Behind the capital letters sits something considerably smaller: an effective 55% share of the output of a joint venture with a private operator.

No text has been published. Interim president Delcy Rodríguez said on Saturday the agreement runs for 25 years, a long way short of the century-long lease that pre-deal reporting had described.

Specialists note there is no precedent for Washington leasing and operating Venezuelan fields, and that the structure may collide with the country’s constitution and its hydrocarbons law.

Meanwhile, none of it matters to the tape below.

An energy shock wearing an inflation costume

Read July’s inflation split and the diagnosis writes itself. As shown in the table above, headline inflation ran at 3.4%. Strip out food and energy and it ran at 2.5%. Energy prices rose 14.7% over twelve months, gasoline 24.6%.

America does not have a broad inflation problem. It has a Strait of Hormuz problem that arrives at the pump, where a gallon costs $4.09 against $3.20 a year ago. Transit through the strait has collapsed from 18 million barrels a day before the war to 4.8 million in July and around 2 million in August.

Caracas cannot fill that hole. Not this year, not next.

Venezuela pumps 1.23 million barrels a day, its best rate since February 2019, and the resource in question is heavy, sour crude that needs upgraders and diluent before a gasoline refinery will accept it. Every serious estimate lands in the same place.

This is a 2029 story wearing a 2026 headline. Read the estimates below, then read the last row twice.

The Fed is the transmission mechanism

Kevin Warsh used Jackson Hole on Friday to say the better summer prints do not tell him underlying trends have meaningfully improved.

Odds of a September hike jumped from 39.9% to 57.0% in a single afternoon.

The two-year Treasury yield tore 12.8 basis points higher to 4.356%, its largest one-day move since March and the biggest reaction to a Jackson Hole address by a Fed chair since Alan Greenspan in 1996.

Step back from Friday and the picture is broader. Treasury yields have risen between 36 and 106 basis points across the yield curve this year, with the short end moving fastest, yet the episode that genuinely unsettled allocators was the abrupt jump in long-dated sovereign yields earlier this quarter, a fiscal signal rather than a monetary one.

The result is that duration has been repriced twice, for two entirely different reasons, and investors have been treating both as if they were the same trade.

Share

Now add the calendar.

History is unkind to the hawkish case: over the past two decades the Federal Reserve has shown little appetite for raising rates in an election year, and this November brings a midterm. At 57%, markets are pricing a coin flip. And given that ninety four of 104 economists surveyed still expect no move this year, somebody is going to be badly wrong.

Then there is the part of Warsh’s speech the tape ignored while portfolios were being rebalanced into the close:

Read that alongside his remark that the potential for substantially higher growth is on the rise.

This is not a chair building a case for choking off an expansion. This is a chair who sees productivity, artificial intelligence included, doing structural work on the supply side of the economy. The hawkish headline was priced within minutes. The growth message was not priced at all.

Subscribe now

Why this is a growth story, not an energy story

Here is the chain that matters. Hormuz sets crude. Crude sets gasoline. Gasoline sets headline inflation. Headline inflation sets the front end of the curve. And the front end sets the multiple investors will pay for artificial intelligence infrastructure earnings in 2032.

Long-duration equities sit at the end of that chain and they are geared to every link in it. Most of their value lives in cash flows a decade out, so each basis point added to the discount rate strikes their present values hardest.

Which makes last week’s sequence so instructive. Nvidia delivered a quarter that should have ended the debate: revenue of $96.2 billion, up 106% year on year, earnings of $2.22 a share against $2.10 expected, data centre revenue of $89 billion, up 117%, and guidance above consensus with Vera Rubin described as the fastest product ramp in the company’s history. The stock rose 8.7% on Thursday, its best session since April 2025, adding roughly $440 billion of market value.

Then Warsh spoke, and it gave back 4.6% on Friday.

Nothing about the AI supercycle changed between Thursday and Friday. Only the discount rate did. That is the whole argument in two sessions.

And it cuts both ways. A discount rate that compressed a supercycle in one session can just as easily hand the multiple back, which is precisely what the market is refusing to price today.

Everything above is the diagnosis, and it is free. What follows is the part that pays: the names geared to the reversal.

Below is a table of some of the major beneficiaries of the AI supercycle, with the reason each one trades on the front end of the curve rather than on a barrel of crude.

Join The Disruptive Investor now and read on. The re-rating will not wait for consensus.

Read more

That is the fourth consecutive quarter of accelerating year-on-year growth, at a scale where deceleration is supposed to be a law of physics.

The details are where it gets serious.

Data center revenue hit 89 billion dollars, up 117% year over year. Gross margin held at 75%. Non-GAAP earnings came in at 2.22 dollars against 2.09 expected, the fifteenth consecutive beat.

Then the guidance, which is the number that actually matters. Nvidia guided the October quarter to 108 billion dollars, four billion above the Street, the first hundred-billion-dollar quarter in the history of the company. And it assumes zero data center compute revenue from China.

Now the sentence from the call that should reframe how you look at every supplier in this chain:

“We have supply for 70% growth. Our demand is much higher than that.”

Share

Read it twice. The constraint is not demand. The constraint is capacity.

Huang was explicit that agentic AI consumes 15 to 100 times more compute than a human typing prompts, that sovereign AI, neoclouds and enterprises now represent roughly half the business and are growing at 100% a year, and that a year ago a single lab was driving this buildout. That concentration is gone. The demand base has broadened.

Here is the read-through nobody should miss. If the bottleneck is physical capacity rather than appetite, the value migrates to whatever is scarce: memory, optical bandwidth, power, materials and test.

You cannot ship an accelerator without high-bandwidth memory. You cannot cluster it without optics. You cannot switch it on without electricity. Nvidia just told you that the entire chain is sold out for the next eighteen months.

Meanwhile, that same chain spent July and August being liquidated by a hedge fund unwind (Situational Awereness) and thin-market de-grossing, while its earnings estimates went straight up through both episodes. Prices marked down. Estimates marked up. Capacity sold out.

The Disruptive Investor holds fourteen positions across every scarce layer of this buildout, each with a five-year price target in triple-digit percentage terms. Paid subscribers get the full book, the framework and the reasoning behind every target.

