is one of just 4 stocks in our hyper-concentrated $18M Asymmetrical Bets fund, which holds at any given time the handful of stocks we believe have the most asymmetrical upside.

After selling off hard today with macro pressure, dropped another 10% after it released its earnings.

While the timing is unfortunate, the deep dive we were preparing for you guys couldn’t come at a better time. With this stock down over 15% in one day, it’s the perfect time to educate yourself on one of the AI buildout’s biggest winners.

Connectivity has become nearly as important as memory over the past months, and every company that’s in the supply chain runs into the the same exact problem:

They can’t ship enough.

We’re not gonna pretend that we’re reinventing a wheel with this company, coming up with some undiscovered stock no one has written about. However, sometimes the most obvious answer is the best one.

We think of CRDO as the ultimate stock for data center connectivity, and today you’ll learn why.

In This Article:

1. Credo Technology ($CRDO)

2. AECs Market

3. Optics Expansion

4. CPO/NPO Angle

5. Customers

6. Q1 FY27 ER and Guidance

7. Q1 Earnings Call

8. Valuation

9. Risks

10. Thoughts

Before we begin, this Substack is for informational and entertainment purposes only and does not constitute financial advice. Past performance is not indicative of future results. Never invest more than you can afford to lose. Our writers may hold positions in the securities discussed and may buy or sell them at any time without notice.

[1] Credo Technology ($CRDO)

Credo Technology is a fabless semiconductor company founded in 2008, based in San Jose. The company started as a SerDes IP licensing business but eventually evolved into a full connectivity product and solution company. Their core technology is already mentioned SerDes (Serializer/Deserializer), that takes parallel data from a chip and converts it into a single high-speed serial stream for transmission, and then reconstructs it at the other end.

When we said that is basically an ETF for connectivity we didn’t lie, because this company’s product lineup spans copper and optical connectivity, selling into every networking layer:

• Scale-Up (XPU to XPU)

• Scale-Out (Rack to Rack)

• Scale-Across (Data Center to Data Center).

At first, copper:

• Active Electrical Cables (AECs)

  • Copper cables with Credo’s silicon embedded in the connector heads, connecting GPUs to switches at distances up to 7 meters. They use 50% less power than optical alternatives and 75% less volume than traditional passive copper cables. AECs are the core revenue driver today.

• Retimers

  • They extend signal reach for Ethernet and PCIe (the standard connection between GPUs and the motherboard) connections inside servers and across racks.

• OmniConnect Weaver

  • A memory gearbox for inference GPUs that delivers 12x more memory bandwidth than standard configurations. Bill Brennan (Credo’s CEO) said that the Weaver contributes around 2/3k in revenue per one GPU and its production ramp starts FY2028 (really important, we will get back to it).

Then we get to the optical side of things.

• ZeroFlap Optical Transceivers (Own DSP + PIC)

  • The main product here, it carries data over fiber up to 500 meters. This product was co-developed with Oracle to solve link flaps, which are signal drops that slow AI cluster deployment.

• Optical DSPs (brain inside the transceiver)

  • Robin at 100G/lane, and Cardinal at 200G/lane, sold separately to other transceiver makers.

• Silicon Photonics PIC (the optical engine)

  • Came through the DustPhotonics acquisition, which brought in L3C technology that cuts the laser count (per module) from 8 to 2.

All of these products run on Credo’s PILOT (Predictive Integrity, Link Optimisation, and Telemetry), a software platform that monitors link health and flags signal degradation before it causes a failure in the system.

[2] AECs Market

Since Credo was the company that invented the AEC category it’s obvious that they own most of the market share, and specifically 88% of it. The remaining market share belongs to Broadcom ($AVGO), Marvell ($MRVL) and Astera Labs. Furthermore, 5 out of 6 major hyperscalers use Credo AECs in production, with xAI being fully deployed on them.

The reason AECs took share from DACs and optical modules comes down to power, and reliability, which we talked about in the product line section, but there are also other important factors:

• DACs are thick and heavy, and naturally, it restricts airflow and limits how many cables you can pack inside a liquid-cooled rack.

• Optical modules solve the density problem, but they consume a lot more power at short distances.

Simple reasons, but that’s really all you need to change old DACs and newer optical modules to AECs. Moreover, Credo claims its product delivers “up to 1,000x greater reliability” than commodity optical modules.

There’s also a Rosenblatt preview, in which they noted that the demand signal they’re ($CRDO) picking up is mainly focused in scale-out connectivity for up to 7m connections, where copper is just a lot more affordable than optical solutions.

Last but not least, copper’s power efficiency, which is why copper is ramping faster than optics right now, and we believe the dynamic holds through at least late 2027.

In August, SA reported that NVIDIA is working to extend copper’s range even further by qualifying PTFE-based laminate materials for the NVSwitch board on Rubin Ultra. This means that they’re trying to push signal performance past what current materials can deliver, and this extends the adoption of AECs.

[3] Optics Expansion

Read more

Hey everyone,

One goal of ours for the rest of the year is to bring you guys more visibility and transparency into our funds/performance to help you get maximal value out of our subscription and stay convicted in core trades as we face increased volatility.

As many of you know, one of the most interesting parts about our publication is that we have over $55M AUM on Autopilot, an app where thousands of retail investors follow our thematic portfolios trade by trade, without needing to self-manage strategies.

For paid subs we share all these updates and daily performance in the VIP discord, as those who do not live in the United States or don’t have a compatible broker can not follow our funds on the Autopilot app.

Michael also publishes his high-risk, high-trade-volume trading portfolio on the VIP discord server for paid subscribers (like his bullish options position ), alongside a group chat where all of our members and analysts chat. If you ever had trouble connecting, we have fixed those issues so you can now connect to it using an email passcode if your discord account wasn’t syncing.

Before we begin, this Substack is for informational and entertainment purposes only and does not constitute financial advice. Past performance is not indicative of future results. Never invest more than you can afford to lose. Our writers may hold positions in the securities discussed and may buy or sell them at any time without notice.

Current Overall Market Thoughts

The stock and bond markets are pricing two different realities right now. The S&P 500 is trading near 7,711, up about 21% over the past year, as AI spending enthusiasm keeps pushing prices higher. At the same time, the 10-year Treasury yield sits above 4.70% and the 30-year is over 5.20%. When you can earn a guaranteed 4.7% risk-free from government bonds, you need a much bigger potential payout to justify taking risk in stocks. With the S&P 500 priced at roughly 23 times earnings, stocks are historically expensive and leave almost zero margin of safety if corporate profits don’t grow at record speeds.

Bond yields aren’t coming down anytime soon because of basic supply and demand. The federal government is running a nearly $2 trillion annual deficit, flooding the market with new debt that investors demand higher yields to absorb. At the same time, mega-cap tech companies have issued around $200 billion in corporate bonds this year alone, competing directly with the government for investor dollars. While Treasury Secretary Scott Bessent stepped in with a $4 billion bond buyback program to calm the market, it’s essentially a temporary band-aid swapping long-term debt for short-term bills without fixing the root problem. To make matters worse, Fed Chair Kevin Warsh made it clear at Jackson Hole that current financial conditions aren’t even restrictive and that inflation is still running too hot, effectively killing off hopes for immediate rate cuts.

Despite stubborn inflation, ongoing geopolitical tensions, and interest rates staying higher for longer, stock market volatility is asleep, with the VIX sitting near 14. This means many traders are pricing in an ideal world where nothing goes wrong, leaving them vulnerable to sudden pullbacks.

Three critical catalysts are set to dictate whether this market disconnect can hold: the September 4 Non-Farm Payrolls report (testing whether recent labor market cracks are widening), the September 11 CPI print (revealing whether energy disruptions and sticky core components are re-accelerating inflation), and the September 16 FOMC decision (where the Fed will formally address whether its hawkish tone translates into actual policy tightening).

In a market pricing in perfection with zero room for error, staying disciplined and thinking long term here is the only real edge. Here is where our strategies sit today and how we are positioned heading into this month's macro catalysts.

Autopilot Portfolios

AUM is live Autopilot total COA as of Sunday 8/30. Performance chips as of Sunday 8/30 (last session Friday 8/28). Holdings as of Friday close. All trade updates since August 1.

*A note on the math: the +18.4% blended return weights each fund's one-month performance by its AUM, so the three biggest sleeves (Photonics $19.7M, Asymmetric Bets $18.2M, Flagship $15.1M) drive most of that number. The +11.9% equal weight average counts all six portfolios the same regardless of size.

Read more

For the first time in the history of US, the national debt crossed $40T, and the 30-year yields hit its highest level since 2007. On the same morning Treasury Secretary, Scott Bessent, announced he’s doubling the government’s bond buyback program.

Stock market loved it and reacted almost instantly with something we like to call “the dead cat bounce”, because when we put the $4B buyback program next to the $40T in total debt, the math is pretty disappointing.

Bessent is giving us time, and the real question is, how much left of it do we got? This week Kevin Warsh will give his first big speech at Jackson Hole, and what’s going to be said there, will probably have immediate consequences on the market. Furthermore, we have upcoming midterms, and Trump needs both low gas prices as well as a strong stock market (currently 0/2), so understanding the macro is really crucial for anyone interested in the space.

