, no office, no employee, no named project, in any company material or third-party source I could find.

I went looking for a regulatory reason this might be true and didn’t find one. The US-UK Mutual Recognition Agreement, signed August 2024, is making it easier for US-licensed engineers to practice in the UK, not harder. What actually explains the gap is commercial, not legal: entrenched local relationships, Uptime Institute Tier-designer accreditation, and framework agreements with hyperscalers and sovereign developers like Humain or G42 that TIC simply hasn’t built. When Kent wanted into Middle East data center engineering, it didn’t build organically. It bought Sudlows in February 2025, calling it its strategic entry into the international data center engineering market, and rebranded the unit Kent Data Centres in November with 400+ specialists. I think that’s the only realistic path into either region for TIC as well, not a multi-year organic build.

Management’s own M&A framework reserves $100-150 million a year, with “selective international beachheads driven by customer demand” as an explicit deal category, but no named target, no rumor, no analyst note pointing to anything specific in Europe or the Middle East has surfaced. For the thesis, this is the cleanest binary signal in the entire data center story: a platform acquisition above roughly $200 million enterprise value that brings a European or Middle East entry would be real evidence that management is willing to spend M&A capital on this optionality rather than reserving it entirely for the core TICC integration. Continued silence on this front through the next two or three Investor Days would tell me the data center growth story stays a North America-and-legacy-APAC story, full stop, and that the $68 billion of EU/ME demand on the slide is decorative.

This is the one region where I think the company is making a capital allocation choice rather than running into a wall, and that’s a different kind of risk than the one I started out worried about. TIC isn’t locked out of Europe or the Middle East by anything structural, the way it’s somewhat locked out of the US top tier by procurement habits, or the way India and Indonesia are capped by a fee pool that’s just not that big. It’s locked out because leadership hasn’t spent the money yet, and every dollar they do spend on integration, on debt paydown, on the core TICC business competes directly with a hypothetical Sudlows-style deal. I don’t fault them for prioritizing the balance sheet over a beachhead acquisition this early in the integration. I just don’t think investors should treat the $68 billion gray zone on that map as upside until management actually writes the check, because right now it’s a TAM slide wearing the costume of a growth plan.

What this changes in the numbers I’m actually using

None of the five regions move my re-rating timeline. The thesis was never built on data centers closing the multiple gap to Bureau Veritas or Intertek; it was built on leverage coming down and the combined entity proving out a track record over 8-12 quarters. If anything, that gap got a real-world data point in June, when EQT agreed to take Intertek private at roughly 12x EV/EBITDA, right inside the 11-14x range the thesis already used. That's not the market guessing what a TICC peer is worth. That's a private equity firm writing the check at that multiple.

Data centers sit inside CE's growth rate, and CE's growth rate is one input among several feeding into the stock's three-year return scenarios out to 2028: Bear (+34% total return from here), Base (+177%), Bull (+346%). Below, I translate the four regions into a specific revenue range for the segment under each of those scenarios over the next two to three years, and lay out the one signal per region that would actually move me off this framing, the things worth watching instead of the quarterly growth headline.

Read more

One thing to keep in mind: the model reflects what I know when the thesis is published. If the facts change, the numbers change too, and the update becomes the version that matters. Whenever I publish one, I'll link it here.

1. INVESTMENT CASE

Those who follow Colubeat ID regularly know the kind of companies I’m drawn to: assets that can’t be replicated easily, businesses with a clear structural advantage that I can understand without a PhD in financial engineering, and situations where the market is pricing something at a discount that I can explain in concrete terms.

Rayonier looked like a standard timber REIT at first glance. Stable cash flows, slow-moving assets, nothing unusual.

But then you run into something harder to ignore: in 2023–2024 the company was selling land to private buyers at 44x 3Y timber EBITDA, while the public market implied something closer to ~18x. That gap has persisted for years.

I wanted to understand why.

The more I looked, the more it stopped looking like a “timber REIT” and started looking like a land business in the middle of a structural transition. Asset sales, New Zealand exit, merger with PotlatchDeltic, and a real estate platform that behaves very differently from the timber segment.

The thesis comes down to four things: land value, development optionality, wood products cyclicality, and the private–public valuation gap.

RYN trades around $21 as of June 2026, ~38% below its 2022 highs, while the underlying land base hasn’t really gone anywhere.

2. BUSINESS MODEL: How Rayonier Actually Makes Money

Rayonier is structured as a Real Estate Investment Trust. In exchange for distributing at least 90% of its taxable income to shareholders, it pays no federal corporate income tax on qualifying timber and real estate activities.

The land is the real estate asset. The trees are what the land produces. And crucially, the land doesn’t depreciate the way a factory or a piece of software does. The same acre can generate timber income while the trees grow, a solar or hunting lease in the interim, and a real estate sale when urban expansion eventually reaches it. Rayonier can shift between uses as conditions change, and has been doing that deliberately for decades. That’s what makes this business structurally different from a pure timber operator on one end, or a pure residential developer on the other.

As of early 2026, Rayonier operates across 4 reporting segments:

Each responding to different cycles and valued by different metrics. Section 4 goes deep into each one.

3. INDUSTRY & COMPETITIVE POSITION

In the timber business itself, Rayonier isn’t really protected in any meaningful way. A mill in Georgia doesn’t care whose trees they’re buying. When sawtimber prices fall, they fall for everyone, Rayonier, Weyerhaeuser, private landowners. No brand, no structure that smooths the cycle. It’s a commodity business.

What matters isn’t the timber. It’s the land underneath it.

The U.S. South can’t be expanded. The pine belt from the Gulf Coast through the Carolinas is fixed geography. Loblolly pine gets to merchantable size in 21–28 years, which is fast by global standards, and that’s why capital has been sitting in this region for decades. Rayonier now owns 3.16 million acres of it. You don’t really replicate that, not at scale, not in the right locations.

Everything in the portfolio is constantly being re-evaluated for its highest and best use: timber, or something else. Some parcels sit above CO₂ storage formations. Others sit in the path of urban expansion along Florida and Georgia corridors and get flagged years before anything happens.

You can actually see this showing up in the data. In 2015–17, rural HBU land sold at a 55% premium to the NCREIF South timberland benchmark. In 2021–24, that had widened to 117%. The land didn’t change. The selection and timing did. That improvement in per-acre realization is what section 4.4 gets into in detail.

4. SEGMENT ANALYSIS

4.1 Southern Timber

3.16 million acres across Alabama, Arkansas, Florida, Georgia, Louisiana, Mississippi, Oklahoma, South Carolina, and Texas. About two-thirds of that is intensively managed pine plantation: Rayonier plants a genetically improved seedling, thins and fertilizes the stand over 21 to 28 years, then cuts and replants. The land never sits idle.

The two products

Pine sawtimber (the larger, older wood) goes to sawmills and becomes dimensional lumber. It's the more valuable product and it moves with housing. Pine pulpwood (the smaller wood from early thinnings) goes to paper mills and packaging plants. Worth much less per ton, responding to entirely different demand drivers. Same forest, two very different businesses inside it.

The table above has the numbers. What it doesn’t show is the direction, or why.

Sawtimber prices fell year-over-year. Part of that is the market; part of it is geography, PotlatchDeltic’s acreage prices lower on average than Rayonier’s legacy Georgia and Florida core, so the merger itself pulled the weighted average down. Pulpwood is worse. Delivered pulpwood dropped from $37.83/ton in Q1 2025 to $30.20/ton in Q1 2026. That’s not a rounding error.

Three things drove it, and they’re not all the same kind of problem.

The structural overhang. When housing collapsed in 2008, landowners across the South stopped harvesting. Trees kept growing. Today the South holds 4.4 billion tons of standing private pine, growing 40% more per year than it's cutting. Forisk projects that imbalance only eases to 1.2x by 2029. There's no quick fix for a region that spent fifteen years overgrowing its own demand.

The pulp mill wave. Between 2023 and 2025, more than ten major facilities permanently closed, around 16% of Southern pulping capacity in two years. The shift toward recycled fiber drove the closures. That demand isn't coming back. The mills are gone.

Hurricane Helene. In late September 2024, Helene crossed 8.9 million acres of Georgia forestland and caused $1.28 billion in timber damage. The salvage wave that followed flooded local markets through most of 2025. Some Georgia landowners received $2–3/ton for salvage wood. South Georgia pine pulpwood fell 40% year-over-year in Q4 2025.

Here’s the part that tends to get lost in the South-wide numbers. Helene was the worst short-term event for Rayonier’s Georgia portfolio, and it may also be a medium-term positive for the same geography. The salvage window closed by late 2025. The permanently destroyed inventory doesn’t grow back. North-Central Florida pine sawtimber was up 14% year-over-year in Q4 2025. North Georgia was up 20%. The structural overhang still caps any broad South-wide recovery. What Helene changed is the local picture in the specific markets where Rayonier is concentrated. Both things are true at the same time.

FY2025 Adjusted EBITDA was $130 million, down 14% from 2024, the third consecutive year of compression. Non-timber revenue ($47 million, including $11.2 million from CCS and solar) provides a floor that doesn't move with lumber prices. Q1 2026 came in at $45.5 million on 2.78 million tons, the 76% volume jump was almost entirely PotlatchDeltic acreage arriving.

One more factor worth naming, though it doesn’t change the near-term picture. Loblolly pine productivity has been drifting upward at around 0.3–0.5% per year, documented across 113,000 forest plots between 1999 and 2020 (Hogan et al., PNAS 2024). Half from stand maturation, half from CO₂ fertilization effects. Compounded over a 25-year rotation, that’s a 10–12% increase in sustainable yield per acre. The catch: the CO₂ effect is partly cancelled out by the fertilization Rayonier already applies, both compete for the same nitrogen pathway, so they don’t stack. I’m using 0.4%/year in the model, not the 30% some projections assume. It’s enough to say the NAV doesn’t erode over time. I wouldn’t build a bull case on it.

The diagram below breaks this down visually, left to right, from planting to harvest to the constraint that caps the upside.

When does this get better.

Southern timber price recovery runs through one chain:

Mortgage rates down → housing affordability up → single-family starts up → Southern Yellow Pine lumber demand up → sawmill utilization up → log demand up → stumpage prices up.

The chart below shows where that chain is stuck right now.

The blue line is single-family housing starts since 1997. The red dashed line is the 30-year fixed mortgage rate. The relationship is hard to miss. Rates rise, starts fall. Right now both lines are telling the same story: rates at 6.3%, starts at 930,000 SAAR in April 2026, well below the 1.0-1.1 million level where sawmill utilization begins to recover. Fannie Mae projects 919,000 for full-year 2026. Essentially flat.

The 4M home deficit (Realtor.com) is real. The demographic demand is real. None of that translates into building activity until buyers can afford to buy, and at 6.3%, they can’t, at scale. Even when starts recover, sawmill utilization needs to cross back above roughly 80% before mills start bidding aggressively for logs. Look at the chart: the last time starts held above 1.1 million for an extended period was 2004-2007.

The realistic base case is sawtimber flat South-wide through 2026, localized firming in the Helene-affected markets, meaningful tightening a 2028-2030 story. Pulpwood may never recover to 2022 levels. The mills that closed aren’t coming back.

What Rayonier can do while it waits. The sawtimber share of total pine volume rose to 56% in Q1 2026 from 43% in Q1 2025. Part of this is deliberate, Rayonier has been deferring pulpwood harvest where prices don’t justify cutting. But the larger part is geographic: PotlatchDeltic’s acreage carries a structurally higher sawtimber mix than legacy Rayonier, so the merger itself shifted the portfolio composition. Both effects are real; the merger effect is the bigger one.

U.S. South mills have moved from 26% to 40% of North American lumber production since 2010. All-in cost sits around $301/MBF, the lowest-cost region on the continent. That doesn't fix stumpage prices in 2026. It does mean the South is where the lumber recovery lands when it arrives, and Rayonier's 3.16 million acres sit inside that geography.

4.2 Northwest Timber

Northwest Timber isn’t a larger version of Southern Timber.

The species are different, the rotation cycles are longer, and the buyers are different. A Douglas-fir log harvested after 40 years on the Oregon coast is a larger, denser, higher-grade piece of wood than loblolly pine cut at 25 years in Georgia, and the market prices it that way. $94.37/ton delivered for Northwest domestic sawtimber in Q1 2026, versus $44.59/ton for Southern pine. These aren’t interchangeable products.

Post-merger, the segment covers 930,000 acres: 623,000 in Idaho, 301,000 in Washington, 6,000 in Oregon. Idaho is now 67% of the segment, and as we’ll get to, that shift changes the risk profile considerably.

The Pacific Rim dimension

When Chinese or Japanese buyers want logs, they compete directly with domestic mills for PNW timber, pulling regional prices higher. That tension has largely collapsed. China was the dominant buyer (60% of total U.S. offshore log exports at peak) but its property crisis gutted construction-driven demand. Total Chinese log imports fell from 63.57 million cubic meters in 2021 to 31.38 million in 2025. In March 2025, China suspended U.S. log imports entirely as a trade retaliatory measure; exports fell 95% in the January/August 2025 period.

China lifted the ban in November 2025, tied to the Trump–Xi meeting in South Korea. Management cited this as a 2026 catalyst. The optionality is real. But a property market recovery isn’t expected before 2027, and without that, log demand doesn’t follow. I’d treat China as upside if it materializes, not a number in the base case.

The Canadian supply story

This is the structural factor that actually matters for the segment right now. Canadian lumber competes directly with PNW lumber in U.S. markets. The more of it that stays out, the more PNW mills need to source locally, and the more aggressively they bid for PNW logs.