UPGRADE TO ACCESS ALL FOURTEEN NAMES

Subscribe now

Not investment advice. For informational purposes only. Read the full Disclaimer here.

Read more

Bloom Energy trades at around 213 dollars today, roughly 40% below its June high near 350. Over the same stretch, the analyst consensus for what this company will earn in 2028 has gone up, not down, and it now sits at its highest level on record.

Both statements are true. Only one of them describes the business.

FIRST, THE PRICE. IT HAS BEEN A VIOLENT YEAR IN BOTH DIRECTIONS.

Look at what this chart actually contains, because the eye goes straight to the drawdown on the right and misses the shape of the whole thing.

The name went from under 30 dollars a year ago to 350 in July. That is the move that matters. Everything since then is a correction inside a trend, not a reversal of it. The February to April consolidation between 176 and 116, then the April breakout, then the July blow-off top, then two months of giving back gains. Anyone who bought the breakout in April is still comfortably ahead. Anyone who bought the July top is down 40%.

That is what a high-beta position in a structurally growing market looks like. It is not a malfunction. It is the toll you pay on the road to a triple-digit return, and the toll is always collected at the least convenient moment.

The question is what happened underneath the price while all that was going on.

NOW THE PART NOBODY PUT ON A SCREEN THIS SUMMER.

Three lines, three fiscal years, and one direction.

The 2028 estimate has gone from roughly 1 dollar in early 2025 to about 6.20 dollars today. The 2027 estimate has gone from under 1 dollar to nearly 3.90. The 2026 estimate has roughly quadrupled from where it started.

Every one of those lines has made a fresh high in the past few weeks, including after the July peak in the share price, including through the August selloff.

Read the sequencing carefully, because it is the entire point. The share price topped in July. The estimates did not. They stepped up again in May, and again this month. Not a single one of the three has rolled over.

There is a specific reason this happens with Bloom Energy and it is worth stating plainly.

Compute is constrained by electricity long before it is constrained by silicon. Every incremental gigawatt of AI data center capacity has to be powered by something, grid interconnection queues in the United States now run for years, and on-site generation stops being an option and becomes the only path.

The order book reflects that. The estimates reflect the order book. The share price reflected, this summer, a thin August market reducing exposure because the calendar said so.

WHICH BRINGS US TO VALUATION, AND THIS IS THE CHART THAT SHOULD MAKE YOU SIT UP.

Price on top. Forward twelve-month price to earnings multiple on the bottom. Watch the bottom panel.

Bloom Energy traded at a forward multiple near 190 times in November 2025 and again in January 2026. Today it trades at 57.85 times. That is a de-rating of roughly 70% in the multiple, and it puts the stock close to the lowest valuation of its entire recent history, back near the floor it last visited in 2024 when this business was a fraction of the size it is now.

Understand what just happened mechanically. The price fell while earnings expectations rose. Multiples are a ratio, and both sides of that ratio moved in the direction that compresses it. You are not being asked to pay up for optimism here. Specifically, for a company that is going to see EPS average annual growth north of 76% between 2026 and 2028, a price earnings ratio of 57.85 is not expensive.

You are being offered the same growth trajectory at a third of the price of admission that the market demanded nine months ago.

I have no interest in claiming this is cheap in absolute terms. Fifty-eight times forward earnings is not cheap. But it is the cheapest this name has been relative to its own growth in three years, at the exact moment its earnings visibility is the best it has ever been. That distinction is where returns come from.

SO WHAT REVERSES IT.

Two catalysts, and both land inside the next four weeks.

Nvidia reports tomorrow evening, 26 August, after the close. This is the single most important data point in the entire complex, not because Bloom Energy sells to Nvidia, but because the capital expenditure narrative for every supplier below it either gets confirmed or gets broken in that call.

Watch the data center commentary and the supply constraint language, not the headline beat. If the buildout guidance holds, the power layer is mechanically the next thing repriced, because you cannot deploy the accelerators without solving the electricity first.

Then September, which brings a dense technology conference calendar. These events matter more than they should, and for a simple reason: they put management teams in front of investors with fresh order books at a moment when positioning is clean, leverage has been flushed and the crowding has thinned.

A trigger does not need to be a surprise. It only needs to give a market that has already de-rated a reason to look again.

THE PATTERN IS NOT UNIQUE TO BLOOM ENERGY. THAT IS THE REAL POINT.

What you have just read is not a single-stock story. It is a template, and it is currently repeating across the AI infrastructure complex.

The sequence began in July, when Leopold Aschenbrenner’s Situational Awareness fund came apart. A concentrated, levered book in exactly these names had to be liquidated into whatever bid existed, and the mechanics of forced deleveraging do not distinguish between a good business and a bad one. They sell what they hold, in the size they hold it, at the speed the margin desk demands. August then finished the job, with thin desks reducing gross exposure because the calendar said so rather than because anything had changed.

So the template reads the same across the complex: prices marked down by two consecutive rounds of mechanical selling, estimates revised up straight through both of them, multiples compressed to the low end of their own range, and catalysts stacked into the next four weeks. Not one earnings estimate was cut in either episode. That is the tell.

Which raises the only question that matters from here: where in the chain is the divergence widest, and which names have the most ground to recover when the multiple re-rates rather than compresses. The Disruptive Investor has run that work name by name.

Paid subscribers get the full book, all fourteen names below.

Not investment advice. For informational purposes only. Read the full Disclaimer here.

Fourteen positions. Fourteen price targets, all of them triple-digit. Same divergence, same setup, same calendar.

Read more

It is not a cynical question. It is a structural one.

Banks are paid for issuance, for volume, for financing, and for access to management teams, and each of those revenue lines quietly prefers a particular conclusion. That does not make the analysis wrong. It makes the distribution of errors asymmetric, tilted in a direction the reader is never shown. So a bank note begins in my hands with a presumption against it, and it has to argue its way out.

Last week, Goldman’s trading desk published something that argues its way out.

It does so for a simple reason: the argument is not a recommendation, it is a mechanic.