There’s a lot to go through, so we’re going to take you through everything, one piece at a time.

Before we begin, this Substack is for informational and entertainment purposes only and does not constitute financial advice. Past performance is not indicative of future results. Never invest more than you can afford to lose. Our writers may hold positions in the securities discussed and may buy or sell them at any time without notice.

Read more

We wrote about tungsten on April 1st, when we first published our deep dive on the critical minerals trade. Tungsten prices had 10x’d, and we laid out the thesis that the supply demand imbalance was structural and wouldn’t ease for many years.

Since then, the critical minerals trade hasn’t worked the way we expected, and the sector has been weak. is now trading around $16, roughly 34% off its ATHs, despite revenue growing 498% YoY, and the company sitting on C$1.2B in cash.

The fundamentals have gotten stronger, and on August 7th Trump committed another $3B to critical minerals projects, and the executive order banning Chinese tungsten waivers for defense killed remaining hope that the supply chain was going to revert to normal.

To check back in on the trade, we sat down with Lewis Black, the CEO of Almonty (), for 30 minutes to talk about the company and the industry directly.

This conversation was recorded before Q2 earnings were released on August 11, and interview responses are quoted directly from our call unless otherwise noted. You can also watch this interview on Michael’s channel.

At the end we’ll share where we think the stock can trade.

Before we begin, this Substack is for informational and entertainment purposes only and does not constitute financial advice. Past performance is not indicative of future results. Never invest more than you can afford to lose. Our writers may hold positions in the securities discussed and may buy or sell them at any time without notice.

Why Should Investors Care About Critical Minerals?

As we mentioned in the LPTH interview, many defense and critical mineral names have underperformed other sectors.

Michael asked Mr Black why a retail investor should still be interested in the critical minerals and overall mining trade.

Almonty’s CEO answered:

“They’re absolutely right [about the performance of the industry]. The issue is the dreaded junior miner. The person who has the greatest thing since sliced bread, if you give him money, he’s going to change the world. And that’s really the end of it. Everyone’s making promises. They’re like meeting a guy in San Francisco back at the gold rush. Everyone’s got a gold mine, buy me a drink and I’ll give you a certificate for the greatest thing ever.”

He then followed with what investors should look for (when investing in the space)?

“Has the management done this before? Not just been in some metal, in this metal they’re now working on. Have they put a lot of their net worth into this? Are they willing to follow you dollar for dollar into it? Does the project have longevity? Does it have grade? And most importantly, is it permitted? Because if it’s not, it’s a multi-year journey.”

And then he mentioned how there’s no AI without critical minerals:

“No AI without critical minerals. It’s like being able to make the greatest birthday cake in the world. But if you don’t have any wheat, there’s no way to make the cake. Tungsten is in everything. You can’t make semiconductors without the tungsten gas. Just a tiny little bit, but without it you can’t do any of it.”

This is probably the most important quote in this entire interview, because the AI trade has always been about chips and memory (with power now catching up). However, every semiconductor on the planet requires tungsten gas during fabrication.

The Supply-Demand Imbalance

Michael’s question:

“Companies like Micron, we’re just living in this mania around supply demand imbalances of important things in these super cycles. How long do you see this lasting? How permanent is this supply demand imbalance and this pricing power that you’re coming into as a company?”

Mr Black’s answer, which shocked us:

“Five, six years before we find balance. Because until China really develops out their recycling there will be this continuing imbalance in the sector, and there’s nothing new coming online in tungsten except for our mine that just opened in Korea.”

The next thing that surprised us was the fact that China itself doesn’t have enough tungsten. Lewis Black confirmed they’re a net importer, chasing the exact same material his Western customers need:

“They have the same problem as us. They don’t have enough generated in China to feed their own domestic consumption, so they have to also pursue the same material we do.”

Michael asked where prices settle long term:

“Many years ago, I think eight, nine years ago, we were hired by the Chinese government to help all the Chinese mines comply with new environmental codes that came in. We got to look at the Chinese mines and they all worked with subsidies and they elicited about a 12 to 18% margin because of this subsidy. If I now factor in recycling, which is a much higher energy footprint than primary mines, we’re going to have to see a price of between a thousand and fourteen hundred. Right now it’s 3,100. So all I can really say with some authority is that the price won’t ever go below a thousand again.”

We would like you to remember that Almonty was built to operate at $250 and $350 of tungsten prices, which means with a floor of $1,000, it’s still 3x what the business was designed around.

Almonty’s Revenue Ramp

Michael asked Lewis what revenues look like over the next few years as Sangdong comes online.

Mr Black answered:

“We should do with the current price around 100 million from Portugal this year, probably a little bit more. As Korea comes online, it’s going to add another, you know, in phase one another roughly ballpark 400 million, 400, 450 million. So I’d say that during 27 we should be pushing, you know, if the prices remain at these sort of levels, you’re going to be closer to 700 million in revenue.”

Then there’s also molybdenum, where Almonty has a fully permitted moly deposit in Korea. The second largest moly consumer in the world, Sear, is in-country and as of right now, imports 100% of its feed. Lewis already has a contract with them:

“They’ve given me a $19 a pound floor price, no cap on the upside. It will be a revenue driver. At these prices it’ll add another 300 odd million dollars of revenue to me a year.”

But when Michael started focusing on revenue, Lewis cut him off, saying that his main point was margins:

“Forget about revenue, it’s margin. That’s what’s insane. When we modeled Sang Dong and we worked to the price of $350 we had a 30, 40% net income margin on that. We’re now at $3,100. So almost 10x. I don’t even model it at this price because it looks like I’m just making the numbers up.”

Contract Structure and Pricing Power

Michael’s question:

“Is it too good to be true? Like these deals, there’s a price floor and you get paid spot. So every month you get basically just increased, the price keeps going up, you just get increased to spot each month for each delivery.”

Lewis on why Almonty gets these terms:

“We’ve done this a long time. We’re the only ones who get this. But we get it because we’ve always delivered, even in the lean times. And my customers, we all helped each other when things were really bad. Sometimes they would give me a fixed price contract for a year above market so that we could survive. But we kept supplying and supplying on time. We need each other. We can’t stand each other. We both want to kill each other almost on a daily basis, but we’ve worked together for so long. They rely on me and I rely on them.”

Furthermore, Almonty extended its off take agreement with Global Tungsten and Poweders (a member of Austria’s Plansee Group), by adding six years to the term and improving pricing by roughly 6.3% on all contracted volumes. The offtaker voluntarily gave Almonty more money on an existing contract he didn’t have to renegotiate, Lewis said:

“The offtaker gave me more money for an existing contract that he already had. He didn’t have to. He already had a price. He gave me actually more money because that’s the state of the market. People don’t believe there’s a huge amount of material coming online.”

Iran, and the Tungsten Allocation Problem

Michael asked Lewis about the military applications around tungsten and how the Iran conflict changed the dynamic.

Lewis’s answer:

“You have to bear in mind that right now the defense prime in the US does not have the additional capacity, output capacity, that hasn’t been completed yet. So we’re still supplying the normal amount of tungsten that has always been consumed by the defense base. When that additional capacity comes online sometime during next year, next 12, 14 months, you will see it have a meaningful impact on tungsten consumption. The question is where does the tungsten come from?”

His answer was particularly interesting:

“The only thing you can really do is sort of borrow from Peter to pay Paul, and that means other sectors are going to have to have less availability of tungsten. My bet having been in this for a long time now is that less essential areas like the car space will see bottlenecks of tungsten components because it’s being diverted towards defense. And rather than risk recessionary implications of the car space slowing down, especially the Europeans, they will allow the Chinese to provide those components. And then China begins the drive to take market share in the downstream.”

This means that China spent 30 years feeding cheap raw materials to competitors to the West, and now they’re pivoting to finished components. Defense munitions demand creates the opening, and European auto manufacturers accept Chinese tungsten components to avoid production slowdowns, so China gains downstream market share.

Michael also asked about the recent dip in tungsten prices on certain exchanges, Lewis had a specific explanation:

“You noticed recently the Trump administration issued an executive order banning the waivers that the defense industrial base in the US were looking for, because after the end of this year you can’t use Chinese, North Korean or Russian tungsten, and they were looking for waivers. And so traders were looking to jam prices or have the illusion of lower prices in China to jam margin to go buy material and then feed it into the US. Once that executive order was actually issued, you notice the price in that particular publication started to climb back up. By complete coincidence, I’m sure.”

Dilution, and the Kazakhstan Competitor

Michael’s question on dilution:

“You have over a billion dollars of cash available to you. How and through what mechanisms is that available to you? And with equity issuance, share count was up 32% in the last year, what is the future of liquidity for Almonty and how much more do you think you’ll have to lean on the equity market?”

CEO’s answer:

“We don’t envisage going back to market because there’s no need to. What we are looking at right now is what Warren Buffett always used to say, wait for there to be blood in the streets. Better to have cash in the bank and just wait. It’s getting more and more difficult to raise funding for entire projects in critical metals.”