BC’s timber base has been hit from multiple directions at once: mountain pine beetle kill, indigenous land rights restrictions, old-growth protection policies, and punitive import duties. The provincial harvest fell from around 90 million cubic meters annually to around 40 million over four to five years. Since 2020, 16 sawmills, three pulp mills, and four paper mills have closed in BC. Fastmarkets projects BC’s share of North American lumber production will fall from 22% in the early 2000s to under 10% by 2027.

On top of that physical collapse, U.S. tariffs added a financial wall, currently above 45% all-in, with a preliminary ruling in April 2026 suggesting it could fall to around 35% when finalized in August. Even at 35%, Canadian lumber doesn’t suddenly get cheap.

The diagram below shows both pressures separately, the physical collapse on the left, the financial wall on the right. Independent causes, same effect on PNW log demand.

As BC mills curtail and close, PNW mills gain market share and run at higher utilization. More utilization means more aggressive bidding for PNW logs. The 2025 annual report noted that PNW sawmills "benefited from the continued curtailment of competing lumber supply from Canada." The data supports it.

What PotlatchDeltic’s Idaho acreage changes, and the risk it adds

The Canadian story is good for Northwest Timber. The Idaho story is more complicated.

The 623,000 Idaho acres more than tripled the segment, but Idaho operates on different terms. Around 75% of Idaho sawlog sales are indexed to lumber prices on a four-week lag under long-term supply agreements. That means Idaho revenue doesn’t follow regional log market dynamics. It follows Western SPF 2x4 lumber prices directly. The Q1 2026 10-Q is explicit: the merger “has increased the segment’s overall sensitivity to lumber price movements.”

Western SPF 2x4 was trading near $380/MBF in early 2026. Most BC mills need $450–$500/MBF to break even, which is why curtailments continue. If lumber stays below that range, BC supply stays constrained and Idaho revenue holds. If prices climb above $500/MBF (which would happen in a genuine housing recovery) some BC capacity restarts, adding supply and softening prices at exactly the moment Idaho’s indexed revenue would otherwise be peaking. The upside and the risk arrive together.

The flow below makes the transmission mechanism concrete: from housing demand to lumber price to Idaho log revenue, with the four-week lag that connects them.

I’d run both outcomes (BC curtailments holding versus BC capacity restarting) in any model of this segment.

There’s a third risk that gets less attention. Federal lands in Idaho were a major timber source for decades until environmental restrictions cut that flow. Executive Order 14225 (March 2025) explicitly targets restarting federal harvests. Management’s own filing is direct: more federal timber on the market would depress prices and hurt results. The near-term obstacles are real (road access, legal challenges, mill calibration) but this is a 3-to-5-year story, not a 12-month one. If it materializes, it adds supply pressure in the exact geography where 75% of Rayonier’s Idaho sales are already indexed to lumber prices. That connection is direct.

Where the segment stands

The numbers are in the table above. What the table doesn't show is the forward picture: management expects Q2 volume of around 500,000 tons as Idaho spring break-up conditions ease, with higher sawtimber prices. Full-year 2026 volume guidance is 2.0–2.3 million tons. There's no EBITDA figure in that guidance, Idaho's lumber-index exposure makes any profit forecast dependent on where lumber prices land, and management isn't guessing at that.

4.3 Wood Products

Wood Products didn't exist inside Rayonier before January 30, 2026. The entire segment came through the PotlatchDeltic merger: six sawmills in Waldo, Warren, and Ola (Arkansas), St. Maries (Idaho), Gwinn (Michigan), and Bemidji (Minnesota), plus one industrial plywood mill in St. Maries. Combined nameplate capacity: 1.155 billion board feet of lumber per year and 150 million square feet of plywood. That makes Rayonier a top-10 U.S. lumber producer by volume.

Read more

A few weeks ago I was going through GATX’s annual report, a company I have been looking at closely. I was reading the usual numbers when one line kept pulling me back. Not the revenue. Not the profit margin. The depreciation charge. And what happened every time GATX sold one of its old railcars.

That observation became this article.

There is a business in this issue that reported $333.3 million in profit last year. That same year, it charged itself $431.8 million for how much its assets aged and wore down over the year. On EV/EBITDA, the ratio that strips depreciation out of earnings before measuring value, those two facts make the company look like it earns far more than the accounting admits.

I want to show you why, in the right business, that conclusion is completely backwards.

There are companies that are better than their own numbers suggest. They don’t always look cheap on the standard screens. The signal, when you know where to look, tends to sit in a corner of the annual report that most investors skip: the line that records what the company got when it sold something old, and whether that price was higher or lower than what the accounting said the thing was still worth.

The guess in the footnote

Every long-lived asset, a factory, a railcar, a server, a plane, comes with a problem the accounting has to solve. The company paid for it upfront, but the asset does not stop working on day one. It keeps running for years, sometimes decades, and slowly gets less useful over time. So accounting spreads the original cost across the asset’s estimated life, charging a slice each year. That annual slice is depreciation.

The key word is estimated. Someone inside the company has to decide two things: how many years the asset will stay useful, and what someone would pay for it at the end of that period, when it is old and worn. Those two guesses determine how large the annual depreciation charge will be. Choose a shorter life and more depreciation flows through the accounts each year, which means lower reported earnings. Choose a longer life and less flows through, which means higher reported earnings. Same asset. Same purchase price. Different guess. Different number on the page.

Most investors see this line labelled as a non-cash expense and move on. Non-cash means no money is leaving the bank account this specific year, which is true. But the money already left when the company bought the asset, and it will leave again when the asset needs replacing. The label is true. The conclusion most people reach from it, that the expense is not real, is wrong. A building that needs a new roof in ten years has been costing its owner money every year for the past decade. The accounting is just the company acknowledging that, one year at a time.

The scoreboard

Every estimate eventually gets tested. For depreciation, the test happens in the second-hand market.

When a company sells an old asset, it compares the price it gets to what the accounting currently says that asset is worth. If the sale price is higher, the company records a gain. If lower, a loss. That number is the market telling you whether the original depreciation estimate was realistic.

A company that consistently sells old assets at a gain is showing you something specific: its depreciation was too conservative. The asset lasted longer than the accounting predicted. The books said it was nearly done. A buyer disagreed and paid more. Which means the earnings the company reported each year, after absorbing all that depreciation, were actually lower than what the business was really making.

This matters because the two directions carry very different consequences for the investor. When management charges more depreciation than the asset strictly needs, the reported earnings are lower than the real value the business is creating. When management charges less, the opposite happens: earnings look better than they are, and the bill arrives later, when the assets need replacing and the cash has to go out all at once.

This is not a warning about a common mistake. It is a way to find something.

A business that charges more depreciation than its assets strictly need is quietly understating its own earnings, year after year, in the most verifiable way possible. The second-hand market confirms it every time an old asset sells for more than the books said it was worth. That information is public, sitting in the notes section of every annual report, and it tells you whether the accounting has been careful or generous about what the company actually owns. Almost nobody reads it.

That business is GATX Corporation.

— GATX Corporation ($GATX) —

GATX leases railcars. About 156,000 of them, one of the largest privately owned fleets in the world, most of it steel tank cars carrying chemicals, compressed gases, and food-grade liquids somewhere along a track that most investors will never think about. Total assets of $18.0 billion, made up almost entirely of railcars. Not glamorous by any measure, and exactly the right kind of company to learn from.

In 2025, GATX reported net income of $333.3 million, or $9.12 per diluted share. The depreciation charge that same year was $431.8 million. The single line the standard screens remove is larger than the entire reported profit. Take it out and the company appears to earn more than twice as much. That is the whole appeal of EBITDA, which stands for earnings before interest, taxes, depreciation and amortization: set aside the accounting for how fast the assets are wearing down, and the number looks much cleaner.

The money, though, really went out. GATX spent $1,316.7 million buying new railcars and maintaining the fleet in 2025. The total depreciation charged over the years sits at $4,251.7 million on its balance sheet, the running record of how much of the fleet has aged on paper since the company started keeping count. Steel tank cars wear out. They get scrapped. They get replaced. Depreciation is the company saying so, one year at a time.

Now look at what the sales record actually shows.

GATX’s accounting policy assigns each railcar a useful life somewhere between 15 and 45 years. A new steel tank car, well maintained, can run for 40 or 45 years. An older or more specialised one might have a shorter commercial life. That spread carries real consequences: a railcar written off over 15 years generates a depreciation charge roughly three times larger each year than one written off over 45 years. Same car, same purchase price, dramatically different annual cost on the income statement. Where management places the estimate within that range is the decision that matters.

In 2025, GATX booked $140 million in net gains on asset sales. In 2024, $137.8 million. In 2023, $130.9 million. Three years in a row of selling cars the accounting had nearly written off for more than the books said they were worth, with gains holding consistently between $130 and $140 million across very different market conditions.

Read those numbers the other way around. Consistent gains on asset sales mean the depreciation was too heavy. The cars lasted longer than the estimate said they would. The accounting was making the business look worse than it actually was, quietly, every single year.

GATX’s $431.8 million depreciation charge, the line that disappears inside an EBITDA calculation, is the financial fingerprint of a management team that has spent years placing its estimates on the conservative side of that 15 to 45 year range. They charged more than they had to. The old cars kept selling above what the books said they were worth. The pattern keeps repeating.

A fair point, and I will say it directly: the gain on asset disposals is a mixed number, and separating the parts matters. Some of it comes from a strong secondary market for railcars. Leasing utilisation has been high, renewal rates have climbed, and tight supply has pushed prices up across the industry. In a market like this, any owner of used railcars is going to sell above book. That part of the gain tells you more about industry conditions than about GATX specifically.

But there is a structural element that survives the market argument. GATX’s accounting policy sets residual values at the low end of what a used car might realistically fetch. That is a policy choice, written into the footnotes, not something inferred from results. It means the book value floor is built to be cleared regardless of where the market is. An ageing tank car in average market conditions will typically sell above a conservatively set book value. In a strong market it sells even further above. The gains visible in the filing mix both effects, and it is honest to admit that the exact split between market conditions and accounting conservatism is not something I can answer from the outside.

What I can say is this: the policy is conservative by design, the gains have held consistently across three years and different market conditions, and the pattern fits a management team that would rather report less today than promise more and correct later. That combination is worth paying attention to, even if the precise weight of each factor is something only GATX’s internal numbers could settle.

The mirror image

The same thing can run in the other direction, and that direction is where the problems are.

Some companies stretch the estimated life of their assets on purpose. Less depreciation flows through each year. Reported earnings go up. The business does not change. And the gap between what the books say and what reality will demand keeps growing.

In recent years, several large technology companies extended the estimated useful life of the computer servers that run their services, from four years to five or six. The servers did not improve. They did not last longer. The estimate changed. The depreciation charge fell. The profits went up. The market has not had its say yet, but the clock is running.

GATX charges more than it has to and shows its quality in the asset sale record. Companies that charge less than they should hide a future problem in the same place. The test is the same. The story is not.

Three tests worth running on any company with significant physical assets.

Test 1. Is depreciation larger than net income? When a company leads with EBITDA and its depreciation charge is bigger than its reported profit, ask whether that heavy charge reflects careful accounting or a business that simply burns through a lot of assets. At GATX: $431.8 million in depreciation against $333.3 million in net income. A business that runs on steel tells you so in the notes.

Test 2. What does the asset sale record say? Look for three years of gains or losses on asset sales in the footnotes of the annual report. Consistent gains mean the depreciation was conservative and the assets lasted longer than the accounting said they would. Consistent losses, or a sudden large write-down, mean the depreciation was too light and the bill is arriving all at once. GATX: $140 million, $137.8 million, $130.9 million. Three years in a row, consistently above $130 million. Three years of the same answer.

Update, 17 june: a reader pointed me to Heartland Express ($HTLD), a trucking company that also books consistent gains on equipment sales. Worth digging into, because the reason turns out to be different from GATX. HTLD depreciates its tractors using the declining balance method, which front-loads the charge: depreciation is high in the early years and tapers off later. That makes book value fall faster than the truck's real market value early on, which is what lets the company sell at almost any point in a tractor's life without booking a loss. HTLD says this directly in its annual report, and even notes it can add volatility to gains and quarterly earnings. The gain, in other words, is partly built into the accounting choice itself. GATX is doing something different: straight-line depreciation with conservative residual values, so its gains show up because reality beat the estimate, not because the method was designed to produce them. Same line on the cash flow statement, very different story underneath. Test 2 needs one more question: not just whether the gains exist, but how the company depreciates to get there. Thanks to the reader who brought this up.

Test 3. Has the useful life estimate quietly changed? One small change in the footnotes that extends an asset’s estimated life lowers this year’s depreciation and raises this year’s earnings, with nothing changing in the real world. Look for the useful life assumption in the notes, compare it to the year before, and ask what actually changed to justify the revision. If nothing changed, the earnings improvement is accounting. The answer has not come yet.

Depreciation is not the cost of an asset wearing out. It is a guess about how fast it will. The price someone pays for it second-hand is the day you find out whether the guess was honest.

GATX has spent years placing its depreciation estimates on the conservative side of a range that runs from 15 to 45 years. The proof is what someone paid for an old tank car, reliably, year after year, above what the books said it was worth. The standard screens hand that $431.8 million back and call the company cheap. The asset sale record tells a different story: a business that has been understating its own quality, carefully and consistently, for years.

The cars still wear out. The cash still goes out the door. And sometimes the line that looks most like a cost is the clearest way a business has of telling you it is better than it looks.