It describes how capital is actually being reallocated inside the AI theme, with observable measurements behind each claim, and it points toward the unloved parts of the complex rather than the crowded ones. That is the opposite of a note written to move product.

So for once, I am going to take it seriously and explain it properly. Here is the shift, from first principles, and what it does to a portfolio built on this theme.

FIRST, WHAT ACTUALLY HAPPENED LAST WEEK. IT WAS PLUMBING, NOT PROPHECY.

August is the month when trading desks are half-staffed, liquidity is thin, and risk managers reduce exposure rather than defend it. The industry term is de-grossing: cutting both longs and shorts to shrink the book, regardless of what you think of the underlying companies.

Understand what that means, because it matters.

De-grossing is not selling because the thesis broke. It is selling because the calendar says so. In a thin market, the same volume of selling moves prices two or three times further than it would in October. That is why the magnitude of the moves shocked a market that had already written off the late-July washout as ancient history.

NOW THE PART THAT ACTUALLY MATTERS: THE QUESTION IS CHANGING.

For three years, owning AI meant owning the buildout. Who sells the chips, the memory, the optics, the power, the racks. Capital expenditure was the scoreboard, and every upward revision to a hyperscaler budget lifted the entire supply chain.

Goldman’s read is that the market is now transitioning toward a second question: who wins from mass AI adoption, as the cost of compute deflates.

Take the logic slowly, because it is not intuitive. Every year, a given quantity of AI compute gets cheaper. That is what falling cost curves do.

For the companies selling the shovels, deflating unit cost is a headwind that volume growth has to keep outrunning. For the companies using AI to transform an existing business, deflating unit cost is pure margin. The cheaper intelligence gets, the more of it they consume, and the more value they capture.

So the same trend that built the first phase of this trade eventually starts funding the second phase. Not immediately. Not cleanly. But directionally, and the market is starting to price it.

THIS DOES NOT MEAN SELL THE BUILDOUT. IT MEANS OWN IT DIFFERENTLY.

Goldman is explicit that there is still value in owning the buildout beneficiaries, and I agree. What changes is the approach: more tactical, more attentive to entry points, less willing to pay any price for exposure.

Their preferred tool is the one long-term readers of the Disruptive Investor will recognize immediately. Trade the gaps between price and EPS.

The concept is simple and it is the most reliable thing in equity investing. A company’s share price and its forward earnings estimates travel together over any meaningful horizon.

In the short run they separate, sometimes dramatically, because price responds to sentiment, positioning, liquidity and headlines, while earnings estimates respond only to business reality. When the two lines separate, one of them is wrong, and it is almost never the earnings line.

A gap where price has fallen while estimates have risen is not a warning. It is a discount on a business that got better while its quote got worse.

Where does Goldman find the widest gaps in AI right now? Two places: memory and data centers. In both, aggregate market capitalization has lagged aggregate earnings estimates. And in memory, the divergence is now the widest of the entire cycle.

Memory Weighted Average Net Earnings Estimates vs Weighted Average Market Cap:

Source: Goldman Sachs

HOW TO READ THAT CHART, BECAUSE IT IS THE MOST IMPORTANT ONE IN THIS ARTICLE.

Both lines start at the same point in January 2025 and measure the same universe of memory companies. The black line is the weighted average of their net earnings estimates. The blue line is their weighted average market capitalisation. One tracks what the businesses are expected to earn. The other tracks what investors are willing to pay for them.

Until early 2026 they moved together, as they should. Then they parted company.

Earnings estimates for the sector are now about nine and a half times their January 2025 level. Market capitalisation is roughly five and a half times that same starting point, having peaked near seven times in July before rolling over. The profit expectations of these companies have compounded at nearly twice the pace of their share prices, and the gap has been widening rather than closing since April.

Read that once more, slowly. The black line has not paused, not flinched, not rolled over. It went up through the July hedge fund liquidation, up through the Hormuz shock, and up through this week’s de-grossing. The blue line did all the flinching.

There is no third interpretation available. Either the earnings estimates are collectively wrong, in a sector where pricing and volumes are contracted and visible quarters ahead, or the share prices have work to do. The market is not telling you that memory is broken. It is telling you that memory is cheap.

MOMENTUM IS QUIETLY REBUILDING ITSELF, AND THIS IS THE MOST UNDER-DISCUSSED SIGNAL OF THE SUMMER.

Before going further, a word on why I pay unusual attention to this particular factor.

Momentum is one of the four predictive factors the Disruptive Investor on Substack is built on, and it is not a preference, it is a conclusion. Since Jegadeesh and Titman first documented it in the Journal of Finance in 1993, momentum has been tested across virtually every equity market, every asset class and every time period researchers could obtain data for, and it has survived.

Three decades of academic literature have tried to explain it away as a data artifact, a risk premium, or a survivorship illusion, and it has outlived every attempt. It remains the single most robustly documented source of predictive excess return in the empirical finance canon.

Which is precisely why a structural change in its composition is worth reading closely rather than ignoring. When one of your four engines starts turning over internally, you want to know what it is now pointing at.

Momentum, as a factor, simply means owning what has been going up and avoiding what has been going down. For three years, momentum and AI were effectively the same trade. Goldman’s high-beta momentum basket fell 12% in the recent unwind, and their AI versus AI-at-risk pair dropped about 10% over five sessions. Same trade, same drawdown.

But here is the mechanical detail that tells you where we are in the cycle. Momentum has posted more single-day declines greater than 5% in 2026 than in the previous five years combined. And critically, its internal composition is turning over.

Goldman measures this by overlap. The overlap between the twelve-month winners and the three-month winners has collapsed to multi-year lows. The overlap between the twelve-month winners and the three-month losers is near its highs. Read that again: the stocks that led for a year are now, mechanically, among the recent laggards.

The consequence is that software has become the largest weight in the long leg of three-month momentum, while the semiconductor and AI complex has migrated into the short leg. Systematic strategies that mechanically buy the long leg and sell the short leg are therefore now selling the buildout and buying the applications. Not because of an opinion. Because of an arithmetic that updates every month.

Go back to the chart above and you can see this happening in real time. That blue line rolling over from its July peak, while the black line keeps climbing, is exactly what factor rotation looks like when you plot it.