We always talked about Fidelity, which over the weekend updated their size in Almonty’s to 31.8M shares, calculating to just over 11% of the company. Mr Black talked about who let in during the NASDAQ IPO last year:

“I was very careful when we did our IPO on the NASDAQ last year not to get the cats and dogs involved. Just go for the most difficult people to impress. I always believed that if you lift the kimono and someone kicks all the tires and they are a reputable blue chip very fastidious institution, others will follow.”

And fast forward, Michael asked about the $1.6B Cove Cause tungsten project in Kazakhstan that the Trump brothers may be involved in, Lewis replied:

“I’ve seen over the years an enormous amount of capital destruction in tungsten. Every mine that’s open that we haven’t owned has failed because processing is a very difficult metal to process. Kazakhstan, one of the requirements and one of the problems China has found is that you can only use Kazakhstani workers. But there’s no tungsten sector in Kazakhstan. So you’re having to recruit people that have no knowledge.”

He then quoted someone else’s framework:

“Someone said to me the other day that when you’re assessing the viability from a critical metal standpoint of any project, in the event of World War Three, would it still supply the United States? I think that was actually a very simplistic but eloquent way of looking at the geopolitical viability of a project.”

What Goes Wrong by 2030?

This question appeared for the second time in a row, following the question which was in our LPTH article. Michael asked:

“It’s 2030 and Almonty failed. What happened?”

At first, Lewis ran through everything that can’t kill the company. He explicitly pointed to hard floors on all contracts that protect against a price collapse and he also mentioned that China isn’t going to start supplying raw materials again because they’re moving downstream.

Then he actually gave a us a reason why Almonty could collapse:

“I suppose the only thing that can really happen is if my customers, who I hold an equal disdain because I blame them in part for this mess because they were seduced by cheap materials, if they go out of business because I don’t know, they just couldn’t work out how to put their own pants on, that would be the only reason I didn’t exist. And I probably wouldn’t exist as Almonty. I’d probably have a Chinese name above the door. China would have bought me. So my shareholders will be okay.”

He also added that the downstream customers who do have vertical integration like Plansee Group or Sandvik, are already taking market share from the ones who don’t. The ones without secured supply are still in what Levis called a “deer in the headlights” phase. They spent last year in shell shock, and only now they’re starting to realise the situation is permanent.

Our Price Target

Read more

Since we started this Substack we’ve been spotting bottlenecks in the AI supply chain before market priced them in. Optical networking with or even testing equipment with and .

PCBs are the next one.

A printed circuit board is one of the oldest component in the history of electronics. These are copper layers that conduct electricity between chips, and are separated by sheets of insulating fibreglass, drilled with thousands of microscopic holes. For decades this was a low margin commodity, produced mainly in Asia.

There’s one publicly traded pure-play PCB manufacturer based in the US, and it just printed its first $1B revenue quarter. The company is TTM Technologies ($TTMI).

Before we begin, this Substack is for informational and entertainment purposes only and does not constitute financial advice. Past performance is not indicative of future results. Never invest more than you can afford to lose. Our writers may hold positions in the securities discussed and may buy or sell them at any time without notice.

What’s Inside:

1. Introduction

2. What is a PCB

3. The PCB TAM

4. TTM Technologies ($TTMI)

   1. What’s Their Business

   2. PCB Supply Chain

   3. Customers

   4. ASP expansion

5. Q2 Earnings

6. M+N and NVIDIA

7. Valuation

8. Risks

9. Our Take

[2] What is a PCB

You should think of a PCB like a sandwich. There are thin sheets of copper foil which conduct electricity from one chip to another, and between those copper sheets sits an insulating layer (usually fibreglass cloth soaked in resin and baked really hard).

The insulating layer is called the laminate and it holds everything while also stopping the copper from short-circuiting (frying itself).

So, now imagine that a cheap board inside a microwave has one or two copper layers, at max. The board underneath an NVIDIA GPU has at least 30 of them. Each of these boards gets drilled with thousand of microscopic holes called “vias” that connect copper on each layer.

If one of these drilled hole lands a few microns off, you can throw away the whole board. And yield (which is the percentage of boards that come out without being ruined), is the entire game in high-end PCB manufacturing.

Furthrmore, per Morgan Stanley, the PCB content cost is surging by 233% (from GB300 to VR200). Naturally, the boards need far more copper layers stacked on top of each other, using special laminate grades (M7,M8, and M9) that keep signal quality at extreme speeds, because standard laminate would destroy the data. Edwin Roks (CEO of TTMI) talked about how does this affect pricing:

“Because of that complexity, our ASPs are going up, a factor of 4, maybe a factor of 8.”

[3] The PCB Market

The Global Electronics Association’s May 2026 data shows a book-to-bill ratio of 1.60, meaning for every $1.00 of boards shipped there was $1.60 of new orders came in (bookings were up 102.8% YoY).

Moreover, GF Securities estimates the AI server PCB TAM at roughly $13B in 2026, with expansion to over $29B in 2027, and PCB cost per GPU will reportedly rise from $400 to around $750 for Vera Rubin platform. And it’s not Nvidia alone, Google’s new TPU is estimated to have $800 to $1,000 in PCB costs per accelerator.

However, without laminate, there’s no board, and CCL is the raw laminate sheet that every PCB is built from. Per BofA, combined June sales for Taiwan’s biggest CCL manufacturers just doubled YoY. With that kind of demand and rising prices of E-glass cloth (key raw material inside the laminate) can lead to higher prices in PCB boards, as the cost for its making is going to be substantially higher.

As we mentioned in the title, this whole industry is a monopoly taken by Asia. Roughly 90% of global PCB production sits in the east, and China alone accounts for about 59%. NA holds around 4%, down from 30% (at some point in the history).

Luckily, TTMI is the only publicly traded, pure-play PCB company in the US.

The rest of our TTMI report is available to paid subscribers only.

Join hundreds of asymmetry seekers getting access to:

1. Multiple deep dive articles weekly

2. VIP discord server with private chat and daily trade alerts/analysis from 4 analysts

3. Around the clock support/guidance from Michael, Kawz, and AB analyst team

[4] TTM Technologies ($TTMI)

Read more

SanDisk had an investor day yesterday and the stock rose 14% by the end of trading. David Goeckeler opened the conference with:

“as I stand here today, I feel like I’ve finally gotten to the starting line of where the real value creation is going to happen.”

Goeckeler’s confidence comes from a contract book which secures half of fiscal 2027 and two-thirds of fiscal 2028 bits into long term contracts that include a minimum guarantee. There’s a lot to discuss regarding this event; therefore buckle your seatbelts and read about the most significant event for memory of this year.

Remember paying subscribers have access to the VIP Discord where you can see trade alerts from our analysts and authors.

Before we begin, this Substack is for informational and entertainment purposes only and does not constitute financial advice. Past performance is not indicative of future results. Never invest more than you can afford to lose. Our writers may hold positions in the securities discussed and may buy or sell them at any time without notice.

What’s Inside:

1. Introduction (already read)

2. Before Investor Day

3. Long Term Profitability

4. NBM

5. Leveraging Supply

6. HBF

7. Our Take

[2] Before Investor Day

Sandisk reported their Q4 earnings nine days ago. Their headlines for this reporting period included:

→ Revenue was $8.97B, up 51% QoQ

→ Non-GAAP Gross margin was 84.6%

→ EPS was $39.25; beat the guidance of $30-$33.

All week after the report, this quarter has been a hot topic on X. Because of that, we wanted to focus your attention on one particular element; which is one specific sentence that explains this full cycle.

“...Sequential revenue growth came about one third from an increase in volume, and two thirds from an increase in price.”

Price change, relative to unit volume changes, had a significantly larger impact on the overall results. NAND supply constraints are providing pricing power for SanDisk, and therefore, the floor prices locked into their contracts will determine how much of that pricing power they can capture.

[3] Long Term Profitability

Cyclical this, cyclical that… If there is one thing you should remember from this conference, it’s this quote from Visoso himself:

“When you aggregate all of that for 2028 through 2030, we expect to grow revenue mid to high teens, consistent with [bit] growth... We expect non-GAAP gross margin to be around 80%. We expect non-GAAP operating margin to be 75%. How do we get there? We expect to spend about 5% in OpEx... And then you get to 50% adjusted FCF after paying for taxes, working capital, capital spending.”

Take a look at “grow revenue […] consistent with bit growth”. This means that if revenue and volume grow at the same rate, pricing stays roughly flat for three years, so they’re not modelling any future price hikes.

Gross margin near the 80% mark, with FCF margin of 50%. All that while running opex at around 5% of revenue is something unheard of. Not in memory, and DEFINITELY not in cyclical business.

Goeckeler (CEO) on the growth number:

“We’re committed to this kind of mid to high teens volume growth. When we look at the whole market, we think this is sustainable over long periods of time... it’s an input to a process of developing a whole fab strategy.”

As analysts already assigned $48.9B in revenue for FY27 and $58.3B for FY28, we decided to extend management’s CAGR to FY30… and we landed at approx. $79B in revenue.

If we assume 50% FCF margin that comes up to $40B of annual FCF against a $227B market cap today, look below.

After so much FCF, the money has to go somewhere. Someone decided to ask Visoso about the same topic:

“There is no trick. It’s 100% of the cash. Excess cash will go back to shareholders. You look at last quarter, we generated $5 billion. How much did we return to you? 4.5.”