If you made it here, take this test with you. The next time you sit down with an annual report for a capital-heavy business, find what the company got when it sold something old. Three years of numbers will tell you more about how management thinks about their own assets than almost anything else on the page. That is usually enough to see something EV/EBITDA missed.

See you next month.

Disclosure: I don’t hold a position in GATX ($GATX) as of June 2026. This is an educational piece, not a recommendation.

If you read the USLM investment thesis published last week, you know the business: a debt-free regional lime monopoly in the South-Central United States, exceptional geology, a gross margin of 48.9% partly explained by a Permian Basin gas anomaly, and a regulatory moat that makes greenfield competition essentially impossible. That thesis covers the financial structure, the quarry assets, and the competitive position in detail.

What it doesn’t cover is a specific question that kept surfacing while building it: does the AI infrastructure build-out, the semiconductor fabs, the hyperscale data centers, the $710 billion in announced hyperscaler capex for 2026, actually reach USLM in any meaningful way? And if it does, through which channel, and how much of what people are claiming online holds up when you check the construction documents?

The soil geology of specific sites matters. The distance between a lime terminal and a construction site matters. Who else is closer matters. I pulled the documents, mapped the distances, read the filings, and traced the demand chain from the press releases down to the molecule. This is what I found.

Before the sites, one concept worth pinning down. It takes thirty seconds to understand, and it decides the entire analysis.

When engineers build something heavy on clay soil, the relevant number is the Plasticity Index, or PI. Think of it as a soil’s tendency to move: a PI of 5 is stable sand, a PI of 45 is the kind of expansive clay that cracks highway asphalt every summer as it swells with moisture and shrinks when it dries. The fix, standard practice across Texas, is to till quicklime into the top layer of soil before construction. The lime reacts chemically with the clay and locks it in place permanently. TxDOT, the Texas state road authority, mandates this treatment for any subgrade with a PI above 20, and considers it the only practical option above PI 30. Below PI 12, lime isn’t the automatic answer. Above it, the question isn’t whether to treat but who sold the lime.

With that in hand, two demand stories. The first is the one that actually holds up. The second is the one most people are talking about.

The channel that actually works: electric-arc steel

Before the construction sites, the finding that holds up best. It’s documented, recurring, and doesn’t depend on anyone’s soil classification.

Most steel produced in the United States today isn’t made from iron ore. It’s made from scrap metal melted in an Electric Arc Furnace, or EAF. These facilities, called mini-mills, feed scrap steel into a massive electric furnace, melt it, and cast it into structural shapes: beams, rebar, flat-rolled coil. The melt has to be chemically cleaned before casting. Lime is dropped into the furnace to bind the silica, phosphorus, and sulfur in the scrap and float them off as slag. Every ton of EAF steel consumes 40 to 60 kilograms of lime in this process, and that consumption repeats every time the furnace runs, for as long as the mill operates.

Soil stabilization ends when the building is up. Furnace lime is an annual contract.

The connection to AI infrastructure is documented in primary filings. Data centers require structural steel for their frames, rebar for their foundations, and flat-rolled steel for their mechanical systems. Nucor stated in its 2025 investor presentations that structural steel shipments for data centers are projected to grow 50% between 2024 and 2026, and that data centers represent roughly 10% of its order backlog. Steel Dynamics named “data center” demand explicitly in its Q1 2026 earnings filing, with fabrication backlog up 38% year over year. Commercial Metals Company described data center construction for the AI sector as “driving growth” in its quarterly results. These aren’t inferences from market conditions.

Nucor also published a useful unit metric: roughly 3,500 tons of structural steel per 250,000 square feet of data center. A 500,000-square-foot facility embeds about 7,000 tons of steel and, at 50 kilograms of lime per ton, roughly 350 metric tons of lime in the EAF flux demand it generates.

On bulk lime, the producer closest to the mill wins on delivered cost. Lime is cheap per ton and heavy to move, so freight cost swamps small differences in product price. Here's how that plays out across the cluster.

The Arkansas cluster is USLM’s strongest position, and it’s larger than most write-ups acknowledge. Nucor-Yamato Steel in Blytheville, Mississippi County, produces over 2.5 million tons per year of structural steel: wide-flange beams, H-piling, sheet piling. USLM’s Arkansas Lime plant in Batesville is roughly 100 to 120 road miles away. The nearest Lhoist plant, the main competitor in this market, is in Texas, over 400 miles out. There’s no competing lime producer within 280 miles of Blytheville.

U.S. Steel's Big River Steel campus in Osceola, also in Mississippi County, completed its Big River 2 expansion in late 2025 at a cost of $3 billion, described by U.S. Steel as "the largest private project in Arkansas history, completed ahead of schedule," doubling campus capacity to 6.0 million tons per year of flat-rolled steel. USLM Batesville sits roughly 100 to 110 miles away. Lhoist is hundreds of miles out. At 50 kilograms of lime per ton and 75% capacity utilization, Nucor-Yamato plus Big River together imply roughly 318,000 metric tons of annual lime demand with essentially no competitor within reach. That number was undercounted in earlier write-ups because Big River 2 wasn't yet at full capacity when most of them were written.

Gerdau's Midlothian mill in Ellis County, Texas, sits roughly 25 to 30 road miles from USLM's Texas Lime plant in Cleburne. Gerdau produces wide-flange beams and structural sections, references "data halls" as an end market in its own marketing materials, and announced in late 2025 a capacity expansion adding 150,000 metric tons of EAF melt output in the first half of 2026, with a longer-term target toward 2 million tons total. USLM Cleburne is the closest lime producer. Lhoist Clifton is about 50 miles out.

CMC Steel's micro-mill in Durant, Oklahoma, produces 390,000 tons per year of rebar. USLM's Mill Creek plant in Johnston County is roughly 60 miles away. Lhoist Clifton is around 180 miles. USLM wins on distance. One caveat worth stating: Mill Creek produces dolomitic PLS (a lime product made from dolomitic limestone rather than high-calcium limestone, and not what EAF steelmaking typically specifies). The actual USLM supply point for Durant is more likely St. Clair in Oklahoma or Cleburne via rail. The distance advantage is real; the product-type alignment needs verification.

Nucor’s Jewett bar mill in Leon County, Texas, at roughly 1.2 million tons of capacity, sits approximately equidistant between USLM Cleburne and Lhoist Clifton, both around 120 miles. Call it contested.

Where USLM probably doesn’t win

CMC’s Seguin mill in Guadalupe County, Texas, sits 18 miles from Lhoist’s New Braunfels plant. USLM’s Houston terminal is around 180 miles away. Lhoist’s account.

Steel Dynamics’ Sinton flat-roll mill in San Patricio County, Texas, 3.0 million tons confirmed in SDI’s own filings, sits about 165 miles from Lhoist’s New Braunfels plant, which also operates a terminal roughly 30 miles from Sinton in Corpus Christi. Lhoist wins that one.

The mills inside USLM’s freight radius where no competitor is closer represent roughly 420,000 to 570,000 metric tons of annual lime demand on a defensible distance basis. That range covers the core Arkansas cluster plus Gerdau Midlothian and CMC Durant, excludes the mills Lhoist wins on distance, and uses 75% capacity utilization and 50 kilograms per ton as working assumptions.

Against USLM’s estimated total lime sales base of roughly 3.5 million tons per year, that’s 12% to 16% of volume, recurring annually, growing as the mills expand. USLM confirmed this channel in its Q3 2025 release, attributing volume gains “principally due to the Company’s construction, environmental, and steel customers.”

The Texas fabs: the right soil, the wrong foundation

The soil science here matters for one reason: it’s the basis of the most-repeated claim about USLM and AI infrastructure, and it needs to be checked carefully.

Start with Samsung’s fab in Taylor, Texas.

Williamson County sits in the Blackland Prairie, the geological belt running from Dallas down through Austin, dominated by a clay called Houston Black. The NRCS, the federal soil survey agency, classifies it as a fine clay with 40 to 60% clay content and very high shrink-swell potential. If you pour concrete directly on it, the ground absorbs water, heaves, and physically tears the foundation apart. Published geotechnical literature puts its Plasticity Index in the 40 to 60 range, TxDOT’s treatment threshold is PI 20, and Houston Black sits at roughly double that. A 1,268-acre construction site on this material would, under standard Texas engineering practice, require lime treatment across every road, pad, utility corridor, and detention basin built at grade.

What the Samsung Taylor site actually used for its critical foundation isn’t lime-treated subgrade. Construction crews drilled more than 20,000 shafts, each roughly 110 feet deep, and filled them with over half a million cubic yards of concrete. Those piers anchor the fab’s cleanroom platform directly to bedrock, below the expansive clay entirely. The reason is vibration. A semiconductor fab operates lithography equipment sensitive to movements measured in nanometers. A lime-stabilized subgrade still transmits micro-vibration from passing trucks or ground movement. A pier foundation anchored to rock doesn’t. The part of the site that defines the project, the fab building itself, bypasses the clay and has no lime demand in its critical foundation.

Lime demand at Taylor is concentrated in the peripheral infrastructure: access roads, parking structures, utility trenches, substation footings, laydown areas. Real volume, bounded, one-time.

Texas Instruments in Sherman, Grayson County, is the same geology. TI CEO Haviv Ilan confirmed at the SM1 opening in December 2025 that construction involved “more than two million cubic yards of excavation and more than 240,000 cubic yards of concrete.” Consistent with deep earthwork on expansive clay, though it doesn’t confirm lime use.

For TI Sherman, USLM’s Texas Lime plant in Cleburne is the closest quicklime producer at roughly 115 road miles. Lhoist’s nearest plant, Clifton, is about 155 miles. USLM wins on delivered cost.

For Samsung Taylor it’s the reverse. Lhoist’s Marble Falls plant sits roughly 65 miles from Taylor. USLM Cleburne is around 175 miles. There’s also Austin White Lime, whose McNeil plant in Travis County sits about 30 miles from the Taylor site, closer than anyone. Austin White Lime doesn’t appear in most write-ups on this thesis, but it’s the geographically logical supplier for the Samsung site’s peripheral lime work.

No public construction document, no TxDOT bid tabulation, no earthwork contract retrieved during this research confirms lime tonnage at either fab site. The lime demand is geologically near-certain on the at-grade peripheral work. The supplier identity is unknown.

Louisiana: two assumptions that didn’t survive the documents

The Louisiana leg of the thesis is where the most confident claims circulate online, and where the evidence is weakest. The soil classifications here are specific, but the point they make is simple: the Meta site isn’t on the clay people think it’s on, and that changes who benefits.

The story: Meta’s data center in Richland Parish sits on expansive “Yazoo” clay, lime treatment is mandatory across the site, USLM’s Shreveport terminal makes it the natural supplier, and a clustering wave of hyperscalers along the I-20 corridor turns this into a multi-year demand story.

The Meta site is the former Franklin Farm megasite near Holly Ridge, Richland Parish, roughly 30 miles east of Monroe. It covers 2,250 acres, was developed through a Meta shell company called Laidley LLC under the codename Project Sucre, and is reported as a $27 billion campus funded through a Meta and Blue Owl Capital joint venture. The scale isn’t in dispute. The soil is.

The site sits on the Macon Ridge loess terrace, not on Yazoo clay. Loess is wind-blown silt deposited by wind rather than water, which makes it lighter, less plastic, and far less reactive to lime treatment than the dense smectitic clays of the Mississippi alluvial basin. USDA NRCS soil survey data (the same federal agency referenced in the Texas section) shows the dominant series as Gigger and Gallion silt loams, with documented Plasticity Index values in the 9 to 25 range. Instead of sitting four times above the lime trigger like Houston Black clay, this soil straddles it. Some of it qualifies for lime treatment. Most of it doesn’t automatically.

The genuinely expansive Sharkey and Alligator clays that earn the “Yazoo” nickname, with documented PI values of 35 to 60-plus, exist in the alluvial basins flanking Macon Ridge, not on the ridge itself. The Meta campus is on the ridge. Entergy’s own marketing materials for the Franklin Farm megasite described it as having “above average soil bearing capacity.” That’s not how you describe Yazoo clay.

There’s one data point that complicates the clean negative. An investor who shared publicly on X that they had spoken with USLM’s IR reported that management alluded to having done soil treatment work at the Meta site. The post wasn’t retrievable from public search indexes, and USLM has never confirmed any customer relationship in public filings. I take it seriously as a data point, but it sits at the bottom of the evidence ladder: an unverified second-hand account with no tonnage and no scope. If accurate, it most plausibly describes USLM capturing lime-suitable pockets within a mixed-soil site of 2,250 acres, not supplying a site-wide program that the soil data suggests was never warranted.

The distance: The second problem is distance. The Meta site appears in some versions of this thesis at 56 road miles from USLM's Shreveport terminal. The correct figure is roughly 125 miles. That’s still inside the economic truck radius for bulk quicklime, but it’s not the doorstep proximity the “natural supplier” framing implies. At 125 miles, freight cost is a real variable, not a rounding error.

The competitor nobody mentions: Here’s the data point almost entirely absent from online discussion of the USLM-Louisiana thesis: Mississippi Lime Company operates a plant in Vicksburg, Mississippi, roughly 55 road miles from the Meta Richland Parish site, with barge access on the Mississippi River. At 55 miles versus 125 miles, Mississippi Lime holds a delivered-cost advantage of between $7 and $10 per ton on bulk quicklime, based on standard truck freight economics of $0.10 to $0.15 per ton-mile. On a commodity selling at $150 to $250 per ton delivered, that freight gap matters. If Meta’s site required meaningful lime volumes, Mississippi Lime Vicksburg is the geographically logical primary supplier. Not USLM.