WHY THIS IS AN OPPORTUNITY RATHER THAN A THREAT.

Two reasons.

• First, forced selling has no relationship to value. When a systematic strategy moves a name into its short leg, it sells regardless of the backlog, regardless of the guidance, regardless of the earnings revision trajectory. That is exactly how price versus EPS gaps get created. Somebody has to be selling a good business for a bad reason. Factor rotation is that somebody.

• Second, positioning has been cleaned out. The leverage is gone, the crowding has thinned, and the market ex-AI is now strongly negatively correlated with AI itself. Goldman’s own conclusion is that this produces a healthier setup for re-risking into September than we have had in a long while.

THE CALENDAR FROM HERE IS DENSE.

Nvidia reports second-quarter fiscal 2027 results on Wednesday, August 26th, after the close, with consensus around 92 billion dollars of revenue and 2.09 dollars of earnings per share. It remains the single most important data point in the complex, because it either confirms or breaks the capital expenditure narrative for every supplier below it.

Beyond that, September brings a heavy technology conference calendar, which historically supports the theme by putting management teams in front of investors with fresh order books.

WHAT WE ARE DOING WITH ALL OF THIS.

Nothing dramatic, because the framework already told us where to be.

We own the buildout where the price versus EPS gap is widest, which is precisely where Goldman’s own screens point: memory and the physical infrastructure of the data center. We are not paying up for the parts of the complex where price has outrun earnings, and we are using factor-driven selling as an entry mechanism rather than an exit signal.

The specific positions, the memory names at the center of the thesis, and where the framework puts each of them, are below for paid subscribers.

Fourteen positions. The two memory makers doing the heaviest lifting are named in full, along with every optical, power and infrastructure holding behind them.

Read more

Yesterday, AI-infrastructure names fell 10 to 15% in a single session. Over the two weeks that led directly into that session, several of the same names had run up 40 to 50%. Neither move had anything to do with earnings.

START WITH THE ONLY CHART THAT SETTLES THIS ARGUMENT.

The chart above plots the S&P 500 price against forward 12-month EPS over ten years, from FactSet.

Look at it once and the conclusion is inescapable.

The two lines are the same line. Price wanders above and below in the short run, sometimes violently, and it converges on earnings every single time. February 2020, the 2022 derating, the March 2025 air pocket: each divergence closed, and it closed in the direction the EPS line was already pointing.

What is true for the index is true for the single name. Earnings drive prices. Everything else is timing. A geopolitical headline can move a quote for a session or a fortnight. It cannot move the earnings line, and the earnings line is what the quote eventually has to obey.

Hold that thought, because it is the entire answer to yesterday.

Now what actually happened. The Nasdaq closed down 1.3%. The Philadelphia Semiconductor Index dropped close to 5%, its worst session in weeks. Beneath the index, the damage was concentrated exactly where the AI buildout lives: memory, storage, optical bandwidth, and the power layer feeding the data centers.

Applied Optoelectronics lost 15% on the day. Bloom Energy lost 10%. Both sat squarely in the line of fire, as they always do when the sector is repriced wholesale rather than judged name by name.

NOW PUT THOSE TWO NUMBERS BACK IN THEIR FRAME.

Year-to-date, Applied Optoelectronics remains up +277%. Bloom Energy remains up +141%. A single session costing 15 and 10 points respectively does not dent a move of that magnitude. It barely registers against it.

Share

And the runway ahead is what makes yesterday genuinely anecdotal. Here is where the framework stands on the three names every subscriber to The Disruptive Investor can see:

Against a triple-digit road still to be travelled, a 15-point and a 10-point session are not a thesis event. They are noise inside a trend that is still doing exactly what the framework said it would do.

Subscribe now

THE CAUSE WAS NOT THE INDUSTRY. IT WAS THE BOND MARKET.

The 30-year Treasury yield touched 5.337% intraday, its highest level since 2007, nineteen years. The trigger was geopolitical, not economic: the 60-day US-Iran truce expired without renewal, Iran fired missiles toward the Strait of Hormuz, and air defenses were triggered in the UAE. Oil spiked. Inflation expectations spiked with it. Long-duration growth equity, AI infrastructure included, took the discount-rate hit.

Nothing in that sequence touched a single earnings estimate. Go back to the chart. The EPS line did not move yesterday. Only the price line did.

This is worth saying plainly: the AI industry’s fundamentals remain excellent. The two weeks of gains that preceded the drop were not a fluke. They were the market pricing in an exceptional reporting season and guidance that, across the sector, moved higher rather than lower.

That is what produced positions up 40 to 50% across the 14 names covered by The Disruptive Investor going into yesterday. That reality did not change overnight. A missile over the Strait of Hormuz does not reduce hyperscaler capital expenditure, does not shrink a backlog, and does not revise a single data-center buildout.

Subscribe now

WE HAVE SEEN THIS FILM BEFORE, JUST A MONTH AGO.

In July, the AI trade suffered its sharpest correction of the year when Leopold Aschenbrenner’s hedge fund, Situational Awareness, was forced to unwind a heavily leveraged book. Its concentrated AI-infrastructure positions fell between 35% and 78% from their peaks in a matter of weeks.

Bloom Energy, one of the fund’s largest holdings, lost roughly a third of its value over the month for reasons that had nothing to do with the demand for its product. It is up +141% on the year regardless. At the time it felt existential. It was, in fact, a liquidation event, not a verdict on the AI-infrastructure buildout.

Once the forced selling cleared, the same names rebounded sharply, and the eventual consensus read on the episode was unambiguous: a technical dislocation, not a deterioration in the industry’s fundamentals.

Yesterday is the same pattern with a different trigger. Technical in July. Geopolitical and rate-driven yesterday. The common denominator is that the shock originated outside the industry’s own numbers, and every prior instance of that this has faded once the specific tension that caused it cooled.

THE RULE, STATED ONCE, APPLIED CONSISTENTLY.

Geopolitics and macro do not override a thesis built on fundamentals. They override it only when the shock itself shows a credible path to persist: a closed Strait, a structural repricing of the cost of capital, a genuine change in demand.