[4] NBM

The New Business model works this way:

A hyperscaler commits to buying a fixed volume of bits (for X amount of years) → the trade is backed by financial guarantees (at prices with a floor) → the customer gets supply security → SanDisk gets revenue insurance for the downturn.

Per Visoso:

“These go to the highest level of the companies. They require board approval. We’re talking to CFOs, we’re talking to treasurers, we’re talking to CEOs.”

Their backlog right now is at 10 agreements, with 8 customers (3 of them are hyperscalers). Worth noting that the minimum revenue from these deals, at floor pricing, is $93.9B, and the average contract is around 4 years long.

So what happens to margins if prices crash to those floors?

Visoso said:

“Even at the lowest price, at the floor pricing, at any point in time, we don’t expect these to be below 80%... every single year over the period, we should be able to be, even at floor pricing, around 80%.”

Look at it once more, the downside scenario is an 80% gross margin company.

[5] Leveraging Supply

But we still wanted to know how these floors will survive, and luckily Goeckeler did the math, on stage, for us. He went through the BiCS (Bit-Cost Scalable) roadmap:

“Through the application of innovation, we can grow output per wafer at a rate of 27% a year. Which, by the way, tells you, you can’t just release nodes whenever they’re available. Otherwise you’re going to flood the market with supply and you’re going to have another 23’ situation.”

Explained in easier language: the technology COULD possibly grow output 27% a year but SNDK chooses to ship mid to high teens, which means everything in between is holding supply for pricing reasons.

That’s the sole reason why most of NAND growth through 2027 comes from upgrading existing factories, where each new process squeezes more storage out of the same wafer.

However, the fastest growing AI workload is KV cache (this is the memory that lets chatbots remember your conversation), which means that the demand is pulling another way. People using one chat longer or even models that remember who you are, all of it sums up to this one workload, and SNDK said its size is at zettabyte of storage by 2030.

What even is a zettabyte?

1,000,000,000,000 (1 trillion) gigabytes, nearly as much flash as the entire industry will ship this year (in phones, data centres, etc.).

Add this amount of demand to withheld supply and you get the pricing argument in one.

[6] HBF

HBF is basically a HBM, but with NAND stacked on top of each other, instead of DRAM. One key difference is: 8-16x more capacity.

When first announced 1.5 yrs ago, no one knew what they were looking at, but now they do. $SKHY, $GOOG, Tencent and (updated this week) $META, are all behind it as a consortium (an alliance of two or more independent entities).

The most important thing about it was the benchmark which was pretty insane. Based on a coding agent running on a 490B parameter model, the results were a lot better on HBF in comparison to HBM.

“With four HBF GPUs we are able to match the performance of eight HBM GPUs.”

When KV cache we talked about outgrows HBM, the GPU starts to fetch from slower memory systems which results in waiting almost half the time. HBF is big enough that cache never has to leave.

When asked about timing, and when they’ll ship the product, Ilkbahar (SanDisk’s CTO) said:

“I’m happy to share with you today that we actually taped out our first HBF memory die.”

First samples will go to customers in 2027, and surprisingly none of this sits in the FY30 model, which is what we love. HBF will be financed through opex and capex we covered, but the revenue from it is modeled at zero, so if it works, everything it earns is free upside.

As you probably know, when HBM took off, it ate DRAM production so much, that the regular DRAM got really expensive (because of how hard it was to get it). If HBF ramps the same way, people shorting NAND companies will have a hard time too.

[7] Our Take

Ben Reitzes at Melius put the valuation absurdity better than we could:

“I’ve been going to tech conferences and analyst days since about 1992, and I’ve never seen a company guide for a year three years out and be trading at less than three times that number.”

And we completely agree with that angle. is now trading at around 6x fwd P/E, essentially same as . However, if you look at or this multiple doesn’t make any sense, even looking at the growth itself (without touching books).

TLDR:

We remember that every time memory crashed in the past, customers found a way out of LTAs. The problem is even bigger when you account for and others adding new capacity starting in 2027 (these floors haven’t faced a test yet).

However, if the company wanted to convince people (memory investors specifically) that the old cycle of memory is gone, the proof would’ve looked exactly like this (Withheld supply, floor pricing, guarantees from banks, upgrading rather than expanding).

We’re long memory, and yesterday only reassured that.

You can follow us directly on Autopilot, where our memory portfolio is up over 100% in the last 6 months.

In June, we mapped the optical supply chain in our shortage report, and LITE scored at the top of the entire stack. The company grows the InP laser chips that every 800G and 1.6T transceiver depends on, and quoting their representatives:

“We can’t make enough of them.”

Demand from other layers of the chain keeps confirming it too, with AAOI telling us on Thursday’s call that the demand for their products runs 20-40% above what they can build.

This is a two-part piece.

Part one breaks down the company and why it sits where it does in the optical stack. Part two covers tonight’s earnings print.

Remember that our subscribers get access to the VIP Discord, where our analysts, as well as our co-founders (Michael and Kawz) share their trade ideas.

Here’s everything you need to know if you want to invest in .

Before we begin, this Substack is for informational and entertainment purposes only and does not constitute financial advice. Past performance is not indicative of future results. Never invest more than you can afford to lose. Our writers may hold positions in the securities discussed and may buy or sell them at any time without notice.

What’s Inside

Part 1: The Company

1. What Does Lumentum Sell?

2. The Laser Position

3. CPO / NPO Timing

4. Capacity Build

5. China Ban

6. The Margin Rebuild

Part 2: The Earnings

7. The Print

8. The Guide

9. Key Information From Q4 Call

10. Capex and Capacity

11. Valuation

12. Risks

13. Our Take

What Does Lumentum Sell?

Lumentum reports two product lines: Components and Systems.

The components side is why this stock look appealing for anyone invested in optical. Lumentum makes the EML (Electro-Absorption Modulated Laser) laser chips that go into majority of 800G transceivers shipped today.

Their customer list includes the biggest names in the module industry: Innolight, Eoptolink, Fabrinet, who assembles for NVIDIA.

The conclusion here is easy = if Lumentum can’t ship enough lasers, the whole transceiver supply chain slows down.

EMLs are the volume product and driver of the company.

However, there are also pump lasers and narrow linewidth laser assemblies that both go into the scale-across links between data centers, and both lines are growing faster than EMLs right now.

→ Pump lasers grew +80% YoY last quarter

→ Narrow linewidth shipments grew +120% YoY (it grew for ninth straight quarter)

Hurlston said pump supply got tight enough that they’re now choosing which customers get allocation.

The Systems side of the business came together through the Cloud Light acquisition, which turned Lumentum into a transceiver maker shipping 800G and 1.6T modules directly to hyperscalers. Cloud transceiver shipments were supporting the growth of the company, and this part of the business grew over 40% Q/Q. Then there’s OCS (the optical circuit switches) that networks which are Google-ish use in place of electrical ones.

That product line is backed by a purchase agreement worth several billion dollars, announced at OFC in March. Hurlston called OCS:

“The biggest single tightrope that we’re walking”

He said that because the ramp depends on 2nd tier supply chain parts showing up on time. Furthermore, on the last call, Hurlston told analysts how tight overall supply is:

“We had conversations today with customers, you know, significant customers looking to really up their demand, and get output from us, and we simply can’t service that.”

The Laser Position

Lumentum ($LITE), Broadcom ($AVGO), and Mitsubishi Electric ($6503.T) control 72% of global EML capacity according to TrendForce, with Lumentum being first. The whole industry is more than 30% short on EML supply right now, and it will only get worse as NPO demand surges on top of existing orders.

Pricing reflects the tightness as the ASP on 200G EML is roughly twice that of a 100G EML, and the majority of Lumentum’s capacity is already locked under LTAs. Negotiations for 2028 and 2029 LTAs are underway at higher pricing. Pump laser LTA discussions are also in progress.

Furthermore, the high-power CPO laser market is a Lumentum and Broadcom duopoly, with Coherent behind by estimated 18 months (depending on the stability of their MOPA architecture), and AAOI roughly 2-3 years out (all according to Irrational Analysis). Neither of them (AOI or COHR) has publicly shown phase noise or linewidth data under real operating conditions.

That leaves LITE (out of the big three: LITE, AAOI and COHR) supplying lasers into every architecture that’s important today and every one on the roadmap.

CPO / NPO Timing

When talking about architectures, we should explain them one by one. There are three main optical architectures competing to connect AI clusters.

Pluggable transceivers are the volume product today. Every 800G module shipping right now uses a laser from Lumentum or one of the few other suppliers. Transition to 1.6T is already underway, with Lumentum ramping higher-ASP 200G EMLs and shipping 1.6T transceivers from the Cloud Light platform.

NPO is the next architecture approaching production. Lumentum has three NPO programs ranging through CY27-28, and management said their UHP (Ultra-High-Power) laser capacity is fungible between both CPO and NPO, which is beneficial on short turnoaround. Worth noting that NPO is attracting a broader customer base beyond NVIDIA, particularly platforms looking for better power efficiency without committing to CPO…

CPO carries the biggest content multiplier (and it also sits further out). A high-power CPO laser consumes 5-10x the InP die area of a 100G EML. NVIDIA remains on track for initial CPO shipments in Q3 CY26 with Q4 inflection. Volume scale-up CPO won’t arrive until the Feynman generation, but the laser orders land well before that.