What the Microsoft number is

The “$3.3 billion Louisiana data center” appearing in some coverage doesn’t refer to a Microsoft facility. Microsoft’s $3.3 billion data center investment is in Mount Pleasant, Wisconsin. The $3.3 billion figure attached to Louisiana is the estimated value of Meta’s tax incentive package, a different number for a different thing that got fused into a phantom project. If it’s in your model, remove it.

Three independent problems with the same thesis, and they don’t cancel each other out.

The Louisiana story worth watching is not Meta

On February 23, 2026, Amazon and STACK Infrastructure announced a $12 billion data center investment across Caddo and Bossier parishes in northwest Louisiana. STACK is the developer and owner; AWS is the tenant. Confirmed by Louisiana Economic Development, Amazon’s own release, and STACK’s GlobeNewswire announcement.

Two campus sites are confirmed in public records. The first is in Bossier Parish, on the west side of Louisiana Highway 3, roughly five miles north of Benton, platted as the “Highway 3 Technology Park.” Plans on file with the Bossier Parish Police Jury show eight 400,000-square-foot data center buildings totaling 3.36 million square feet. General contractor is Haskell, out of Jacksonville, Florida. The final plat was approved March 18, 2026, and earthwork was underway shortly after. The second site is in Caddo Parish, north of Blanchard-Latex Road near Blanchard, close to the Texas state line. General contractor is EE Reed, a Houston-based firm. Clearing and dirt work had already begun at the time of the announcement. A possible third site, the Resilient Technology Park on Greenwood Road in west Shreveport, 313 acres adjacent to I-20, was subject to a special-use permit upheld by Caddo Parish in April 2026, with an appeal pending.

The distance from USLM’s Shreveport terminal at 6000 St. Vincent Ave. to the Bossier Highway 3 site is roughly 20 to 30 road miles. To the Caddo Blanchard site, roughly 15 to 20 miles. Compare that to 125 miles for Meta — these sites are essentially at the terminal’s door. No competing lime producer has a kiln within 150 miles. Lhoist’s nearest plant is Clifton, Texas, at roughly 250 miles. Mississippi Lime’s Vicksburg plant, the logical supplier for Meta, is around 190 miles from Shreveport. On delivered cost, USLM holds a freight advantage of roughly $20 to $35 per ton over any competitor for these two sites.

USLM’s Shreveport terminal has quicklime storage, hydrating capability, and slurry production, which is the product form soil stabilization work requires. The terminal is rail-served from USLM’s Arkansas Lime plant in Batesville and Texas Lime plant in Cleburne, so surge capacity depends on upstream rail logistics rather than local kiln output.

The soil question for Caddo-Bossier remains open. Neither site has been publicly characterized at the parcel level in any retrieved geotechnical document. NRCS soil survey data shows both sites sitting on Coastal Plain uplands west of the Red River alluvial valley. The dominant surface soils are Ruston and Bowie, fine sandy loams with PI values generally below 20, which means lime stabilization isn’t automatically triggered across the whole Bossier site.

The Caddo Blanchard site is more interesting. That parcel has higher probability of encountering Sacul and Woodtell series soils in the subsoil, both clayey Wilcox-formation residuals with PI values in the 25 to 55 range. Those do trigger lime treatment. The Bossier site leans low-PI on the surface; the Caddo site near Blanchard is the stronger lime candidate. Neither is the blanket high-clay opportunity the Meta narrative promised. But a 1,000-acre construction site with clay subsoil pockets, built by a GC that has already held local supplier outreach sessions, is a real demand event for the terminal that sits 15 miles away.

Louisiana is also attracting data center investment beyond Meta and Amazon, including Hut 8’s River Bend campus in West Feliciana Parish and Applied Digital’s Delta Forge 1 in Rapides Parish. Northwest Louisiana is drawing a concentration of large-scale construction at a moment when USLM has a terminal positioned to serve it, and the question of how much lime that generates will show up in quarterly construction volumes before management names it explicitly.

What management says, when they say it, and what they leave out

USLM doesn’t host earnings calls. No conference call, no webcast, no live Q&A. CEO Timothy Byrne’s scripted quote in each quarterly press release is the closest thing to management commentary that exists. Every silence is a deliberate editorial choice, not a dodge of a question that was never asked.

Through all of 2023 and 2024, quarterly releases describe demand in standard category language: construction, industrial, steel, oil-and-gas. No mention of data centers, including in Q4 2024, released February 3, 2025, despite Meta having announced the Richland Parish Hyperion project on December 4, 2024.

Q3 2025, released October 29, 2025, is where it becomes explicit. CEO Byrne: “Demand from our construction customers remained solid, supported by the construction of large data centers in the regions that we serve.” The 10-Q MD&A goes further: “a 13.3% increase in sales volumes, principally due to increased demand from our construction, including for the construction of large data centers, environmental, and steel customers.” First time data centers appear as a named revenue driver in a filed document. Q3 2025 revenue was $102.0 million against $89.4 million in Q3 2024, up 14.1%. Q4 2025 repeated the same language, with full-year revenue of $372.7 million against $317.7 million in 2024, up 17.3%.

Q1 2026, released April 29, 2026: revenue fell to $87.8 million from $91.3 million, down 3.7%. Management attributed it to weather disruptions and soft demand from construction, oil-and-gas, and roof shingles. Data centers aren’t mentioned. One quarter after being the headline demand driver, the category vanishes entirely.

In the FY2025 annual report, “data center” appears in MD&A and in Risk Factors as a growth area that “may not continue.” It wasn’t added to Item 1, where USLM lists its seven customer categories, unchanged from 2022. A permanent structural demand category would appear in Item 1. This one doesn’t.

The margin engine is gas, not GPUs

USLM’s gross margin went from 36.6% in 2023 to 45.3% in 2024 to 48.9% in 2025. That’s a large move for a company selling the same product it sold five years ago. The instinct is to credit AI demand or a hot construction market. The documents point somewhere quieter.

Making lime means heating limestone in a kiln to around 900 degrees Celsius continuously. Natural gas is the primary fuel for USLM’s Texas kilns, and the price of that gas is the most important variable cost in the business.

A clarification worth making explicitly: Cleburne is in Johnson County, in the Fort Worth basin, sitting on the Barnett Shale. It is not in the Permian Basin, and it does not physically buy gas at the Waha Hub, which is located roughly 300 miles to the west in West Texas. USLM never names Waha or any specific pricing hub in its filings, and does not disclose its gas contract structure. What follows is an inference from the correlation between regional Texas gas prices and USLM’s margin, not a confirmed fact about how USLM contracts its fuel.

With that caveat stated: the entire Texas gas market is structurally interconnected. The Permian Basin produces enormous quantities of natural gas as a byproduct of oil drilling and has historically lacked enough pipeline capacity to move all of it east. When that supply overwhelms takeaway capacity, prices collapse at the Waha Hub and that pressure spreads, with some lag and some dampening, across the broader Texas grid. Waha prices averaged negative on 107 distinct flow days in 2025, reaching a low of negative $8.79 per MMBtu in October. A plant in north-central Texas does not receive those literal negative prints, but it benefits from structurally cheap regional gas pulled down by the same oversupply dynamic. USLM also says explicitly in its filings that it varies the mix of fuels across coal, petroleum coke, and natural gas depending on relative prices, which adds another layer of flexibility.

The correlation between the Texas gas cost environment and USLM’s margin expansion across 2023 to 2025 is not a coincidence. Q1 2026 offered the first direct signal of what happens when that environment tightens: gross profit fell to $41.8 million from $46.2 million in Q1 2025, a drop of $4.4 million, on revenues that were only $3.4 million lower.

New Permian egress pipeline capacity scheduled for H2 2026, including Hugh Brinson (1.5 Bcf/d, connecting the Permian to the Dallas-Fort Worth area) and Blackcomb (2.5 Bcf/d), will move cheap Permian gas east and compress the regional discount that has benefited USLM’s Texas operations. The normalization effect reaches Cleburne indirectly, through the same grid it benefited from on the way down.

Using a proxy of 1.2 million tons per year for Texas lime output and 3.3 MMBtu of natural gas per ton of lime produced, both flagged as estimates because USLM discloses neither, a conservative normalization of plus $1.50 per MMBtu in regional Texas gas adds roughly $5.9 million in annual fuel cost, enough to compress gross margin from 48.9% toward 47.3%. At plus $3.00 per MMBtu, it’s roughly $11.9 million and gross margin moves toward 45.7%, still above the 2024 level of 45.3%.

Two factors partially offset the normalization. USLM’s new vertical kiln at Cleburne, a $65 million project commissioning in summer 2026, is designed to reduce thermal energy consumption from roughly 4.8 gigajoules per ton on the existing rotary kilns to roughly 3.4 gigajoules per ton, a 29% reduction in energy intensity on Cleburne’s output. And USLM’s Arkansas and Oklahoma plants operate on different regional gas markets, so the margin impact of normalization is concentrated in Texas, not company-wide.

The base case for margin going forward isn’t 48.9%. It’s probably in the 45% to 47% range once Waha normalizes and the vertical kiln reaches steady-state. A forward model anchored on 48.9% is probably 200 to 300 basis points too optimistic.

FINAL THOUGHTS

The semiconductor fab story rests on genuine geology, but the fab buildings use deep pier foundations that bypass the expansive clay entirely. Lime demand at these sites is real, confined to peripheral infrastructure, one-time, and split between suppliers by distance, with USLM favored at Sherman and not at Taylor. No public document confirmed lime tonnage at either site.

The Meta Richland Parish story has three problems: the soil is Macon Ridge loess, not Yazoo clay, with a PI that straddles the treatment threshold rather than sitting far above it; the distance from USLM’s terminal is 125 miles, not 56; and the geographically logical primary supplier is Mississippi Lime’s Vicksburg plant at 55 miles, a competitor almost nobody covering this story has named. The Microsoft $3.3 billion Louisiana data center doesn’t exist. If it’s in your model, remove it.

Two things hold up. Here is what the research implies for each, with the assumptions made explicit.

Caddo-Bossier: near-term, geographically captive, modest in tonnage

The Amazon and STACK campus is the cleanest freight story in this entire investigation. USLM’s Shreveport terminal at 15 to 30 road miles, no competing kiln within 150 miles, general contractors already running local supplier outreach. The lime tonnage from soil stabilization is modest by industrial standards. Using standard Louisiana application rates for mixed-PI terrain at roughly 25 to 30 pounds of lime per square yard of treated subgrade, and assuming 30% of each 1,000-acre site requires treatment (roads, utility corridors, detention infrastructure), each campus generates roughly 1,500 to 2,500 metric tons of lime demand during the earthwork phase. Across two confirmed sites plus a possible third, that’s a total construction-phase demand of roughly 3,000 to 7,500 metric tons attributable to USLM on distance. At an estimated realized price of $105 per metric ton, that’s $300,000 to $800,000 of incremental revenue, modest, one-time, and front-loaded into 2026-2027 as earthwork proceeds. I'll be honest: that's not a number that moves the annual P&L on its own, and I considered whether it's worth naming at all. I think it is, because the freight position is real even if the tonnage is small, and because it's the setup for what comes next if more Louisiana sites follow.

The more important signal isn’t the tonnage. It’s whether Caddo-Bossier shows up in USLM’s quarterly language. Management named data centers when Meta’s earthwork was generating volume. If Q2 or Q3 2026 releases name northwest Louisiana or show a step-up in construction segment volumes, the Shreveport terminal is working. That’s the indicator to watch, not the absolute numbers.

EAF steel: larger, recurring, already in the numbers

This is the durable channel, and it’s already partially reflected in USLM’s revenue. The question is how much it grows as the mills expand. The table below summarizes the defensible EAF cluster using confirmed capacities, estimated distances, and standard lime consumption ratios. All volume and revenue figures are estimates based on publicly available mill capacities and USLM’s estimated realized price; they are not disclosed by USLM or the mills.

*Gerdau at announced expansion target. †Mill Creek produces dolomitic PLS; actual supply point may be St. Clair or Cleburne via rail. Distance advantage holds, product alignment needs confirmation.

At 70% USLM capture across this cluster (allowing for split supply and partial competitor penetration), the revenue attribution falls to roughly $30 million. Against FY2025 revenue of $372.7 million, that’s approximately 8% of the total, recurring annually, and growing as Big River 2 ramps through 2026 and Gerdau Midlothian completes its expansion. The Arkansas cluster, Nucor-Yamato plus Big River, has no competing lime producer within 280 miles of Blytheville.

The timing differs between the two channels. Caddo-Bossier is a 2026-2028 construction-phase event, front-loaded and then done. The EAF steel demand is already running and grows as capacity utilization increases at the expanding mills. Anyone modeling USLM’s AI infrastructure exposure should treat them as different shapes of demand: one short and sharp, one long and compounding.

The margin picture: The forward gross margin isn't 48.9%. Waha normalization through 2026-2027 likely brings it to the 46% to 47.5% range. Applied to flat revenue, that implies gross profit of roughly $171 million to $177 million versus $182 million in FY2025. If EAF volume growth adds 2% to 3% to revenue over 2026-2027 as Big River 2 and Gerdau Midlothian ramp, that partially compensates. These are estimates, not a model.

USLM isn’t an AI infrastructure stock in the way that framing is used online. It’s a debt-free regional lime monopoly with irreplaceable geology and a regulatory moat that prevents new competition, with a gross margin that is partly structural and partly borrowed from a Permian gas anomaly that is fading.