That is not today’s setup. The Strait remains open to traffic even as rhetoric escalates, and a yield spike driven by a war-risk premium rather than a change in the growth or inflation outlook tends to mean revert once the premium is repriced.

THIS IS THE COST OF ADMISSION, NOT A REASON TO LEAVE.

A portfolio built for triple-digit annual returns will never travel in a straight line. Sessions like yesterday are not the exception to that path. They are the toll paid for it. +277% and +141% were never going to arrive without days that take back 15 and 10, and neither will the +183.1% and +151.7% still on the table.

The trajectory that produced 40 to 50% gains across those two weeks is the same trajectory that just gave back 10 to 15% in one session, and it is the same trajectory expected to resume shortly, once markets finish repricing a risk that has not, so far, materialized into anything lasting.

Three names above. Eleven below. The remaining eleven positions, with entry prices and full targets, are for paid subscribers.

PAID SUBSCRIBERS ONLY. THE OTHER ELEVEN NAMES AND WHERE THE FRAMEWORK PUTS THEM.

Read more

A wave of paid subscriptions arrived immediately following substantial advances across positions available to free subscribers. A handful of cancellations followed the correction that came next.

Read that sequence again, because it describes with uncomfortable precision the single most costly behavioural pattern in all of investing: buying strength and selling weakness.

Entering after the move has already happened. Exiting when the discomfort peaks. Doing, in short, the exact opposite of what the evidence supports.

This is worth addressing directly, with numbers.

WHAT ACTUALLY DRIVES OUTPERFORMANCE

The approach behind The Disruptive Investor rests on a body of academic research that has been replicated, stress-tested, and published across four decades.

David Blitz and his quantitative research team at Robeco have catalogued in 2024 in the renowned Journal of Portfolio Management the factors that predict outperformance across markets and time periods.

Their work identifies 15 distinct factors with persistent explanatory power. Among them:

• VALUE, the discipline Warren Buffett built a career upon, formalised in the academic literature by Fama and French

• MOMENTUM, documented by Jegadeesh and Titman in 1993, extended by Carhart into the four factor model that remains the industry standard

• QUALITY, measured through profitability and investment discipline, added by Fama and French in their five factor framework

These are established, peer-reviewed, and independently verified. The work of The Disruptive Investor consists of blending some of the 15 factors in different weightings and calibrating that mix to the prevailing economic regime.

The same methodology governs the “Great Visionaries” fund I manage for qualified investors in Switzerland. Year-to-date performance: +45.5%.

Subscribe now

THE PRICE OF ADMISSION

The two charts below show what this approach produces pro forma and, more importantly, what it costs along the way for impatient investors.

Chart 1: The long run, since 2009

A twenty-stock portfolio built on this factor framework:

The end-result? Six times the cumulative return of the S&P 500 index. And, as shown in the table below a 59.48% peak-to-trough decline at the worst point.

Chart 2: Regime-calibrated factor weighting

The same twenty-stock portfolio approach, with factor weights recalibrated to the current economic regime and rebalanced every two weeks:

A NUMBER THAT INVITES SKEPTICISM, AND SHOULD

NINETY PERCENT PER ANNUM. Top right of the table above. To many readers, that reads less like a performance statistic and more like a typographical error.

That reaction is reasonable. It is also the correct reaction to have toward any number of that magnitude presented without context.

Context matters here. The figure is the result of a historical backtest, applied to a twenty-stock, factor-based construction, rebalanced on a fixed two-week schedule.

What matters is the origin of the factors themselves. Value, momentum, and quality were documented decades before this test was ever run:

• Fama and French established the framework.

• Jegadeesh and Titman identified momentum in 1993, extended by Carhart.

• Blitz and the Robeco quantitative team catalogued fifteen such factors using markets and periods entirely separate from this one.

The number also carries a visible cost, and that cost is disclosed rather than hidden.

A standard deviation of 54.77% and a maximum drawdown of 44.04% sit directly beside the return figure, in the same table above. A backtest this strong, without volatility of that magnitude attached to it, would be the one worth questioning.

Which, stated differently, means the investor who sells into the drawdown is the one paying for someone else's return.

Share

THE POINT THAT MATTERS

Look closely at the right-hand side of both charts above. The recent decline is severe. It is also entirely consistent with the historical pattern.

• A strategy delivering a 26.93% annualized return over sixteen years traveled through a 59.48% drawdown to get there.

• A strategy delivering 90.17% annualized carried 54.77% volatility as the price of that outcome.

Those two facts are inseparable. The return exists because the volatility exists. Investors who require the former while rejecting the latter are asking the market for something it has never provided.

The Sharpe ratio of 1.15 in the regime-calibrated version tells the story cleanly: the compensation per unit of risk exceeds that of the index. The volatility is real, and it is paid for.

WHY THIS MOMENT DESERVES ATTENTION

The recent correction across AI infrastructure names has been sharp. It has also arrived at a point where the underlying spending cycle is accelerating rather than maturing.

Worth stating plainly: AI infrastructure is not a sector selected by preference within The Disruptive Investor. It is the current output of the quantitative process, the sector where the factor scores presently concentrate given the prevailing economic regime. The same process pointed toward different sectors in different regimes before, and will again. A shift in the macro backdrop could just as easily rotate the highest scoring names toward biotech, pharmaceuticals, or utilities. The discipline is the constant. The sector is simply wherever the process currently leads it.

And right now, the process leads directly to the spending decisions of the four companies writing the largest checks in the industry.

As extensively covered in prior articles of The Disruptive Investor, Alphabet, Microsoft, Meta, and Amazon have each raised capital expenditure guidance. The buildout is expanding in scale, and the demand curve for compute, optical interconnect, power infrastructure, and test equipment continues to steepen.

This rally remains in its early innings. Which means the most rational moment to establish positions is precisely the moment that feels least comfortable: now, after the drawdown, rather than after the next advance.

That is the entire lesson of the two charts above, expressed in a single sentence.

AND WHILE THIS ARTICLE WAS BEING WRITTEN

Consider what happened over the last two trading sessions.

Applied Optoelectronics, a position released to free subscribers on April 6, closed up 9.2% on Friday and now shows +26.2% since entry, with a price target sitting +174.3% above current levels.