So the takeaway for the InP demand is simple, each architecture transition increases laser content per system, and all architecture run through the same fabs, increasing revenue streams for LITE.

Capacity Build

Luckily, same as AAOI, LITE is expanding at every level of the supply chain.

The current capacity add comes from completing the transition to 4-inch InP wafers and expanding the Japan fab. Together those should push laser output up roughly 50% from December 2025 to December 2026, additional capacity will be added in the UK fab layers during CY27.

The big one is Greensboro. That fab comes online in 2028 with roughly $5B of annual revenue capacity, and half of it is already committed to NVIDIA through a LTA. Needham expects that remaining capacity to be sold out quickly.

Moreover, as most of you probably know, in March, NVIDIA took a $2B equity stake in Lumentum and Coherent combined, with a purchase commitment attached worth several billions. Preferred stock converts into common shares and adds roughly 7M to the diluted count this quarter, bringing it to about 102 million.

The dilution here is hard to ignore, so you should be aware of it when investing in the company.

In addition, the substrate bottleneck is starting to ease. AXTI is on track to 2x InP substrate capacity in 2026 and expand further in 2027. Export permits out of China are becoming more regular and the only thing that could stop it is an InP substrates export ban coming from China (we’ll talk about that in a second).

Morgan Stanley’s assessment is that internal fab capacity, is now the primary constraint on how many lasers Lumentum can ship.

China Ban

Read more

We’re entering an optical trade war.

just lost a fifth of its market cap after Chinese scrutiny of dual-use optics deferred $17 million of revenue out of the quarter, and the FCC is drafting a ban on new Chinese optical transceivers that could land before year end. Every company that touches light and depends on China is now a liability, and a small handful that don't are about to get repriced.

LightPath Technologies has been on the radar of many small cap stock pickers and defense specialists because of its unique non-China positioning. The stock is up 3.5x in the last year on that exact thesis.

Most infrared optics run on germanium. China controls the supply and cut off exports to the US, which left the entire Western defense supply chain scrambling for a substitute.

LightPath spent years developing BlackDiamond, its own chalcogenide glass that does the same job with no germanium in it. They own the material, they mold it in-house, and it is already designed into programs.

The second piece is that they stopped selling just lenses. With the Visimid acquisition they can now deliver the full thermal imaging module, sensor and electronics included. Same customer, much bigger dollar content per unit, and far stickier once it’s designed in.

Put those together and you get a components supplier turning into a systems supplier at the exact moment drone demand goes vertical.

We got CEO Sam Rubin on a call for half an hour to ask him what’s actually going on inside the business.

In this article, we’ll save you time from watching the entire interview, sharing the best snippets and what they mean for investors in a quick read.

Before we begin, this Substack is for informational and entertainment purposes only and does not constitute financial advice. Past performance is not indicative of future results. Never invest more than you can afford to lose. Our writers may hold positions in the securities discussed and may buy or sell them at any time without notice.

Read more

reported Q2 2026 after the close today.

• Revenue $191.9 million, up 86% YoY, up 27% QoQ

• Non-GAAP EPS $0.06, beating consensus of $0.02 by 200%

• Non-GAAP gross margin 29.8%

• Data center revenue $107.7 million, up 140% YoY

• 800G revenue $12.8 million, up more than 10x YoY

Every number landed inside or above the guided range. Non-GAAP EPS of $0.06 came in above management’s own guidance of ($0.03) to $0.03, but also the $0.02 Street consensus. This was AOI’s 5th consecutive record revenue quarter, and the first quarter of non-GAAP profitability since the ramp began.

CFO Stefan Murry called AOI’s high power narrow linewidth laser

“the best in the industry, the best in the world right now.”

AOI has always been a laser company and management reaffirmed it so people could stop calling them the transceiver assembler. They grow their own InP lasers on proprietary equipment while every other module maker sources from the already sold out suppliers.

Read more

VIAV reported after the close today and beat the top end of guidance on every single line.

• Revenue $443.1 million, up 52.5% YoY

• EPS $0.34, up 161% YoY

• Operating margin 24.0%, up 960 bps YoY

• NSE revenue $353.9 million, up 69.2% YoY

Every number cleared the high end of the guided range. Full year EPS landed at exactly $1.00, up from $0.47 last year, confirming…

Read more

Fluence Energy is the largest standalone battery storage integrator in the Western world. Every piece of a utility scale battery system, the cells or the power electronics, is sourced by Fluence. They assemble it, and run it with their own software for decades.

Think of a general contractor for the power grid.

FLNC reports fiscal Q3 on today after close and the stock sits at roughly $15, down 58% from its ATH.

The stock has built a ton of hype on X, as many retail investors position into the name with influencers like Kevin Xu calling it his all in play for the week.

On the May call, CEO Julian Nebreda said the first hyperscaler purchase order was expected “within the third quarter” and that quarter ended June 30.

The stock jumped 40% on the last earnings because of a mention of two hyperscalers. If FLNC cites one or both of those purchase orders landed in Q3, the stock could go parabolic, making an earnings swing trade set up very interesting.

Remember paying subscribers have access to the VIP Discord where you can see trade alerts from our analysts and authors. Michael snatched 15% off Aster’s 2x ETF last night before AMD’s report dragged markets down.

Here’s everything you need to know ahead of going into this earnings release.

Before we begin, this Substack is for informational and entertainment purposes only and does not constitute financial advice. Past performance is not indicative of future results. Never invest more than you can afford to lose. Our writers may hold positions in the securities discussed and may buy or sell them at any time without notice.

Read more

We’ve been tracking as part of our testing infrastructure coverage since April, when we published our piece on who captures the photonics testing TAM.

It’s a core position in our well known Photonics portfolio published publicly on Autopilot.

The thesis was that every transceiver that ships has to be tested and VIAV owns a dominant share of the optical test market. We allocated 20% of our photonics fund to testing names off the back of that research. Since then, , , and have all reported blowout quarters confirming that test demand is running hot across the semis supply chain.

VIAV has beaten its own guidance four quarters in a row.

Both revenue and EPS cleared the top end of management’s range and the stock responded with a 32% move on earnings day in April, hitting $60 before getting dragged back down in the broader Black July selloff. Shares sit at roughly $37 as of Friday, about 38% off that high.

Remember to join the VIP discord to see if any of our analysts will be swing trading VIAVI earnings.

Before we begin, this Substack is for informational and entertainment purposes only and does not constitute financial advice. Past performance is not indicative of future results. Never invest more than you can afford to lose. Our writers may hold positions in the securities discussed and may buy or sell them at any time without notice.

Read more

Applied Optoelectronics () reports their Q2 2026 on Thursday, August 6 at 4:30pm ET.

In a few days, the stock has sharply reversed since its lows under $80 a share and is exploding 14% today as the market appreciates Iran diplomacy, falling oil, and absent selling pressure after the Situational Awareness implosion.

We’ve been covering since $53/share and have frequently cited it as one of the most asymmetrical AI stocks long term. It peaked at a roughly 4x position for us when it surpassed $200, and more recently has given up much of those gains. Even with a sharp rebound, the stock is off its ATH by roughly 50%.

If you own AAOI stock, it is critical you know exactly what to expect on the earnings call, because this is a business still early in its inflection.

In this article we’re going over exactly what you need to understand going into this week’s set up.

Before we begin, this Substack is for informational and entertainment purposes only and does not constitute financial advice. Past performance is not indicative of future results. Never invest more than you can afford to lose. Our writers may hold positions in the securities discussed and may buy or sell them at any time without notice.

The Setup

Management has said on four consecutive earnings calls that customers want more 800G and 1.6T transceivers than AOI can physically build. In May they put a number on it. Real demand this year is $1.4 to $1.5 billion against a revenue guide of “over $1.1 billion”.

That leaves one open question. Can they bring capacity online fast enough to convert those orders into revenue?

H1 2026 was guided to be roughly a third of the year. The 2nd half carries everything:

• new Texas and Taiwan capacity,

• the 800G ramp going vertical,

• first 1.6T shipments, and

• the profit inflection.

So, Thursday is the first stress test on whether that back half is on schedule.

This means that the Q2 numbers themselves are close to irrelevant but there are three things that matter:

Read more

We’re coming off one of the worst Julys in recent history for semis stocks. The SOX dropped roughly 24% from late June highs, and it was the sharpest drawdown since 2022. Thankfully, the last week happened, with amazing earnings for companies such as:

• , which went up 50% after hours,

• gained 13%, and

• added 17%.

Capex guidance was raised by Amazon, Microsoft, Meta and TSMC and then we finally caught a bid.

This week is one the biggest earnings week of the quarter, and we picked out 7 names that we think every investor should be watching that could possible make for good earnings trades. For each stock, we’ve listed what to look for, and what numbers are the most important. Per usual, paid subscribers can see which of these stocks our authors might swing trade in the VIP Discord.

Before we begin, this Substack is for informational and entertainment purposes only and does not constitute financial advice. Past performance is not indicative of future results. Never invest more than you can afford to lose. Our writers may hold positions in the securities discussed and may buy or sell them at any time without notice.