The AI build-out reaches it through two channels with different profiles: a geographically captive soil stabilization opportunity in Caddo-Bossier that generates modest construction-phase tonnage starting now, and a recurring EAF steel demand worth an estimated $30 million to $43 million annually from mills where USLM holds a structural freight advantage, growing as Big River 2 reaches full output and Gerdau expands. The thesis I'd build, isn’t “USLM benefits from AI because data centers sit on clay.” It’s “USLM holds a freight monopoly on lime supply to an EAF steel cluster that is growing because AI infrastructure is the largest driver of structural steel demand in the current cycle, and that demand recurs annually for the life of the mills.”

The Texas fab subgrade story, the Meta Louisiana narrative, and the margin trajectory implied by treating 2025 as a normalized baseline, those don’t reach it, or reach it far less than claimed.

The demand travels through a steel furnace in Arkansas, not through the dirt under a server hall in Louisiana, and that's a distinction that changes most of the model inputs.

What would break this

There are three things that would make me change my mind. First, if Q2 and Q3 2026 show construction volumes recovering but data centers staying absent from USLM’s disclosure, the 2025 data-center demand was a project-phase pull-forward, not a durable channel. Second, if Lhoist’s Clifton expansion completes ahead of schedule and prices aggressively into the Shreveport market, the Caddo-Bossier freight monopoly erodes faster than expected. Third, if Big River 2 runs significantly below 75% utilization through 2026, the EAF revenue estimate compresses proportionally. None of these are base-case expectations. But they are the specific signals worth monitoring, and the right response to any of them is to revisit the thesis, not explain them away.

Two inputs in the valuation published last week need updating. The forward EBITDA base should be modeled at normalized gross margins of 46% to 47.5%, not at the 2025 peak, which moves gross profit roughly $5 million to $11 million lower before demand assumptions. On the demand side, the EAF steel cluster in Arkansas and north Texas represents a more defensible and larger recurring revenue base than the original thesis isolated, estimated at $30 million to $43 million annually from mills where no competitor can match USLM's delivered cost. The moat analysis, capital structure, and regulatory armor remain unchanged. What changes is the composition of the demand story: less dependent on unverifiable one-time construction events, more anchored to annual furnace lime contracts at mills where USLM wins on distance.

As always, if something here doesn't add up or you see it differently, the comments are there. This one took a while and I'm sure I got some things wrong.

Thanks for reading!

Disclosure: I do not currently hold a position in U.S. Lime & Minerals (USLM) as of the date of publication. This article is for informational purposes only and does not constitute investment advice.

One thing to keep in mind: the model reflects what I know when the thesis is published, not six months later. If the facts change, the numbers change too, and the update becomes the version that matters. When I publish one, I'll link it right here.

1. INVESTMENT CASE

USLM looks like a commodity business. The economics are materially better than the label suggests, and the market is still underweighting why.

In 2022, USLM’s gross profit margin was 29.8%. In 2025, it was 48.9%. Almost 19 percentage points in three years, in a company that sells a heavy material derived from a rock. That’s the number I keep coming back to. That kind of expansion in this type of asset doesn’t happen unless something structural changed.

USLM holds irreplaceable, permitted limestone deposits in Texas and the South-Central U.S., inside a geography where transporting a heavy, low-value material makes outside competition economically impossible. Add a permitting environment in Dallas-Fort Worth hostile enough to make building a new lime kiln a near-impossible task, and you see why this business has pricing power its sector label doesn’t suggest.

Four things drive the 2026-2029 story: data center demand in Texas showing up explicitly in 2025 volumes; a new vertical kiln expected to start up in summer 2026, still under construction as of Q1 2026; a structural energy cost advantage tied to the Waha natural gas hub in the Permian Basin that most people haven’t fully mapped onto the margin story (I’ll try to be honest about where the numbers get imprecise); and a CEO with no named successor whose contract runs through 2028.

Things I’m not comfortable ignoring: CEO Timothy Byrne is 69. One controlling family holds 61.58% through Inberdon Enterprises Ltd., creating governance and liquidity constraints worth understanding, and a succession question that goes beyond the CEO. Q1 2026 came in soft. And the valuation, even after a sharp correction, is 14.5x EV/EBITDA.

Current Market Context

The stock fell 17% on April 30, 2026, the day Q1 results came out, putting the price about 22% below the 52-week high reached in late 2025. I started writing this in the days after that drop, using $105 as the entry price. By the time this publishes, the stock has recovered to around $111. The model and the scenarios use $105 throughout. The final section explains where I stand on that gap and why the thesis doesn't change, but the entry math does.

2. Business Model

2.1 It starts with a rock

Limestone is a sedimentary rock made mostly of calcium carbonate (CaCO3), with large deposits across the central and southern United States. In its raw form, you crush it and sell it as a construction aggregate. The more valuable thing you can do with high-quality limestone is heat it, and that’s where USLM’s business begins.

Purity matters. USLM's deposits contain at least 96% calcium carbonate. Rock with too much silica or magnesium produces lime that won't meet industrial specifications.

The 96% purity floor isn't a quality standard. It's a functional requirement. Steel producers can't substitute lower-purity lime without contaminating the melt. Water utilities can't use high-magnesium lime without generating treatment byproducts that violate their own permits. TxDOT specifies lime by reactivity index, not just composition. The deposit either meets the spec or it doesn't. USLM's Johnson County deposit does. Most deposits in the region don't, which is why most competitors in this geography aren't competing for the same customers.

2.2 From rock to product

Once limestone is extracted, USLM can sell it as-is or process it into progressively more valuable products. Each step requires more energy and produces a more specific product:

The competitive protection is not uniform across this ladder. Lime slurry, USLM's most commercially important product in Texas, can't travel more than 50-75 miles. It settles. It needs agitation. It can't sit at the customer site waiting to be used. The protection is tightest exactly where the margins are highest and Texas construction demand is most concentrated.

Unlike quicklime or pulverized limestone, slurry has no spot market and no benchmark price a contractor can reference when negotiating. The TxDOT subcontractor in Johnson County doesn't call three suppliers to compare delivered costs. There's one supplier within the delivery window that can meet spec and timing. And because lime slurry typically represents around 2-4% of the total cost of a highway subgrade project, customers have limited economic incentive to push back hard on price even when they'd like to. The material is essential to the project. It just isn't a big enough line item to fight over.

2.3 Where USLM operates

USLM operates across six states: Texas, Arkansas, Oklahoma, Missouri, Colorado, and Louisiana. The primary production site is Texas Lime Company, near Cleburne in Johnson County, about 35 miles south of Fort Worth.

Other extraction sites: Arkansas Lime in Batesville; Love Hollow Quarry in Cushman, Arkansas (connected to Batesville by a company-owned rail line); St. Clair Mine in Marble City, Oklahoma (underground); Carthage Crushed Limestone in Missouri; and Mill Creek Dolomite in Oklahoma. Distribution terminals in Shreveport and Houston receive product by rail and distribute by truck to customers too far for direct delivery.

Johnson County generates the overwhelming majority of USLM’s EBITDA. The diversification across Arkansas, Oklahoma, and Missouri is real in terms of reserve tonnage, but marginal in terms of current earnings contribution.

2.4 How much rock do they have?

As of year-end 2025, total proven and probable limestone reserves stood at 223.15 million tons, all with minimum 96% CaCO3 purity. In 2025, the company mined 4.687 million tons. At that rate, the aggregate reserve base is 47 years of production.

The site-level picture is more striking. The Texas Lime Quarry (5,200 acres in Johnson County, limestone bed averaging 25 to 35 feet thick) has reserves for about 70 years. The Love Hollow Quarry in Arkansas has over 93 years. These are SEC-regulated disclosures prepared by independent geologists.

No patent expiration. No technology disruption. No competitor that can replicate the same deposit in the same county.

2.5 Energy: the biggest cost and the biggest variable

Making quicklime requires heating limestone above 1,800 degrees Fahrenheit continuously inside industrial kilns. USLM’s existing Texas kilns are preheater rotary kilns consuming approximately 4.8 gigajoules of thermal energy per metric ton of quicklime produced. Think of burning 160 kilograms of coal to make one ton of product.

Energy is the single largest variable cost. USLM uses a blend of natural gas, coal, and petroleum coke, adjusting the mix based on relative price. That flexibility is an operational advantage, and also one of the reasons the energy cost calculations later in this thesis are less precise than I’d like.

In 2024, USLM began building a new vertical kiln at Texas Lime. Vertical shaft kilns consume approximately 3.4 GJ/ton instead of 4.8, a 30% reduction. Total estimated cost: $65 million. Expected startup: summer 2026. This is both a capacity addition and a structural improvement to unit economics, assuming the design specs hold in production, which I address in section 5.4.

2.6 Who buys, and why they can't easily switch

USLM sold to approximately 675 customers in 2025, with no single customer above 10% of revenues.

Construction (largest segment). Texas clay soils expand when wet and contract when dry, cracking roads built on untreated ground. The Texas Department of Transportation mandates lime or lime slurry in the soil subgrade before paving. One mile of treated highway subgrade requires 60 to 100 tons of lime. Texas approved a record $39.92 billion for its transportation budget for 2026-2027.

"Construction" is not one thing. Private residential and private nonresidential behave completely differently. After the Fed started raising rates in 2022, residential fell sharply and has stayed depressed. Nonresidential went up and kept going through 2023, 2024, and into 2025, driven by data centers, semiconductor fabs funded by the CHIPS Act, and TxDOT highway projects, all institutionally financed and largely indifferent to mortgage rates.

The 10-Q cited weakness from "construction customers" in Q1 2026. The residential side has been depressed for two years. USLM's core product is lime slurry for highway soil stabilization. That demand tracks TxDOT's budget, not housing starts.

Steel producers. Quicklime removes impurities (silica, phosphorus, sulfur) from molten metal during steelmaking. In 2025, large data center construction in Texas drove additional steel demand and consumed lime directly in site preparation. Q1 2026 showed steel holding up even as construction and oil and gas pulled back.

Environmental and water treatment. Municipalities use hydrated lime for water softening, pH adjustment, and heavy metals removal. Industrial facilities use it for flue gas treatment. Demand ties to regulation, not economic cycles.

Oil and gas. Used for oilfield road treatment, drilling mud conditioning, and produced water processing at well sites. Volatile by nature.

Q1 2026 oil and gas softness was directly traceable to WTI in the $70-80 range, below the threshold that drives aggressive Permian drilling programs. That picture has changed materially since. A geopolitical shock from the Strait of Hormuz closure pushed WTI above $100 in early May before a partial correction toward $90 on signals of a potential US-Iran deal. If Hormuz normalizes, WTI likely settles back toward $75-85 and the weakness persists. If it doesn't, Permian activity accelerates and this segment recovers. What hasn't changed: a higher WTI environment is doubly positive for USLM, because more Permian drilling generates both more lime demand at well sites and more associated gas production, which maintains downward pressure on Waha Hub prices.

2.7 Delivery and seasonality

USLM runs its own trucking subsidiary. Rail handles longer hauls from the Arkansas, Oklahoma, and Missouri plants, feeding the Houston and Shreveport distribution terminals.

Revenue peaks in Q2 and Q3. The January 2026 winter storm interrupted shipments, contributing to the Q1 revenue decline. The plants sustained no damage.

2.8 The unit economics

In 2025: 4.687 million tons mined, $372.7 million in revenue, $182.4 million in gross profit. That’s approximately $39 of gross profit per ton extracted.

In Q1 2026, fuel, energy, and transportation costs were $22 million against $87.8 million of revenue, 25% of the top line just for energy and freight. When energy rises and volume falls simultaneously, the impact is amplified: Q1 2026 gross profit fell 9.5% against a revenue decline of only 3.7%.

3. Industry & Competitive Position

3.1 Why lime is regional

Lime is heavy and low value per ton. Transportation costs eat into margins quickly. A truck delivers lime profitably within 150-250 miles of a plant. Rail extends that to 400-500 miles. For slurry, the radius shrinks to 50-75 miles.

USLM’s Cleburne plant sits 35 miles from Fort Worth. No competitor can reach that market from outside at a lower delivered cost. Not for slurry. Not for any construction lime product. The three largest North American producers account for more than two-thirds of total production capacity, concentrated regionally by exactly these economics. In any given geography, one or two producers serve customers economically. That’s not a market failure, it’s a direct result of the physics of the product.

3.2 Why no one can replicate it

Even with unlimited capital, a new entrant would need to identify a comparable high-purity limestone deposit near DFW (essentially none available and unclaimed), acquire thousands of acres in a rapidly urbanizing Texas county, and survive years of permitting before producing a single ton. I haven't found evidence of high-purity deposits in Johnson County that are unclaimed and unpermitted. Maybe someone has. It wouldn't change the timeline for this thesis regardless.

3.3 The regulatory shield

The Dallas-Fort Worth area is classified as a severe nonattainment zone for the 2008 federal ozone standard. Building a new lime kiln near DFW requires a New Source Review permit from the TCEQ (Texas Commission on Environmental Quality), meaning proof the new facility won’t worsen local air quality, purchase of emissions offsets, and deployment of Best Available Control Technology. No new lime kiln has been permitted in the DFW area in decades.

USLM’s Cleburne kilns operate under grandfathered permits that predate the current regime.

But this cuts both ways. The same environment constrains USLM’s own future expansion at Cleburne. The new vertical kiln already required navigating TCEQ. A third or fourth kiln in the 2028-2030 window may not be permittable. If volume growth beyond the new kiln’s capacity is needed, it likely has to come from Arkansas, Oklahoma, or a greenfield outside the DFW zone. That’s a materially different capital story.