That one was given away to all subscribers at no cost.

The most recent recommendation reserved for paid subscribers, the fourteenth position in the portfolio, gained 6% on Friday. Its price target stands +116.9% above where it trades right now, and this remains among the most conservative targets in the entire book.

Here is the part that deserves a moment of reflection.

Every reader on the free list saw the Applied Optoelectronics move. The eleven positions carrying the highest conviction, the deepest quantitative scoring, and the widest gaps to fair value sit behind the paywall. The median distance to the 5-year price target across that group exceeds +244%.

The table below shows exactly where those numbers stand today. Every ticker, every entry level, every target unlocks in full for paid subscribers immediately after the paywall. Nothing in it stays blurred.

The correction of recent weeks has done something specific to those numbers: it has widened them. Lower entry prices with unchanged targets produce larger gaps. That is arithmetic, and it favours whoever acts while the discount exists.

The complete portfolio follows below. Every ticker, every entry level, every target, and the reasoning behind each.

Upgrade to paid and the full table opens immediately.

Read more

If you've been waiting for an entry point in AI optics, this is it.

The number

Alphabet: $195-205B capex for 2026, 2027 “increasing significantly.”
Microsoft: $175B, with a single quarter above $50B.
Meta: $130-145B, low-end raised.
Amazon: $220B, up from $200B.

Total: $720-745 billion in one year, from four companies. Every one guiding 2027 higher.

And here’s the part nobody is pricing: three of the four are now at NEGATIVE free cash flow. Meta down 91% to $784M. Alphabet at minus $5.9B. Amazon at minus $7.6B.

The most profitable companies on earth are burning cash to build AI capacity they’ve already sold.

Microsoft CFO Amy Hood: “AI demand continues to exceed available supply.”
Meta CEO Mark Zuckerberg: “There’s just nowhere near enough compute for all the demand.”
Meta CFO Susan Li: “The industry has under-built historically.”

This is not a cycle rolling over. It is a cycle compounding.

Applied Optoelectronics: the bottleneck finally pays

AAOI ($124.22, Nasdaq) reported last night. Straight to it:

Revenue $191.9M, +86% y/y, a fifth consecutive record quarter. Adjusted EPS $0.06, beating by $0.05, and back in the black. Data center revenue +140% y/y, now 56% of the business. 800G revenue up more than TENFOLD year-over-year.

Guidance for Q3: $255-290M, +130% y/y, above consensus. 800G revenue to quintuple sequentially. Over $70M of 1.6-terabit revenue as early as Q4. Full year reiterated at ~$1.1 billion, +137%.

Why 800G matters in one sentence: in a cluster where tens of thousands of GPUs, like those of Nvidia, must talk constantly, the optical link is the bottleneck, and each generation doubles bandwidth per port. AOI makes its own laser chips in house, which is why it can ramp when everyone else is queueing for supply.

The stock had already run +38% in five sessions. This morning it’s up another +15%.

Its all-time high is $233.67. From yesterday’s close, that’s +88%.

Given a demand curve set by three-quarters of a trillion dollars of capex, reclaiming that high in the coming weeks would surprise nobody.

Subscribe now

And AOI is one position. This isn’t a one-day pop. The whole Disruptive Investor selection is up high double digits over the last five sessions, and it’s still climbing this morning.

Several positions still carry 200% to over 700% upside to target. After the rally.

🔒 The rest is for paid subscribers

Read more

Three weeks ago, Leopold Aschenbrenner’s Situational Awareness sat atop $45 billion in assets, riding the greatest hedge-fund run anyone had seen in a generation.

Then the physical-AI longs and the software shorts turned on him simultaneously.

The prime brokers came calling. And by July 30 the entire public book had been dumped on Citadel in a single block trade, assets cut to roughly $10 billion.

“We let you down this month,” Aschenbrenner wrote to his investors, disclosing a -67% month. Still net +80% for the year. Read those two numbers next to each other and you understand everything about what just happened.

Because that kind of forced deleveraging doesn’t just liquidate a portfolio.

It liquidates positioning: the crowded, levered, momentum-chasing positioning that made the AI infrastructure trade so violently reflexive on the way up, and so violently reflexive on the way down.

And positioning, once purged, tends to get rebuilt.

Wall Street’s strategists just said the quiet part out loud

Right on schedule, the sell-side chorus turned this weekend. Morgan Stanley’s Michael Wilson called the recent slide “one of the worst momentum selloffs in history”.

The arithmetic backs him. A Goldman Sachs basket of momentum stocks has fallen 35% from its June peak and is now up just 9.4% on the year, dead in line with the S&P 500. A mean-reversion setup if there ever was one. Wilson sees the index reaching 8,000 by year-end on improving earnings breadth. It closed Friday at 7,489.72.

His peers are converging on the same read. Goldman’s Ben Snider called the AI trade’s recent consolidation unremarkable by the standards of past momentum rallies, noting that investor deleveraging, the very thing Aschenbrenner’s fund was just marched through, typically precedes an improved outlook.

JPMorgan’s Mislav Matejka still expects broadening beyond tech in the second half, but flagged that with semiconductors near oversold levels and EPS momentum still rising, the cluster should stabilize from here.

Read that again. Deleveraging as a bullish signal. Oversold semis. Momentum resetting, not ending.

Subscribe now

Wilson, Snider and Matejka just handed that call three of the loudest microphones on Wall Street.

Share

The tape is already answering

I am writing this mid-session on Monday, and the market is not waiting for permission.

The names gutted hardest in the Aschenbrenner-adjacent rout are the names leading today’s tape: double-digit green across optical, power and specialty-semi names inside the first hours of trading.

Take Bloom Energy as the case study. BE bottomed near $157.33 during the July 27 to 29 leg of the rout, the very days Situational Awareness was fielding margin calls on its power and compute longs.

As of mid-session today, it trades around $224.00. That is a move of +42.6% off the low. In four sessions.

And here is the part that should stop you cold. Against my Disruptive Investor target of $526, Bloom is still sitting on more than +137% of remaining upside.

This is not a name that has had its move. This is a name that just got started.