Read more

We’ve been getting the same message from you all week, especially on Tuesday. Different words, same question:

• “Is It A Bubble?”

• “Is The Memory Trade Over?”

• “Is The Credit Spread Widening A Fundamental Crack?”

• “Can Capex Keep Going Up?”

We understand it because when everyone is screaming “bubble”, the dotcom comparisons start to feel oddly real.

We’ve been seeing these headwinds build on top of each other for weeks, the AI trade had gotten crowded, and the market had priced in a world where capex only goes up and nothing goes wrong.

So we did what we always do when the noise gets too loud → we went back to the data.

We pulled the numbers from every major crash of the last 25 years: the dotcom bust in 2000, the subprime crisis in 2008, covid in 2020, the 2022 bear market, and the 2025 tariff selloff. Every single one of them had a clear macro catalyst behind it.

Black July doesn’t have any of that, this selloff is something completely different from everything on that list, and understanding why matters for what comes next.

What’s Happening Now

The SOX doubled during the first six months of 2026 and from its April 2025 lows, the index rallied roughly 300%. Semiconductor stocks led the entire market higher, powered by an AI infrastructure buildout that showed no signs of slowing down.

Then July hit.

The SOX and DRAM both peaked on June 18 and have fallen respectively by roughly 25% and 40%. Samsung is down 30% from its ATH, followed by SK Hynix down almost 40% from its highest price. Furthermore, CXMT went public in Shanghai at a 472% pop, and the timing poured gasoline on an already burning memory complex.

Meanwhile the earnings estimates behind theses stocks haven’t moved. Sandisk’s forward P/E has compressed to 6.4x despite being up over 350% YTD, and Micron trades at 5.7x forward earnings. The stocks are falling while the estimates are rising, which means the multiples are shrinking from both directions at once. Moreover, NVIDIA is trading at its lowest point in over 8 years in terms of forward multiples, despite growing over 65% YoY.

The S&P 500 is sitting near all time highs, and SOX is almost in its bear market state. That divergence, the broad market holding steady while the sector that led it higher gets gutted, is one of the defining features of this sell-off. In prior crashes, everything went down together. This time, the pain is concentrated in one trade.

What These Crashes Have In Common

The dotcom bust started in March 2000 and lasted two and a half years. The NASDAQ fell 77% from 5,048 to 1,139. What’s more interesting? The NASDAQ P/E had reached 90x by 1999, and the average P/S ratio for companies going public in 2000 was 48.9x.

74% of surveyed internet companies had negative cash flows and most of them had no revenue model. The Fed had hiked rates six times in 10 months to 6.5%, Japan entered a recession and Barron’s ran a cover story documenting how quickly these companies were burning through cash. $5 trillion in market value was erased and the trigger was a speculative bubble built on companies that had NEVER earned a dollar, most of them had one thing in common: “.com” domain.

The subprime crisis started in late 2007 and bottomed in March 2009. During these years the S&P 500 fell 57% from 1,565 to 676. Household net worth declined by $13 trillion, and 8.5 million jobs were lost. Bear Stearns went under in March 2008, Lehman filed in September and investment banks were running leverage ratios of 33:1. Subprime mortgage originations had gone from 5% of total lending in 1994 to 20% by 2006. It’s also worth noting that the semiconductor sector fell roughly 55% despite having nothing wrong with it operationally, the trigger was a credit system that had loaded itself with $1.3 trillion in subprime debt and then collapsed under its own weight.

If any of this sounds familiar, you’ve probably seen The Big Short.

The Covid crash was the fastest 34% drawdown in S&P history. It took five weeks from the February 19, 2020 peak to the March 23 trough. The VIX hit 80 and the trigger was the global pandemic shutting down the physical economy overnight. There was no earnings deterioration story or no credit crisis. The market sold everything because nobody knew if the world was about to stop functioning, and then the Fed stepped in with unlimited liquidity and the entire drawdown was erased within months.

The 2022 bear market was slower and more methodical. The S&P 500 fell 25% from January to October. The Fed raised rates from 0.25% to 4.5% in a single year, the fastest tightening in over 40 years, and the bond market had its worst year since the 1970s. Consumer electronics demand collapsed and Micron’s revenue fell 49% in one fiscal year. The trigger was the fastest rate hiking cycle in decades crashing valuations first, and then demand actually deteriorated to confirm what the market had already priced in.

The tariff crash in April 2025 was the most avoidable crash out of all on this list. The S&P 500 dropped 12% in the first week of April after the Liberation Day tariff announcement, nearly touching bear market territory. Then Trump paused reciprocal tariffs for 90 days and the S&P rallied 9.5% in a single session. The full drawdown was erased by end of may and the S&P set new all time highs within 20 weeks. The trigger was a trade policy shock that reversed itself almost immediately.

Noticed how we pointed the trigger in every single crash mentioned? The reason is pretty simple:

Every situation described had a macro catalyst you could point to:

• A speculative bubble built on horrible companies with astronomical multiples

• A financial system that was leveraged 33:1 on subprime debt

• A pandemic that shut down the global economy and you couldn’t go into the story for a couple of months.

• The fastest rate hiking cycle in 40 years and a trade war that slapped 125% tariffs on Chinese imports overnight.

EACH time, you could look at the wreckage and trace it back to one specific problem in the real economy. However, Black July doesn’t have that, and why this selloff is happening will be described in the upcoming sections.

The Changing Factor

Every crash we studied had one variable that determined whether the recovery took months or years:

Were the earnings actually declining when the stocks fell?

Starting with the dotcom bust, there were no earnings to decline. Most of these companies had never generated a dollar of profit. Netscape went public at $2.2 billion market cap without ever making money, and it was one of the better ones. When the bubble popped, there was nothing underneath to catch it. It took 15 years to recover to the peak levels because the profits that were supposed to justify the valuations never showed up.

In the subprime crisis, the chip companies were making money through the entire drawdown, but SOX still fell 55%. Once credit market stabilised, the recovery came relatively quickly. The S&P reclaimed its pre-crisis level within four years of the March 2009 bottom.

Covid genuinely shut the economy down and earnings took a hit across most sectors. But the companies themselves weren’t broken. Amazon added $105 billion in revenue in a single year as physical retail closed its doors. Zoom went from a small conferencing tool to the way the entire world communicated. The businesses adapted, the Fed stepped in with unlimited liquidity, and the full 34% S&P drawdown was erased within months.

In 2022, earnings were actually getting worse. Samsung’s semiconductor revenue plunged 38% as the memory market cratered. Intel dropped 17% on collapsing PC and server shipments. The global semi industry lost 8.8% in aggregate revenue. The stocks were falling and the fundamentals underneath were confirming the decline. The market didn’t bottom until October, once the trough in earnings became visible.

The tariff crash followed the covid playbook. Earnings were intact, the threat reversed within days, and the S&P recovered within weeks.

Black July has accelerating earnings, and record margins. NVIDIA trades at its lowest forward multiple since before the AI cycle existed. Memory names sit at single-digit forward earnings. TSMC just printed a record quarter. The stocks are falling while earnings go up. Every time we found that pattern in the data, the recovery came in months.

Why Is This Happening?

We don’t see a credit system falling or global pandemic that could cause this kind of drawdown in semis. What’s been happening is a pile of narratives all hitting at the same time:

• The Iran war reigniting inflation and rate hike fears

• Chinese competition from CXMT

• Kimi K3 questioning whether frontier labs can hold their margins

• OS models threatening the moat of closed models AI companies

• AI effiency concerns about oversupply

• Rising credit spreads on hyperscaler debt

Three months ago, the market treated AI capex as untouchable. All of the fears we listed above have been testing and breaking the narrative over the course of a single month. The market started to have concerns whether the spending can even continue.

Oil, possible rate hikes and credit spreads made the buildout look more expensive to finance. Kimi and OS models made it look unnecessary, because if intelligence gets cheaper, maybe you need fewer GPUs, and china made the returns on all of it look less certain. All of these are different fears but they point in the same direction:

“Capex cuts are coming.”

However, so far, the actual capex numbers haven’t changed. and both raised their guidance and capex this quarter and language from their representatives was positive.

The narratives gave the market a reason to sell and leverage gave it the mechanism to do so.

South Korean retail spent H1 2026 loading up on Samsung and SK Hynix through leveraged ETFs and margin accounts. Over a third of total market margin in Korea sat in just those two stocks. When Korea raised rates 25 basis points in mid-July, the first hike in three years, it jenga blocked the most crowded trade in Asia.

Furthermore, Goldman Sachs data reported that roughly 1.2 million retail accounts faced margin calls in a single week and 350 thousand were forcibly liquidated. Cumulative forced liquidations in July hit 344 billion won. SK Hynix dropped 15.37% in one session, then posted back to back 10%+ down days after that. Korean retail kept buying while foreign and domestic institutions sold, making the forced selling worse with each leg down.

But leverage wasn’t the only thing hitting the sector. The Iran conflict pushed oil prices higher and brought inflation fears back, raising the chance of rate hikes right when the market had priced in the Fed staying put.