What the grandfathered permit does economically: in practice, there are very few economically viable alternatives within the delivery radius that TxDOT projects require. The 21.1% price increase in 2023 wasn’t absorbed reluctantly by customers with no alternative. It was absorbed by customers with no alternative.

3.4 Concentration risk

Texas Lime represents roughly 60-65% of USLM's limestone mined and, given its higher-value product mix, likely more than 70% of gross profit. If Texas Lime loses 60 days of production (kiln failure, environmental enforcement, severe weather), the EBITDA impact at $39/ton and approximately 8,000 tons/day is $18-20 million. Not thesis-ending given the balance sheet, but meaningful quarterly volatility with no quick substitute.

The January 2026 winter storm didn’t damage the plants but interrupted shipments long enough to contribute to a 3.7% revenue decline in Q1. The tail risk is larger than that data point implies.

The partial mitigation: customers within the delivery radius have limited substitute supply during an interruption, so a portion of deferred demand would likely return once production resumes, particularly for contracted highway projects that can’t source elsewhere mid-schedule. Some projects do get delayed, some contracts do get rerouted. But a 60-day shutdown is more likely to produce a one-quarter revenue hole than a permanent volume loss.

The longer-run concentration risk is different and less forgiving. If post-2028 volume growth has to come from Arkansas or Oklahoma because Cleburne can’t get more permits, the margin profile of that incremental capacity is materially worse. The Arkansas and Oklahoma operations don’t carry the same high-slurry, high-margin product mix as Cleburne. Growing in those sites means growing at lower EBITDA/ton. That’s not a crisis. It’s a different business than what the current returns imply.

3.5 The competitors

USLM's main competitors are private: Lhoist (Belgium), Graymont (Canada), Carmeuse (Belgium). None has meaningful presence in the core Texas market. When a DFW highway contractor needs lime slurry, they call whoever can deliver within their window, at the right specs, from within 75 miles. That's USLM. Demand is cyclical and can fall. Energy costs can spike. A single bad quarter can look alarming. None of that changes who the customer calls next time they need slurry in Johnson County.

4. Ownership & Management

4.1 Control structure

Inberdon Enterprises Ltd., a private holding company owned indirectly by George M. Doumet, holds 17,653,780 shares, 61.58% of USLM’s outstanding equity. George’s brother Antoine M. Doumet sits on the board. The remaining 38% is the free float. With that structure, no activist campaign is possible, no hostile bid gets traction, and no minority investor forces a conversation the family doesn’t want to have.

I want to be precise about what the observable evidence says and what it doesn’t, because most analyses of USLM run something like: “the family is long-term oriented and disciplined, which benefits minority investors.” That framing is lazy, and I’ve been guilty of it.

Here’s what the record actually shows. From 2023 through 2025, USLM paid $15.1 million in dividends and spent $7.5 million on share repurchases, $22.6 million returned to shareholders over three years. Over the same period, net income totaled $317.7 million and the cash balance grew from $133 million to $371 million. No special dividends extracted, no related-party transactions in the 10-Ks, no debt loaded for financial engineering. The family hasn’t done anything visibly destructive.

That’s different from saying they’re aligned with the minority shareholder. A family collecting $4.23 million annually in regular dividends while $371 million accumulates in a Wells Fargo account has no urgency to change anything. The cash on the balance sheet increases the paper value of their stake whether it gets distributed or not.

What I genuinely don’t know: whether the next generation intends to maintain the controlling stake. Whether there’s an estate plan keeping Inberdon’s 61.58% intact after George Doumet. Whether a CEO transition in 2028 prompts a reconsideration of capital structure. The Proxy doesn’t discuss it. The 10-K risk factors mention “concentrated ownership” but not succession planning at the shareholder level. Anyone who tells you they know the answer is guessing.

Two risks worth pricing honestly. The first is permanent capital inefficiency, the cash doesn’t get returned, the business compounds internally but the gap between intrinsic value and market price never closes. The second is succession uncertainty, both at the CEO level and at the ownership level. These risks are correlated. A CEO transition in 2028 is the most plausible moment when the capital allocation question gets revisited. That’s a timing argument, not a prediction.

4.2 The succession question

Timothy Byrne has been CEO since the late 1990s. Through 2020, when revenues fell to $160.7 million, he didn’t cut the dividend, didn’t draw on the credit facility, and kept the asset base intact. The $65 million kiln is his most significant growth capex decision in years, and it’s operationally sound.

But he’s 69. His contract ends in 2028. No successor is named in any public filing. The succession question doesn’t stop at the CEO: the public record contains nothing about the family’s intentions for 2028-2035. Whether the next generation maintains the stake. Whether $800 million-plus in accumulated cash on a $3 billion market cap becomes a catalyst for a sale or recapitalization. I don’t know. Nobody outside the family does.

4.3 Compensation and capital allocation

Byrne’s cash bonus is tied entirely to absolute EBITDA, no ROIC, no FCF component. The 2025 threshold was $120 million EBITDA for maximum payout; the 2026 maximum is $605,000. Growing absolute EBITDA is what gets rewarded. Deploying capital efficiently is not in the formula.

The $65 million vertical kiln passes any reasonable return test. But the same incentive structure would reward a $200 million acquisition that adds EBITDA even if it destroys ROIC. That’s the latent risk in the comp design, not the current capital allocation.

Now the cash. By Q1 2026, the balance was $383 million, $13.32 per share at 28.75 million diluted shares, on a company trading around $105. With operating cash flow running at $150-165 million annually and capex dropping sharply after the kiln completes, the balance likely reaches $480-500 million by year-end 2026 and crosses $600 million by year-end 2027.

The scenario where the cash never comes back

If the cash is never distributed, the question changes: what is the market implicitly paying for the operating business alone? The arithmetic, with different assumptions about how much of the cash balance a minority investor should credit:

At zero cash credit, the market is paying 16.5x the operating business, not obviously cheap for a regional industrial without the scale or float of Vulcan or Martin Marietta. At 50% credit, 15.5x. At 100% credit, 14.4x, which is where the base case sits.

The right way to underwrite this: does the operating business justify 14-15x on its own merits, without any assumption about cash distribution? If the pricing power in sections 3 and 5 is structural, then 70-year reserves under grandfathered DFW permits with zero debt and 35%+ operating ROE trades at 14-15x in a private market transaction regardless of what sits in the bank account. The cash is upside, not support. If you need the cash catalyst to justify the entry multiple, the thesis is weaker than it looks.

The three-year track record doesn’t support the zero-credit scenario. The dividend grew from $0.16/share in 2022 to $0.24/share in 2025, a 50% increase over three years. Buybacks have been small but consistent. The family isn’t maximizing personal cash extraction, but they’re not leaving the minority shareholder entirely behind.

What I can’t say is whether $500-plus million in projected 2026 cash triggers a change in posture. No mechanism forces it. No activist can get in. The only catalyst is internal. The answer to that single question shifts my base case exit by $8-12 per share. I’ve modeled it at 100% credit and flagged the risk. The reader should decide what probability to assign.

Share

5. Financial Analysis

5.1 Revenue

From 2020 to 2025, revenue grew at an 18.3% CAGR, reaching $372.7 million from $146.1 million. The 2024-to-2025 growth was 17.3%, driven by an 11.7% volume increase and a 5.6% price increase. Both volume and price growing simultaneously in an industrial company is unusual. It suggests demand wasn’t just repriced but genuinely expanding.

Revenue in 2025 was 2.3x the 2020 level. Net income was 4.8x. The gap is entirely explained by margin expansion. 2020 was a genuinely bad year, so the five-year CAGR flatters the starting point. The compounding is real.

One item to isolate: interest income on the cash balance contributed $13.15 million to 2025 net income. About 10% of net income comes from a Wells Fargo account, not from selling lime. The stock screens more expensive on operating earnings than the headline P/E implies.

5.2 Margins: the expansion and what could reverse it

The chart above tracks the Producer Price Index for lime manufacturing in the United States. Two completely different regimes are visible. From 2015 to 2021, the index moves from roughly 240 to 274, 14% over six years, below general inflation. Textbook commodity behavior.

Then 2022. By January 2023, the index is above 400. By early 2025, above 490. A 72% price increase in three years, in a product that barely moved in the prior six.

This is external confirmation that USLM’s own filings aren’t telling a unique story. When the 10-K says average selling prices rose 21.1% in 2023, every lime producer in the country repriced that year. USLM is one company’s slice of what the industry-wide data shows.

After the repricing, the index doesn’t fall back. It settles between 410 and 500. Producers raised prices and customers kept paying. In a real commodity business, a cheaper supplier comes in or customers switch. Neither happened. The prices stuck.

Gross margin expanded from 29.8% in 2022 to 48.9% in 2025. Four drivers:

1. Pricing power: +10.6% in 2022, +21.1% in 2023, +14.2% in 2024, without losing meaningful volume.

2. Waha gas prices: The Waha Hub in the Permian Basin, where USLM sources natural gas for its Texas kilns, went negative 107 days in 2025.

3. Operating leverage: Fixed costs stayed largely flat as revenue grew.

4. Product mix shift toward higher-value processed products: The MD&A doesn’t break this out precisely enough to verify independently.

The Waha estimate: what I can and can’t say

USLM’s Texas kilns buy gas at Waha Hub prices, not Henry Hub. Waha sits inside the Permian Basin, where associated gas from crude oil extraction has chronically overwhelmed outbound pipeline capacity. In 2025, Waha went negative on 107 separate days, hitting lows of -$8.79/MMBtu in October. Henry Hub averaged $2.50/MMBtu that year.

The reason this discount has proven durable: gas in the Permian is produced as a byproduct of oil drilling. Operators drill for crude. The gas comes out whether they want it or not. Because the decision to drill is driven by oil economics, not gas economics, Permian gas production doesn’t slow down when Waha prices go negative. The wells keep flowing because WTI at $70 is still profitable regardless of what happens to gas. That creates a supply of gas nearly inelastic to gas price, which is why adding 2.5 Bcf/day of outbound capacity with the Matterhorn Express in late 2024 still left Waha negative on 107 days in 2025.

A higher WTI environment reinforces this. More Permian drilling generates more associated gas volumes, not fewer. The higher the oil price, the more wells get drilled, the more gas comes out as a byproduct, and the more pressure accumulates on Waha Hub.

I can’t nail the fuel mix from the filings. USLM discloses it burns natural gas, coal, and petroleum coke, adjusting the blend based on relative prices, but doesn’t disclose the split by plant or fuel type. That gap makes precise margin attribution impossible. Here’s what I can do: build a sensitivity table around the two unknowns, gas share of Texas Lime’s energy mix, and where Waha-Henry Hub basis lands in 2027.

Texas Lime produced roughly 2.9 million tons in 2025. At 4.8 GJ/ton, that’s approximately 13.9 million MMBtu consumed annually at Cleburne.

Read that table as the annual gross profit that disappears if Waha normalizes. The mid-range estimate (65% gas, $2.50 discount) puts the 2025 advantage at roughly $23 million. At 65% gas with basis compressing to $0.75, that drops to $7 million. The gross margin impact of going from $23M to $7M on $372M revenue is about 4 percentage points, from 48.9% to roughly 44-45%. That’s the number the base case should be built around, not the theoretical $51 million maximum (100% gas exposure at a $2.50 Henry Hub discount all year, the ceiling, not the center).

On durability: the Blackcomb pipeline, expected late 2026, adds another 2.5 Bcf/day of Permian outbound capacity. The forward curve shows basis narrowing in 2027 but not collapsing to parity. I lean toward meaningful Waha suppression lasting through at least 2027. That’s a lean, not a conviction.

The chart shows the national benchmark price collapsing from above $8/MMBtu in summer 2022 to roughly $2.50 in 2025. USLM's Texas kilns bought at Waha, not at that price. My best estimate of the Waha advantage is $7-28 million annually in 2025, mid-case around $20-23 million.

The PPI for limestone mining sits at 532.3 in April 2026, up from 530.9 in January. It hasn’t fallen.

Q1 2026 was a volume problem, not a price problem. Construction customers pulled back (weather and seasonal), oil and gas reduced activity (WTI-driven), and roof shingle demand was soft. The underlying pricing held. Those are two different diagnoses with two different implications. Competitive position eroding would be a thesis-changer. A soft quarter driven by weather and a weaker oil price is a soft quarter.

Q1 2026 showed the first year-over-year gross margin compression in three years: 35.6% versus 38.0% in Q1 2025. Higher fuel and energy costs were explicitly cited alongside lower volumes. Energy cost rising while volumes fall simultaneously is the scenario where the Waha advantage matters most, if it’s compressing at the same time volumes are soft, gross profit takes a double hit. That’s the variable to watch in Q2 and Q3 2026.

5.3 Free cash flow & Balance sheet

Read more

A few months after I published the Eurofins thesis, I realised I’d left something important unexplained.

The original thesis made the core investment case. The margin gap, the mature scope, the CapEx cycle ending, the buyback. But some of readers came back with a question I didn’t have a clean answer to: what is eLIMS-NG, and how does it actually work?

The question matters more than it looks. Go to Trustpilot and search for Eurofins. You’ll find over ten thousand reviews. A consistent pattern in the critical ones: invoices sent to the wrong project, bills for services already paid, debt collection letters arriving months after a payment was confirmed. Go to any investor forum where Eurofins is discussed and you’ll find the same complaints, almost word for word. The science is fine. The after-sales service is a mess.