Below is the full watchlist. Here is exactly what is free, what is teased, and what is unlocked.

Table 1: All Subscribers (Free)

Three names. Three double-digit recovery days. And the smallest upside on the list still carries +137% to target.

Now here is what the other ten names in the book look like right now. For the moment, blurred.

Table 2: Paid Subscribers

Look closely at row 6. One of the ten names above is sitting on more than 800% of distance to target, and it is trading under seven dollars today.

Everything below this line is for paid subscribers of The Disruptive Investor.

Read more

Eight days. Four earnings calls. And one message, repeated in every transcript: the AI infrastructure build is accelerating.

Alphabet: Capex of $195 to $205 billion for 2026. 2027 “increasing significantly.” Cloud revenue grew 82% to $24.8 billion. Free cash flow: negative $5.9 billion.

Microsoft: Capex of $175 billion for 2026, with Q1 FY2027 capex above $50 billion. Azure guided to 45% growth, accelerating from 40%. Commercial RPO: $627 billion, nearly double year-over-year. CFO Amy Hood: “AI demand continues to exceed available supply.”

Meta: Capex of $130 to $145 billion for 2026, lower end raised. FCF collapsed 91% year-over-year to $784 million. CEO Mark Zuckerberg: “There’s just nowhere near enough compute for all the demand.” CFO Susan Li: “The industry has under-built historically.”

Amazon: Capex of $220 billion, raised from $200 billion. AWS grew 37% to $42.2 billion, fastest in 18 quarters. AWS backlog: $496 billion, growing triple digits. TTM free cash flow: negative $7.6 billion. AWS AI services and custom silicon both crossed $25 billion in annual run rates.

Combined 2026 capex guidance: $720 to $745 billion. Nearly three-quarters of a trillion dollars, in a single year, from four companies. And 2027 is guided higher across the board.

➡️ This is not a cycle rolling over. It is a cycle compounding.

Subscribe

The Backlog and the Cash Flow

The capex numbers are staggering, but the backlogs tell the deeper story. Microsoft’s RPO is $627 billion. Amazon’s AWS backlog is $496 billion. Combined, over $1.1 trillion in contracted future revenue. These are not speculative projections. They are commitments at a scale that dwarfs the GDP of most countries.

Three of the four hyperscalers are now in or near negative free cash flow. Meta at $784 million, down 91%. Alphabet at minus $5.9 billion. Amazon at minus $7.6 billion.

They are ploughing nearly every dollar of operating cash flow into AI infrastructure they have already contracted to deliver. When the largest companies on earth accept negative FCF to build, the supply chain feeding that buildout does not stay cheap for long.

That is where the Disruptive Investor selection sits.

Share

The Rally

The Disruptive Investor selection responded immediately. Yesterday’s session produced broad double-digit gains, with multiple positions up 15 to 27%.

The “cycle peak” narrative that had crushed these names 40 to 70% from June peaks was demolished in 48 hours by four earnings calls.

One position surged +27% yesterday. As of writing, premarket indications point to another roughly +28% move due to the announcement last night of a multi-year revenue visibility, locked in, prepaid, and extending nearly six years forward.

Subscribe

The Setup

Four hyperscalers. $720 to $745 billion in 2026 capex. Over $1.1 trillion in combined backlogs. Demand exceeds supply across every single one.

The Disruptive Investor portfolio is positioned exactly where this demand flows: memory, optoelectronics, substrates, power delivery, and test equipment.

Several positions still carry triple-digit upside to their price targets, ranging from 200% to over 900%, despite yesterday’s rally.

The market is repricing the “cycle peak” narrative into what looks like a generational buying opportunity, and the repricing is still in its early innings.

When four hyperscalers commit three-quarters of a trillion dollars in a single year, the supply chain does not reprice in two sessions.

The catalysts are stacked. A major memory name reports next week. The midterm window opens in November with a 90.9% historical hit rate on the S&P 500. The supply chain is tightening, not loosening.

This is not the end of a cycle. It is the middle of an acceleration.

The full Disruptive Investor selection, with entry prices, current prices, and triple-digit target prices for each position, is available to paid subscribers. Upgrade to unlock real-time entry alerts and the analysis behind each position.

Subscribe

This note reflects the author’s personal research and assessment. It is published for informational and educational purposes only. It does not constitute investment advice, a personal recommendation, a solicitation to buy or sell any security, or a forecast of future price performance. Past performance is not indicative of future results. Forward-looking statements and valuation scenarios are illustrative model outputs based on publicly available consensus data and may differ materially from actual results. Readers must conduct their own due diligence and consider their personal financial situation, risk tolerance and applicable regulations before any investment decision. Read the full disclaimer here.

Last night delivered two earnings calls that could not have been more different in their reception, yet were identical in their message.

• Microsoft reported fiscal Q4 2026 and demolished expectations. Revenue of $90.0B against a $87.6B consensus. EPS of $4.74 versus $4.21 estimated. Azure growth guidance for Q1: 45% in constant currency, an acceleration from the 40% pace maintained for two consecutive quarters. The stock surged 8.9% in after-hours to $425.21.

• Meta reported Q2 2026 and the market balked. Revenue of $60.8B beat the $60.2B estimate, up 28% year-on-year. But EPS of $6.18 missed the $7.10 consensus, weighed down by a $2.4B legal charge. Free cash flow collapsed to $784 million. The stock dropped 7.5% in after-hours to $542.00.

One up 9%, the other down nearly 8%. Read the headlines and you would think these companies are on different planets. Read the transcripts and you realise they are reading from the same playbook.

The AI Capex Verdict

Here is what matters for the AI infrastructure thesis. Not the EPS beats or misses. The capital expenditure numbers and the language around them.

Microsoft:

• Q4 capex was $41 billion, including finance leases of $5.6B.

• Calendar year 2026 capex expectation: approximately $175 billion after a lease reclassification.

• FY 2027 capex: expected to grow year-over-year. CFO Amy Hood’s words: “We expect FY 2027 capital expenditures will grow year-over-year, given demand signals across our portfolio.”

• Q1 FY 2027 capex guidance: over $50 billion in a single quarter.