On top of that, the debt load across AI infrastructure is finally making people nervous. NVIDIA’s CDS costs surged by the most on record after reports that the company is in talks on over $750 billion in AI infrastructure deals. Hyperscalers are spending $700B+ in capex this year, and for several of them that spending now absorbs most or all of their operating cash flow. They’ve responded by issuing massive volumes of investment grade bonds, and the flood of new paper has pushed credit spreads wider.

We think the debt panic is overdone, and Gavin Baker made a great point last week. The companies that signed contracts for GPU compute in 2024/25 are currently paying roughly half of what the spot market charges today. That means hyperscalers are under earning on their existing capacity right now → as those contracts expire and reprice to current spot levels, cash flow accelerates without building a single new data center.

Operating cash flow growth is moving from 31% in Q1 to 50% in Q2, and we think consensus is wrong to expect a slowdown in Q3. If that plays out, the funding gap the credit market is worried about closes on its own.

Let us be clear. None of this represents a fundamental deterioration.

The debt is financing demand with real returns, the war is a geopolitical risk and rate hike fears might come away as soon as the war ends. However, when all three hit a sector that doubled in six months on leveraged positioning, it doesn’t take a crisis to produce a this type of drawdown.

We have found one quote from an analyst we think you should hear:

“the lesson from Korea is not that the AI cycle is over. It is that even a powerful structural story can be broken temporarily by the price paid for it and the leverage used to own it.”

Bottom Line

Take a look at the table below. Five tech selloffs since 1998 and every single one recovered. The median drawdown was 32.4% and the median time to new ATHs was 282 days.

We don’t know where the exact bottom is (nobody does), but the data where earnings are growing and no macro crisis sits behind the drawdown is pretty clear. They recover and they tend to do it fast.

There’s also a catalyst forming that the market hasn’t priced in yet. The US and Iran have paused strikes and Oman is mediating talks over the Strait of Hormuz. If a peace deal finally (!) materializes, oil prices come down → inflation fears ease → the rate hike narrative dies, and one of the biggest overhangs disappears overnight.

We covered this angle in our ERII piece and the logic applies here too. A resolution in the Middle East removes the single macro risk that’s giving the bears a huge argument right now.

Where’s our head in this situation?

We're not calling a bottom and we're not telling anyone to go all in today. However, we’re saying is that the conditions that produced every bear market / bubble pop we studied are absent right now.

The stocks that are down 30-50% in a month are the same ones printing record earnings and trading at the lowest forward multiples in years. If sustained capex cuts from hyperscalers show up or data center orders collapse in the this round of earnings, we'll reassess.

Until then, nothing changes from our side as we are still invested in the industry.

As the air gets sucked out the AI trade, our eyes are on retail favorite Reddit. While it’s not an AI capex beneficiary, Reddit is undeniably a great business that fits the high growth profile investors might look for outside the semiconductor trade. So can this non-AI growth story still catch a bid amid slowing momentum?

It ran from $136 to $203 on the data repricing thesis, got hit for 8% on a WSJ headline about the Google contract, and now sits at $184. In the span of three months the market priced in bear and bull cases.

Today’s print will finally settle the debate. At roughly 40x (NTM) forward earnings with 12% short interest, Reddit doesn’t get the benefit of the doubt. The stock has to earn its spot here.

Before we begin, this Substack is for informational and entertainment purposes only and does not constitute financial advice. Past performance is not indicative of future results. Never invest more than you can afford to lose. Our writers may hold positions in the securities discussed and may buy or sell them at any time without notice.

Key Terms

DAUq (Daily Active Uniques) - Reddit’s count of unique people who visit in a day, logged in or not.

WAUq - The weekly version, softer reach signal, and matters less for monetization.

Logged-out users - These are people without an account, mostly arriving from Google search. Nearly 60% of daily users.

ARPU - Average revenue per user (quarterly revenue divided by DAUq). The core of the bull case since Reddit monetizes each user well below Meta or Pinterest.

Adjusted EBITDA - Reddit’s preferred profit metric, excludes stock comp.

On valuation - When we say roughly 40x forward earnings, that’s NTM (next twelve months) EPS. On calendar 2026 estimates it’s ~37x, on 2027 ~28x.

What Happened Last Quarter

Reddit reported Q1 on April 30 and beat every consensus across the board.

• Revenue was $663 million, up 69%

• Ad revenue hit $625 million, up 74%

• EPS came in at $1.01 against a $0.58 estimate

This sent the stock up 13%.

EBITDA margin crossed 40% for the first time, up almost 1,100 basis points from a year ago. FCF hit $311 million on $1 million of capex. Reddit is printing cash like a mature software business while still growing revenue 69%.

Global DAUq grew 17% to 126.8 million but daily US logged in users grew just 1%. Huffman said on the call that:

“DAUq is the primary focus of the company because revenue is doing very, very well. DAUq is both our mission and also fuel for the business.”

The goal is 100 million US daily users, up from around 54 million. There are already 200 million Americans on Reddit every week, so the target is getting those weekly visitors to come back every day.

ARPU picked up the slack, it jumped 54% to $9.63. Global ARPU was up 44% to $5.23 and international ARPU hit $2.02, up 51%. Jen Wong made the point that the value of an ad impression is basically the same whether the user is logged in or not, logged-in users just spend more time and see more of them.

As we covered in our last post, Huffman said Reddit is the most cited source across all AI platforms and called the Google and OpenAI deals important beyond just revenue. He put it simply on the call:

“there is no artificial intelligence without actual intelligence, and that comes from Reddit”.

What the Stock Has to Prove

Read more

Three months ago capex looked indestructible. Hyperscalers kept raising guidance. Earnings kept beating. That combination is what held the market up through the Iran war, when nothing else could.

That’s not the setup anymore. Over the past two months the capex story has come apart from several directions at once:

• Oil higher, and inflation with it

• Rates staying high, which is what finances the buildout

• Credit spreads widening

• Chinese open source models closing on frontier models, which pressures the margins that justify spending

None of these broke the trade on their own, but they’ve been slowly deteriorating sentiment for tech. In fact, the worst momentum stock drawdown ever recorded has happened over the last month.

That’s the anxiety going into this afternoon.

The FOMC decision will land this afternoon, and markets are pricing a hold at 3.50 to 3.75%, this is also what we expect. We’ll be focusing on Warsh’s language in the press conference afterward, specifically whether he gives any signal about September. This meeting won’t come with updated future projections, so the press conference is all we have to work with. We think that understanding how these two machines work (inflationary demand now, and deflationary productivity later), is the single most useful lens for reading whatever Warsh says.

1. Uncertainties

What’s Still In Question?

The Fed doesn’t have every answer right now, and at this point in the cycle that’s normal. Cook said she’s watching for disinflation and will act if she doesn’t see it. However, the committee is split on where rates should go by year end, and Warsh has stayed out of that debate by not submitting a dot of his own. We don’t think that split is a warning, because the committee arguing over where rates should go while the chair keeps his options open is how the Fed is supposed to work when the data is still moving (especially with a technology as new/uncertain as AI).

The productivity payoff from AI is in the exact same spot, because we know the investment is here and we can see it flowing through the economy. Yet, we can’t see how much of that spending is making the economy more efficient.

The most important factor, outside of the US, is oil. WTI near $80 is manageable but another escalation in Iran pushes that toward $100 and changes inflation math in a way the Fed would have to respond to.

What the Macro Selloff is

The market treated Warsh’s decision to drop forward guidance as the Fed going dark, but he explained why he did that:

“I think financial markets perform best when they react to incoming data.”

During Powell’s cadence markets have been trading off whatever the Fed says it might do next. We believe that Warsh’s way of guiding the Fed, which is trading off what the economy is doing, is much better.

The dot plot got the same market uncertainty treatment as the dropped guidance. Warsh called those submission pencil marks, meaning they reflect where each person sits today, rather than three months later.

2. Where Rates Actually Stand?

The FOMC wraps its meeting this afternoon, and markets expect a hold at these levels, which would make five consecutive meetings with no change.

At 3.75%, rates sit below both the 2006 and 2023 peaks of 5.25%. GDP is growing around 2.2% with unemployment holding steady. Warsh described the effect as uneven because housing feels the pressure but AI investment, the labor market, and corporate earnings mostly don’t.

What the Dot Plot Shows

The median FOMC participant projected rates at 3.8% by year end at the June meeting, which works out to roughly one 25 basis point hike from today. The spread around that median runs wide. Some participants see room to hold near 3%, and others cluster closer to 4. According to CME FedWatch as of last Friday, 62% of the market expects a hold today and the remaining 38% is pricing a hike to the 3.75 to 4.00% range.

Inflation needs to sit comfortably below 3% and keep trending toward 2.5 before cuts become something possible and it’s at 3.7% right now. Getting there takes months of sustained cooling, and that will get tested in September. This will come with fresh projections and a new dot plot.

Trump told reporters aboard Air Force One two days ago that the U.S. should have the lowest interest rate in the world. He called Warsh “fantastic” but said the board members are “very political” and “perhaps have bad intentions”.

3. How Warsh Is Rebuilding the Fed

Warsh spent his first month doing something recent Fed chairs haven’t, he went through how the institution runs and questioned parts that hadn’t been reviewed in years. At his June press conference he said:

”Start with first principles, ask hard questions, examine current practice, consider alternatives, propose next steps”

Five task forces came out of that, each aimed at a different piece of how the Fed operates.