I’d written about eLIMS-NG as a financial variable in the thesis. Capitalised IT spend, DSO trajectory, headcount relative to organic growth. But I hadn’t gone inside it. I didn’t know, for example, that the system doesn’t run on cloud infrastructure. Or that the reason you can’t push a software update remotely is because each hub has physical servers that need to be physically configured. Or that the billing confusion clients experience on Trustpilot isn’t a customer service failure. It’s what happens when a client’s account exists simultaneously in two billing systems that don’t talk to each other, one from the acquired lab’s legacy software and one from the Eurofins central template. When that fragmentation ends, so does the confusion.

This article is what I found when I went inside it. If the thesis left you with the numbers but uncertain about the mechanics, this is the piece that fills that gap.

One more thing on timing. In the Q1 2026 trading update, the company described “the early benefits of digitalisation initiatives” as a factor behind improving profitability despite organic growth of only 2.6%. That phrase does real work if you understand what eLIMS-NG is. If you don’t, it’s just language in a press release.

The regulator is the real customer

The thesis covers the business model, segments, and competitive structure in depth, if you haven’t read it, start there. What matters here is a single mechanic that the thesis touches on but doesn’t explain in detail: why the LIMS, not the analytical method, determines the economics of this business.

Eurofins competes in Testing, Inspection and Certification. In each contract (whether it’s a pharmaceutical batch release, a salmonella test on infant formula, or a PFAS (per- and polyfluoroalkyl substances, so-called “forever chemicals” found in water and industrial waste streams) check on drinking water) the buyer isn’t really the customer. The buyer is the regulator. And the regulator demands a certificate from an accredited third party. The consequence: the contract is won with the list of accreditations and the list of validated test methods, not with the lowest price. Eurofins carries more than 200,000 methods. A competitor needs those methods already validated before it can even bid.

The customer never decides on price alone. A one-day delay in a batch-release certificate for a pharmaceutical product can shut down a manufacturing line. A transcription error on a salmonella certificate can trigger a recall that costs a hundred times the price of the test. Whoever delivers fast and without errors charges a premium. Whoever controls the software that determines delivery speed and error rate controls the margin. That software is the LIMS.

From sample to certificate

It’s Monday morning at one of Eurofins’ food labs in the US. A courier drops off a cooler with twelve milk powder samples from a Midwest client. The receiving technician scans the manifest barcode. Within seconds, the twelve samples exist in the central system as twelve individual entries, each with an identification number, a client account, a contracted test panel, a promised turnaround, a unit price per test, and a billing endpoint. Every time a sample crosses a scale, a testing equipment, a spectrophotometer, or an analyst’s hands, the system adds a line to its history. By the afternoon, the system has cross-checked the last result against the contractual limits and fired off a digitally signed PDF certificate to the client portal, together with the invoice.

Forty or more digital movements per sample take place. A human touches four, maybe five. The other thirty-five are handled by a Laboratory Information Management System, or LIMS. If Microsoft Office is the operating system of corporate administration, the LIMS is the operating system of a laboratory.

The LIMS determines, with very little human filtering, three variables: turnaround time, the percentage of samples that require reprocessing, and the speed of invoicing after the certificate is delivered. Whoever has a good LIMS collects sooner, charges more per test, and reprocesses less. Whoever has fifteen different LIMS doesn’t have a LIMS. They have a problem.

Why Eurofins built its own LIMS

Twenty years ago, Eurofins made a decision that still shows up in every quarterly filing today: it would not license a LIMS from an external software company. It would build one.

Quest Diagnostics runs on Epic Beaker, a commercial LIMS produced by Epic Systems. LabCorp took a similar route. Buying a third-party LIMS has one obvious advantage: no in-house software engineering team on the payroll. The cost shows up as a recurring licence fee, not as CapEx.

Two hidden costs don’t appear on day one. The first is customisation. A commercial LIMS serves the lowest common denominator of many laboratories. When the US Environmental Protection Agency adds new PFAS compounds to its required testing list (PFAS are a family of industrial chemicals in water and soil that regulators worldwide are monitoring with increasing urgency) adapting the workflow inside a third-party system means submitting a change request to the vendor and waiting. The queue can take months.

The second is intellectual property. A laboratory’s testing methods are worth something commercially. If the LIMS runs on a vendor’s servers, that vendor has technical visibility into those methods every time it deploys an update. In a sector where clients pay specifically for proprietary analytical capabilities, that’s a real risk.

Eurofins decided both costs were too high. The consequence: the company capitalises between €105 million and €115 million per year in IT systems, 1.5% to 1.7% of revenues. Quest doesn’t carry that line in its CapEx because it pays through licence fees that go through the income statement instead. Neither model is free. Eurofins just pays differently, and the way it pays makes the cost more visible to anyone reading the cash flow statement without understanding why it exists.

When the time comes to install the same system across a thousand laboratories, the company doesn't need permission from a vendor to move. It controls the code, the deployment schedule, and the update cycle. It also explains why the unification has taken two decades. There was no system to buy. It had to be written.

Thirty years of acquisitions, fifteen LIMS, one integration problem

During those decades Eurofins hasn’t stood still. Each adquisition lands inside the group with its own LIMS, its own internal test codes, its own client database, and frequently its own language. The clinical operations bought from Synlab in Spain arrived with several different systems depending on the lab.

The result is what the company acknowledges in its 2025 European Investor Day: in Food & Feed Testing across Europe alone, more than fifteen different LIMS coexist. One division, one geography.

Eurofins manages this diversity through thirty National Service Centres around the world, which centralise treasury, accounting and financial reporting. But an NSC can only absorb a lab after it’s been technically connected to the central system. That connection can take up to twelve months after closing. During that year, the acquired lab operates under the Eurofins umbrella technically, but financially and operationally it still lives on its own island.

I’m going to call this the shadow period. It’s the gap during which accounts aren’t reconciled with the group, invoices aren’t emitted from the same template, and technicians are still being trained on the new LIMS. When Ernst & Young audited Eurofins’ cash position in 2024, they found one unrecorded entry of €254,000 from a German subsidiary. The entry hadn’t been posted because the subsidiary hadn’t been integrated into the NSC yet. There was no malfeasance. There was a shadow period.

Multiply that by the hundreds of acquisitions closed in recent years. An invoice that arrives late because the acquired lab isn’t yet in the central template. A certificate with the wrong logo because the migration is still pending. A client receiving payment requests in different formats from two Eurofins labs in the same division. That’s the shadow period the retail investor experiences as poor after-sales service.

The public record of what this looks like from the client side is more specific than most investors realise. On Trustpilot, Eurofins carries over ten thousand reviews. A disproportionate number of the critical ones describe the same underlying situation: invoices sent to the wrong project, charges duplicated across billing periods, debt collection letters arriving for invoices that had already been paid and confirmed. These aren’t service failures in the conventional sense. They’re billing system failures. When a client’s account exists in two separate billing platforms, one from the acquired lab’s legacy system and one from the Eurofins central template, a payment registered in one doesn’t automatically appear in the other. The client pays. The debt collection process, running off the system that hasn’t been updated, continues regardless.

The company’s own annual report adds an important detail: each laboratory runs its own system for registering and handling client complaints. A client with a dispute involving two Eurofins subsidiaries acquired at different times has to argue their case to two separate complaints processes. Management knows the customer journey is uneven. The 2024 annual report mentions that Eurofins organises “undercover proficiency testing,” described internally as mystery shopping, specifically to gain visibility on “the entire customer journey, from the first contact with a Eurofins laboratory, through to result delivery.” Companies that need to mystery-shop their own labs are not companies that believe the experience is consistent.

When eLIMS-NG lands at a lab and that lab is fully absorbed into the NSC (National Service Centre), the client account consolidates into a single record. One billing template. One complaints channel. One view of payment history. The billing confusion doesn’t disappear because Eurofins improved its customer service. It disappears because the fragmentation that caused it is retired.

Inside eLIMS-NG

eLIMS-NG (the NG stands for next generation) is the central piece of a wider proprietary suite: a client portal (myEOL), an electronic chain-of-custody system (eCOC) for field samples, a quotation module (eQuote), a sample receipt notification module (eSRN), an electronic data delivery module (EDD), an internal outsourcing system between their labs called TALS. That's a lot of acronyms for what is essentially one idea: every interaction between the client, the sample, and the certificate passes through the same proprietary infrastructure.

TALS manages the assignment of test work between Eurofins laboratories when a local lab doesn’t have capacity or the specific accreditation for a test. The client doesn’t see it, the invoice comes from the original lab. Eurofins has confirmed this subcontracting happens “on an ad hoc basis for specific technical knowledge or to cover demand and capacity limitations.” Volume data isn’t published since inter-company revenues are eliminated in consolidation. TALS is an internal efficiency tool. Clients choose Eurofins for price, turnaround time, and accreditations. TALS is invisible to them.

The technical architecture explains why the rollout is slow and expensive. eLIMS-NG isn’t a cloud-native application. It’s a hybrid architecture with physical nodes. Data is processed and stored on proprietary servers in regional hubs, the Netherlands, Kansas, St. Charles in Missouri, Mississauga in Ontario. “Secure storage on our servers.” Not on AWS. Not on Azure.

You can’t upgrade a recently acquired lab by pushing a patch over the wire. You send a team, configure local servers, migrate the legacy database, reconfigure analytical instruments so they stream data directly into the new system, retire the legacy software, train the staff, and run both systems in parallel for several months. Per-site disruption window: six to twelve months.

Current rollout status: essentially complete in BioPharma Product Testing; in progress in Food and Feed Testing and Environmental Testing; early stages in Clinical Diagnostics. Target date for full completion: end of 2027.

The numbers from North America, where the rollout is most advanced, are concrete. The auto-reporting module generates 35% of monthly reports without human intervention and cuts average turnaround time by 34%. The auto-narrative module generates 78% of report narratives automatically. The login redesign saves between 15% of time on small logins and 64% on large ones. These savings are already landing in margins today.

The rollout in the numbers

DSO deteriorates when a newly acquired lab can’t yet invoice from the central system, and improves when the system absorbs the billing. At 56 days in H1 2025, it’s three days below H1 2024, the direction is right. Administrative headcount grew at more than double the pace of organic revenue in the heavy M&A years, then inverted in 2026 for the first time. SDIs, the reorganisation costs carved out below adjusted EBITDA, declined to €37 million in H1 2025, down from €43 million in H1 2024.

On Glassdoor, the employee complaint is recurring and operationally specific: “the LIMS systems are an absolute nightmare and they have 2 different systems they use which do not communicate with each other.” That's not really a complaint about management. That's someone at the bench describing what dual-running actually feels like from where they're standing.

Read more

When I published the TIC thesis in January, the main question I couldn't answer was: where is management actually trying to take this business?

Revenue of $3B+. Adjusted EBITDA margin of 18%+. Free cash flow conversion of 85%+. Management is calling this target 3/18/85.

The base case in the published thesis modeled 18% EBITDA margins by 2028. Management just publicly committed to the same number, one year later, with a detailed bridge of how they get there.

The logic behind the 2029 targets

When APi Group held its first investor day as a public company, they put out a three-number framework too. The CFO of TIC, Kristin Schultes, was in the room. She mentioned it on stage yesterday. Their numbers were 13/60/80. She said at the time it felt aspirational and maybe repetitive. It became the core operational metric and proved effective.

She's running the same playbook here. Every business decision, every acquisition, every pricing call gets evaluated against those three numbers now.

The market dropped the stock because the capital allocation priority order put debt repayment fourth. Organic first, M&A second, buybacks third, debt repayment fourth. Some investors expected the company to use the FCF generated over the next three years to aggressively pay down the $1.6B in debt. That’s not the plan. The plan is to grow the business while keeping leverage at or below 2.5x, letting the EBITDA growth do the deleveraging work.

2. BEN HERAUD: WHAT THE CEO SAID

Heraud opened the presentation by explaining how they want the market to see the company before talking about any financial numbers. They want to be seen as a lifecycle partner for physical assets rather than just an inspection or engineering firm. Their idea is to position the business as a long-term partner that works with clients during the full life of an asset.

That matters because analysts compare companies using peer groups, and those peer groups affect valuation multiples.

The 3 priorities weren’t new, but this time management explained them with much more detail than during the Q1 call.

Priority 1: Grow in markets with structural, long-term demand.

Priority 2: Improve margins while staying technically differentiated.

Priority 3: Expand the role TIC plays across the full life of each asset.

The AI section

Heraud spent more time on AI than I expected, and he went into more detail than the usual tech comments you hear in earnings calls. He said TIC hasn’t seen, and doesn’t expect to see, real pricing pressure from AI over the next few years. The reason is simple: most of the work still needs qualified engineers, formal review, and client approval. You can’t replace that part of the process with software in regulated inspections like bridges.

Alex Hochman gave a concrete example. A 1,000-page hyperscaler contract can now be processed with AI tools that pull out key risks, highlight force majeure clauses, and list notification requirements. What used to take days now takes minutes. That mainly frees up time for engineers to focus on client work that actually brings in revenue.

There’s also a rough estimate for the CE segment. About 10% of staff is in administrative roles that can be partly automated with AI tools. It’s not about reducing headcount directly. It’s more about freeing up capacity in a business where the limiting factor isn’t money or equipment, but certified technical staff.

Geographic gaps in the business

Heraud showed a map. CE has almost no presence in Canada where I&M is very strong. I&M has no presence in the Middle East or APAC. He described this as a “land and expand” opportunity. Deploy a capability where the client relationship already exists, then expand the service offering locally.