We think this matters more than the market is giving it credit for right now. A Fed that finds its own blind spots before the next cycle is a Fed that makes fewer of the mistakes that cost everyone money in 2021 and 2022. The volatility through July is what it looks like when that kind of review happens in public instead of behind closed doors.

The Five Task Forces

1. The communications task force is reviewing whether press conferences after every meeting and the dot plot add clarity or noise. This group decides what the new communication framework looks like going forward.

2. The balance sheet task force is reviewing the trillions in bonds the Fed still holds from pandemic buying. That pile of bonds created a system called ample reserves, and it’s never been tested through sustained inflation. If this group recommends shrinking the balance sheet, financial conditions tighten without anyone touching rates. Of the five, this one carries the most direct risk for markets if it lands on the wrong side.

3. The data and methodology task force is the one we’re watching most closely. Most of the Fed’s economic data comes from survey methods designed for a very different economy. Private companies already track their own performance in real time but The Fed waits on monthly government releases that get revised for years after they come out. Lack of this task force is what went wrong when the Fed called inflation transitory in 2021 and stuck with that read too long.

4. The productivity and AI task force is checking whether AI makes the economy more productive in a lasting way. If the answer comes back strong, the Fed gets room to hold rates lower than current inflation data alone would suggest, because faster productivity growth means the economy can run hotter without prices spiraling.

5. The inflation frameworks task force is rebuilding the model that produced the transitory call. Warsh said the 2% target stays where it is and the methods and assumptions underneath it are what’s getting rebuilt.

4. Inflationary First, Deflationary Later

Data centers already draw roughly 3% of U.S. electricity, that number is projected to climb toward 8 to 12% by 2030, and every new facility pulls raw materials, and skilled trades away from other projects. The cost pressure spills into the rest of the economy, causing elevated prices.

Cook flagged this in her July 15 speech. Core goods prices are running at 5% annualized in 2026, a category that spent two decades getting cheaper before the pandemic. AI hardware demand is the reason why this trend reversed. Equipment investment grew 8% year over year through Q1 2026, and high-tech spending within that grew nearly 25%.

That’s the inflationary machine running right now.

The deflationary machine takes longer to show up.

A study published in the QJE, cited by Barr in his speech, found AI cut professional task completion time by around 40% and improved quality by about 18%, with the largest gains going to the weakest performers. Once gains at that scale move beyond controlled studies and into the broader economy, the same spending driving prices up today starts working in the opposite direction.

Furthermore, Trump posted on July 15 calling data center jobs and tax revenue “LIQUID GOLD” and criticizing New York for blocking projects. States are now actively competing for these builds offering cheaper land and faster permitting. Federal energy approvals and permitting sit outside the Fed’s control entirely, so even if the Fed wanted to cool this spending through higher rates, the political side is working to accelerate it at the same time.

5. The Task Force Calendar

Task forces start work in late July, findings come out through the fall, and most wrap by year end. Two CPI prints land before then, in August and September, and those will show what the June’s number was.

September FOMC comes with fresh economic projections and a new dot plot, making it the first meeting since June where the committee has to put updated numbers on the table.

6. What Could Turn This Bearish?

• CPI accelerating above 4% is the highest probability risk, Cook named that level as the point where markets drop 2 to 3%. At 4%, a rate hike will be the base case.

• A hyperscaler pulling back on capex guidance would drive semis industry even lower. Microsoft and Meta report today. We’re watching their capex commentary specifically.

• Oil staying above $85 keeps headline inflation hot, and WTI has been volatile through July on Iran tensions, up roughly 20% for the month.

• The September FOMC meeting, new projections and dot plot mean the committee has to show its hand rather than hold and wait. If inflation hasn’t cooperated by then, hike risk goes from a background concern to the main conversation.

• Credit spreads widening in high yield bonds tend to show stress before GDP data does.

• Taiwan escalation carries low odds but would physically disrupt the supply chain the entire buildout runs through.

7. Takeaway

The FOMC decision lands at 2:00 PM ET and we expect a hold. Warsh’s press conference is at 2:30 PM and we’ll be listening in order to find any suggestions about September.

We’ve been through most transcripts from this month. The Fed isn’t hostile to the AI buildout, but they’re cautious about inflation while it also rebuilds its own framework. The political side is pushing the opposite direction on rates, but if we trust the core assumption of Fed (which is independence from the President), Trump’s comments aren’t changing their decision.

On Wednesday Reddit closed at $170.38, down 8.32%, losing roughly $3 billion of market value. The trigger was a Wall Street Journal story reporting that Reddit executives have internally discussed cutting off Google’s access to Reddit content for AI training.

The contract pays about $60 million a year and the talks are still open, per people familiar with the negotiations. No decision has been made yet.

We think the market got this selloff wrong. The stock moved by roughly 50x the annual value of a deal that, based on the Journal’s own reporting, looks more likely to be renewed at a higher price than killed. Wednesday’s reaction treated Reddit like just another publisher losing traffic to AI but Reddit is renegotiating with the customer, rather than losing him.

Before we begin, this Substack is for informational and entertainment purposes only and does not constitute financial advice. Past performance is not indicative of future results. Never invest more than you can afford to lose. Our writers may hold positions in the securities discussed and may buy or sell them at any time without notice.

What the Journal Reported

Reddit was one of six companies described in the piece. There were also:

• USA Today,

• Politico,

• Reuters,

• The Economist and

• People Inc.

They’re all rethinking whether to keep supplying Google with content because their search referral traffic is collapsing. Reddit’s complaint is different because they have a $60 million annual contract for AI training rights, and the fight is over whether that price still reflects what Google is getting.

On top of that contract, Google also crawls reddit.com through a Googlebot that indexes pages so Reddit threads show up in search results. Reddit’s compensation for that second arrangement was always informal, with only the referral traffic from people clicking those results back to the site.

That worked for 20 years, and then Google started putting AI generated summaries at the top of the search page and the clicks stopped coming back at the same rate.

What Crawling Means

There are two separate crawls running between Google and Reddit and WSJ only talks about one of them.

1. The training crawl, which gives Google permission to use Reddit data for building AI models.

2. The search crawl is separate. Googlebot uses it to index pages for regular search results, and AI Overviews run on it too. Blocking it means your site vanishes from Google completely.

Even without the training contract, Google's AI summaries would still pull Reddit content through the search crawl. The only way to cut that off would be blocking Googlebot entirely and disappearing from Google search, which nobody in the article has done.

Pew Research went through almost 69 thousand Google searches in July 2025.

With an AI summary present, 8% of users clicked through to a regular result. Without one, 15% did. The traffic that Google used to send back to Reddit has been falling since AI Overviews rolled out, and the $60 million price tag was locked in before anyone realized how much would disappear.

Read more

Back in March, our co-founder Michael flew out to Nvidia GTC, asked Jensen Huang about his $2 billion in optics investments, and sat down with Gilad Shatner, Nvidia’s SVP of Networking, to talk about the shift from copper to CPO. That trip produced our biggest article of the year and a Youtube video where Michael called Soitec his number one stock pick on camera. The stock ran 200% after that video went live.

Soitec became more than just a trade idea for us. The March deep-dive helped us build this newsletter and put Asymmetrical Bets on the map with the photonics community. A lot of you reading this are holding this stock or held it at some point because of that piece.

The stock has pulled back hard from its peak at €200.50, and trades at €117 today. But that €117 is still well above the ~€55 where we first called it and now Soitec just reported Q1 fiscal 2027, and everything Michael laid out in March is showing up directly in the print.

Revenue hit €113 million, up 23% YoY at constant currency and scope, well above the company’s own guidance of roughly 15% and Shares gained almost 22% on the day.

What moved the stock the most was Photonics-SOI. Management’s language on yesterday’s call tracks almost exactly what we described in March. CEO Laurent Rémont broke Photonics growth into three waves.

1. The first was telecom, going back roughly a decade.

2. The current wave is scale-out in the data center.

3. The next wave, still building, is scale-up, moving from pluggable transceivers toward CPO.

We also flagged Soitec’s CPO content multiplier in March, the idea that a CPO module consumes several times more Photonics-SOI wafer area than a standard pluggable transceiver. Rémont didn’t put a number on that this quarter, but he confirmed the mechanism. CPO adoption in scale-up is coming, and Soitec sells into pluggable, NPO, and CPO architectures without much preference for which one wins.

Quarter Review & Highlights

Read more

When we published our deep dive piece a few weeks ago, the standing objection to this company was always the same: beautiful margins, but no growth.

Our argument was that the growth was about to kick in.

• A factory that had just doubled

• A capex bill about to vanish

• A defense business coiled off a near zero base

• A growth incentivized engineer taking the wheel

• The spintronics market seeing explosive growth

On July 22, NVE reported the first quarter with that new capacity fully in service.

→Revenue $11.03 million, up 81%

→ Net income $6.39 million, up 79%

→ EPS of $1.32 versus $0.74, up 78%

We’re not here just to toot our own horn though, we believe NVEC is actually much cheaper than it was before we published our first article on it. Our new price targets are as follows:

Read more