The data center business is the clearest example of this working in practice. It started in APAC because that’s where TIC already had strong relationships with hyperscalers. Once those clients needed support in the US, TIC was able to follow them and deliver there too. Now the US data center business is growing fast, and Heraud said it will likely reach the same size as APAC. Right now, APAC data centers are about $64M out of $80M over the last twelve months, so if the US catches up, total revenue in this segment would roughly double.

Canada represents the next logical step for this strategy. The plan is to use existing I&M relationships to cross-sell CE engineering. Heraud was careful not to put a number on it, but Kevin Woit in the M&A section named CE expansion in Canada as one of the top-priority geographic targets for acquisitions.

Cross-selling financial impact

Dan Coloma presented the cross-selling update. The headline: YTD 2026 cross-selling has already equaled the full-year 2025 total. The 45% of that cross-selling coming from Acuren-to-NV5 and NV5-to-Acuren cross-selling between Acuren and NV5 shows the integration is driving revenue, not just cost cuts.

The main driver is the project margin profile rather than volume. When TIC cross-sells, win rates double and average rates increase significantly. The municipal RFP example he gave: a project that I&M couldn’t have won alone, now bundled with CE and GEO, came in at 50% higher rate and doubled the total revenue from that client for the year.

That's not a one-quarter trend: Schultes confirmed in the financial section that essentially no revenue synergy is baked into the 3/18/85 framework. The 5% mid-single digit organic growth target is the floor, built on holding existing clients and recovering some lost sites. Everything Dan showed yesterday is upside to the published targets.

3. THE THREE SEGMENTS

Consulting Engineering: Alex Hochman

Hochman started with a question: does anyone know what CE revenue was in 2010? CE revenue grew from $20M in 2010 to $714M currently, with a management target to exceed $1B.

Most CE contracts use Quality-Based Selection: TIC wins on technical capability and reputation, then negotiates the fee. Price isn't in the initial decision. That's why CE margins have held around 47% even as the business grew. No one undercuts them on rate.

The data center detail

Gary and Keith, the Hong Kong-based leaders of the international data center business, presented for the first time publicly. The numbers were much more detailed than in the earlier reports.

700 dedicated data center engineers across 10 countries. Working with 8 of the top 10 hyperscalers globally. Clients give construction pipelines of 24-36 months as standard. One gave five years.

Keith gave an example from Osaka. A typical Japanese data center development would take 2-3 years just for design and another 2-3 for construction. TIC completed the design in under a year. That speed is the product.

AI workloads require 3-8 times the power density of traditional compute. Every time a hyperscaler refreshes its AI infrastructure, it needs to re-engineer the cooling, power distribution, and structural systems. That’s not a one-time project. It’s a recurring engineering engagement every time the underlying compute technology advances.

TIC isn’t exposed to the construction risk or fixed-price construction risk. The model is pure engineering hours plus expertise. More complexity per megawatt means more hours and more specialized expertise, which means higher revenue per project and better margins. This is the opposite of a commodity business.

Gary presented a service penetration chart showing how each TIC segment currently serves the data center lifecycle. CE is deeply embedded across most phases. I&M and GEO have significant grey areas, meaning services they could deliver but currently don't in this market. Gary's comment: in the next few years, those grey areas will fill with much darker shades.

What Hochman changed in my model

The original thesis used $70M as the 2025 data center baseline and $100M as the 2026 target. The trailing twelve months to March 2026 is already $80M. APAC accounts for roughly 80% of that. Management expects the US data center market to reach parity with APAC. If this occurs and APAC continues growing, the path to $200M in data center CE revenue within two to three years isn't a bull case assumption. It's the base case.

The CE white space in Canada is the other piece. Zero CE presence yesterday in a market where TIC has 45% of its I&M workforce and decades of client relationships. Hochman showed the map. The density of CE offices in the US coastal markets is strong. The interior, Texas, and all of Canada is effectively empty. That’s the acquisition and organic build target for the next three years.

One project Hochman highlighted that’s worth naming specifically. TIC is the lead engineering partner on the Fifth Avenue redesign in Midtown Manhattan, a $300M project covering 21 blocks between Bryant Park and Central Park. That’s ten times the size of the Lexington Avenue project they completed previously, which was itself the first infrastructure BIM model ever built for NYC public infrastructure.

What makes it relevant for the thesis isn’t the size. It’s the model. TIC used LiDAR scanning above and below ground on Lexington Avenue to build a complete digital twin of the corridor, mapping every utility, its depth, its relationship to adjacent buildings, all without digging. The city now knows exactly what’s under the street without putting a shovel in the ground. That’s the product.

Fifth Avenue follows the same approach at ten times the scale. And Hochman made the point that matters most: if this model works on one corridor, it works on every corridor. The city of New York has thousands of blocks of aging infrastructure with utilities installed in the 1800s still in active use. The addressable market for that kind of work isn’t one project. It’s a city.

One more CE project worth naming: TIC has already completed 3D scanning of 1,000+ retail locations for a single national chain. Six business units coordinated. The output is a digital twin of each store that enables the client to deploy AI-driven sensor and tracking technology. When the client renovates a location, the model needs updating. It's a recurring data relationship with a non-infrastructure client, which shows CE's addressable market extends well beyond bridges and power grids.

Inspection & Mitigation: Seamus Sullivan

Sullivan relocated from Canada to Houston six months ago. The Gulf Coast is a $200M region where TIC holds less than its fair share.

He spent the first part of his presentation on why I&M wins. Three reasons: scale and density, workforce and safety, and breadth of services. None of these are new to anyone following the thesis. But he quantified them in ways the filings don’t.

Their recordable injury rate is 0.11, which is about 21x better than the construction industry average. That’s not just a safety KPI. It directly affects sales. Large clients in oil & gas, aerospace, and utilities won’t risk bringing contractors into their sites if safety performance is weak. In practice, it acts as a filter, and it removes a lot of competitors before pricing even matters.

He also reframed the '100+ services' stat. The point isn't the count. It's that those services connect: inspect, diagnose, fix, re-inspect. All by the same team on site, under one contract.

Gulf Coast performance: what actually changed

This was the part I paid the most attention to. The Gulf Coast is still the weakest area, and Sullivan was pretty direct about that.

It’s a $200M region, and he said they’re below their fair share there. The main issues in 2025 were losing some sites and delays in LNG projects. The response has been changes in regional leadership, smaller operating teams with clearer responsibility, and a shift in how they sell. Instead of structuring sales around services, they now focus more on individual clients and how to deepen those relationships.

He said they’ve seen six months of steady margin improvement in the region. He didn’t give exact numbers, but if a business unit like this shows continuous improvement for half a year, it usually means the operational changes are actually working.

He also spent time on rope access, which didn’t get much attention in the room but matters for margins. In North America, rope access penetration is around 5-10%, compared to 15-20% in Europe. TIC helped develop that capability in Europe, and they’re now pushing it more in the US. In Q1 2026, rope access revenue grew 9%, and lab work grew 20%. These are higher-margin parts of I&M, so the mix shift matters for overall profitability.

The numbers management put on the slide: rope access saves clients 30-60% on cost versus scaffolding and 30-80% on time. One mobilization, multiple services, no shutdown required. That's the customer value proposition in three lines.

The Murray MacKay Bridge in Halifax: a 4,000-foot suspension bridge beyond its 50-year design life. TIC installed a modular platform covering 180,000 square feet, designed to last 10 years versus 5 for competitors. One contract, a decade of recurring access. That's the infrastructure rope access model.

On drones and AI

Sullivan addressed AI and automation directly, partly because it’s the question every industrial services company gets from investors right now. His answer was consistent with what Heraud said: The drone is the data capture tool. The value is in what you do with that data afterwards.

He also mentioned an internal tool called Ares, developed for inspection workflow management. With AI advances, the ability to develop and improve that tool has become exponential. That’s not a mature software product yet. But it’s a sign that I&M is building the data infrastructure that eventually supports the recurring asset intelligence layer Sullivan described as a long-term goal.

The most forward-looking slide in the I&M presentation was about what Sullivan called the Asset Intelligence Layer. TIC already sits at the critical data position in thousands of industrial facilities. It has digital radiographs, LiDAR scans, and corrosion history on specific assets across multiple turnaround cycles. The ARES system is designed to connect that data, standardize workflows, and eventually convert it into real-time predictive recommendations for clients. Sullivan said explicitly this isn't a mature product yet. But if it works, I&M stops being purely a labor-hours business and starts generating data-driven recurring relationships. It's the same strategic move GEO is making with GeoAI, applied to the inspection data TIC has been accumulating for 60 years.

The I&M model in the 3/18/85 framework

I&M guides 3-5% organic CAGR through 2029. That’s the most conservative of the three segments. Sullivan confirmed in the Q&A that the model assumes holding existing clients and recovering some lost sites. No cyclical rebound in refineries or petrochemicals is baked in. No outsize Gulf Coast recovery. Just disciplined commercial execution.

If you believe the margin expansion story in I&M, the math is straightforward. More rope access and lab work in the mix, better utilization of the 5,500 technicians in the field, and the overhead leverage that comes from growing revenue against a partially fixed cost base. That discipline, applied daily, is what gets I&M from 28% gross margins toward its required contribution to the 18% company-wide target.

Geospatial: Kurt Allen

The GEO segment looks mispriced compared to how it actually performs financially. It does about $298M in revenue with 52% gross margins, which puts it at the top end of TIC’s portfolio. Despite that, the market still values it like a standard industrial services business at around 7.9x EBITDA.

The way the business works is fairly straightforward. First, it collects geospatial data, then it processes it through the ENVI platform, and finally it turns that into recommendations that clients use for operations and planning.

In practice, it’s not just about flying aircraft and producing maps. A big part of the business is ongoing analysis work for clients who rely on TIC’s data for day-to-day decisions around assets and capital planning. The 500,000 ENVI users worldwide are mostly professionals who have already built parts of their workflow around the platform.

The GeoAI architecture

This was the section I paid the most attention to, especially after the pilot update in the Q1 call. Allen broke the GeoAI work into three levels.

GeoAI Practice: this is mainly services work where TIC helps clients design and set up AI in their geospatial workflows. It’s not tied to one vendor or system. They mix client tools, TIC tools, and third-party software depending on the project. The goal is solving operational problems for clients, not selling standalone software.

Geospatial Desktop Agents: two tools here. The IDL Agent is aimed at developers and helps speed up their workflow. The ENVI Agent sits closer to analysts and helps them plan and run geospatial work while still keeping control over the process.

Geospatial Enterprise Solutions: this includes GeoAgent and GeoIQ. GeoAgent acts like a system that understands client tasks and helps automate full workflows. GeoIQ uses large language models to help clients work with spatial data, like location patterns, distance, and clustering.

The key point is how this is delivered. GeoAgent isn’t sold as a standalone software subscription. It’s introduced through consulting work: first an AI readiness review, then implementation, then development. So every deal brings in upfront services revenue and also creates a longer-term software relationship. Clients don’t just buy a tool, they start relying on TIC’s setup for how they work with geospatial data.

Allen said TIC believes they’re first in this space with an AI system built on top of a full orchestration platform. That’s a strong statement, but it does line up with the scale of the ENVI platform, which has been around for decades and has about 500,000 users. Most competitors in this space don’t have that kind of installed base to build on. Platform itself has 30+ years of deep machine learning and AI experience behind it, built from the ground up rather than adapted from off-the-shelf tools. It's the reason competitors starting to build AI capabilities now are starting three decades behind.

Critical minerals project

One project that came up in the transcript deserves specific attention. TIC is currently conducting a deep-sea hydrographic survey in the Pacific Exclusive Economic Zone as a federal pilot program for critical mineral mapping. The scope: tens of thousands of square nautical miles. The workflow: vessel-based survey, AUV imagery (autonomous underwater vehicle imagery), seafloor data collection, and sample analysis integrated into a single intelligence output.

The US Executive Order on critical minerals gives this type of work longer-term federal funding visibility. If this pilot turns into an ongoing program, it becomes a new part of GEO that wasn’t really part of the original investment case. It also connects directly to federal budgets that extend years into the future.

This helps explain why GEO sits at high gross margins around 52%. It’s not just basic data collection. The work is technically difficult, it depends on proprietary analysis systems, and the clients are often government entities. That combination supports pricing that’s very different from standard survey work.

Utility vegetation management

Allen also described the long-standing Western US utility program for wildfire mitigation and vegetation management. LiDAR and high-resolution imagery identify encroachment, fall-in, overhang, and tree-health risk at the individual tree level. The mid-cycle collection workflow has roughly a two-week reporting turnaround. The output is a 3-5 year risk roadmap that helps utilities justify and sequence rate-base eligible reliability investments.

This is recurring revenue. The same utility, the same lines, flown on a programmatic cycle, generating data that feeds directly into the utility’s capital planning process. When Heraud mentioned on the Q1 call that TIC flies 150,000 miles of utility lines annually, this is what he was describing. The market values this as project revenue because it shows up in the GEO segment without a subscription label. It works like a multi-year recurring contract, even if it's not labeled as one.

What GEO still hasn't disclosed

The GEO Agent subscription conversion metrics that the thesis identified as the key missing data point for May 19 weren’t disclosed in specific quantitative terms. Allen described the architecture, the deployment model, and the commercial logic. He didn’t give a number of active pilots, an average contract size, or a percentage of GEO revenue currently under subscription.

He did explain the structure and how it’s being deployed, but not the scale yet. That’s not necessarily a problem. Early deployments are still tied to consulting work and implementation services, so the subscription part won’t show up clearly until later stages of adoption.

The direction is clear. The scale and timing of the subscription transition are still open questions.

Read more