• Azure guidance for Q1: 45% constant currency growth, accelerating from 40%.

• And the line that could be tattooed on every AI bear’s forehead, from Hood: “Demand continues to exceed available supply.”

Meta:

• Q2 capex was $31.1 billion.

• 2026 capex guidance narrowed to $130 to $145 billion, with the lower end raised from $125B.

• Free cash flow: $784 million, down 91% from $8.55 billion a year ago and the lowest since 2022. Not a typo. Meta is pouring nearly every dollar of operating cash flow into AI infrastructure.

• Announced a partnership with BlackRock for a 1 GW data center in El Paso, Texas.

• CFO Susan Li: “The industry has under-built historically for the wave of AI adoption, making existing capacity, including our own, extremely valuable.”

• Mark Zuckerberg: “There’s just nowhere near enough compute for all the demand.” He added that Meta is receiving offers for its compute capacity “at a meaningful premium over what we paid.”

Combined: Microsoft and Meta alone spent $72.1 billion on capex in a single quarter. Add Alphabet’s $44.9B from last week, and you have three hyperscalers spending $117 billion in three months. Annualized, that is approaching half a trillion dollars flowing into AI infrastructure.

Is the AI Trade Alive?

The question on every desk since mid-July has been: is the AI capex supercycle peaking?

The answer from last night is unambiguous. No.

Microsoft is accelerating Azure growth from 40% to 45% and guiding capex above $50B per quarter. Meta is raising the floor on its $130 to $145B capex range and partnering with the world’s largest asset manager to fund a gigawatt-scale data center. Alphabet last week raised capex to $195 to $205B with 2027 set to “increase significantly.” Amazon reports tonight.

The AI infrastructure trade is alive. What was mislabeled as a peak is looking more like a correction within an acceleration.

What Happens Next To Your Portfolio

The following analysis is for paid subscribers.

Read more

Skepticism about the staying power of AI infrastructure spending keeps popping up in market conversations. Yet a striking data point just landed from the energy side of that story: electricity demand tied to AI isn’t cooling off, it’s building momentum.

Bloom Energy (BE, $166,84), a current pick within The Disruptive Investor, just posted arguably its strongest quarter on record, evidence that AI infrastructure spending is showing up concretely in order books, not just in headlines. Numbers for Q2 2026, released last night after the close, are sending the stock up 12.5% in pre-market trading this morning.

Q2 2026 snapshot:

• Quarterly sales hit an all-time high of $1.065 billion, up 165.5% from a year ago, well past the $827 million analysts had penciled in

• Earnings per share (EPS) landed at $0.78, nearly twice the $0.41 Wall Street was modeling

• Operating income flipped to a $182.2 million profit, a sharp reversal from an operating loss in the same period last year

• Operating cash flow came in at $226.4 million, compared with a $213.1 million cash outflow a year earlier

What really stood out, though, was how aggressively management raised its outlook for the rest of 2026:

• Full-year revenue guidance lifted to $3.9 to $4.2 billion, up from $3.4 to $3.8 billion previously, and ahead of the $3.74 billion consensus

• Full-year operating income guidance lifted to $800 to $900 million, up from $600 to $750 million

• Full-year EPS guidance lifted to $2.55 to $2.85, up from $1.85 to $2.25, and above the $2.17 consensus

Subscribe now

CEO KR Sridhar didn’t mince words on the call: demand keeps building quarter after quarter, and customers who once defaulted to combustion-based generators are now actively choosing Bloom Energy instead. Every major U.S. hyperscaler, plus more than a dozen neoclouds, AI labs and colocation operators, has now validated and approved Bloom Energy’s systems for AI facility power.

Share

Bloom Energy isn’t the only name in the Disruptive Investor’s coverage benefiting from this trend.

Just a day earlier, another company closely monitored inside The Disruptive Investor delivered a blowout quarter of its own, a business sitting right at the center of the AI data center buildout, with results that beat every Wall Street estimate by a wide margin and a contracted revenue backlog that just crossed a historic milestone.

This company recently signed a 15-year, $5.2 billion lease with an investment-grade hyperscaler, is in advanced talks for hundreds of additional megawatts of capacity, and just completed a major corporate spin-off designed to unlock shareholder value.

I break down the full numbers, what the guidance implies, and why I see this as one of the more underappreciated plays on AI power infrastructure, exclusively for paid subscribers below.

If you want the full picture, not just Bloom Energy but every high-conviction name tracked across the AI power and infrastructure theme, with real-time alerts the moment results move the needle, now is the moment to upgrade.

Paid subscribers get complete access to our stock-by-stock breakdowns, The Disruptive Investor portfolio, and earnings analysis like the one below.

Read more

It has been a brutal summer for the AI infrastructure complex.

SanDisk, the pure-play NAND memory standout that stunned the market with a 860% year-to-date gain, has collapsed -39% from its June peak of $2,354 to $1,436.56 as of last week’s close. Micron, its HBM counterpart, was up 320% and has shed -27%, falling from $1,255 to $920.95.

The VanEck Semiconductor ETF fell nearly 9% in a single week in mid-July. The Philadelphia Semiconductor Index dropped over 3% in one session after SK Hynix recorded its largest single-day decline on record. TSMC reported a 34% year-on-year jump in quarterly sales and it changed nothing.

The selling continued.

The trigger was not a single event but a convergence of doubts. Alphabet’s Q2 showed the first negative free cash flow in the company’s history as capex was raised to $205B. SK Hynix’s long-term HBM contracts locked it out of the spot price surge. The question on every desk became the same one: is the AI capex supercycle peaking?

The Catalyst This Week

This is where the hyperscalers’ earnings calendar intervenes.

• Microsoft reports fiscal Q4 2026 on Wednesday, July 29, after the close, with consensus EPS of $4.24 (16.2% y/y) on revenue of $87.7B (+14.8% y/y).

• Meta follows the same evening, with consensus EPS of $7.39 (+3.6% Y/Y) on revenue of $60.3B (26.9% Y/Y).

• Amazon closes the trifecta on Thursday, July 30.

All three have beaten estimates in each of the past seven consecutive quarters.

These are not just earnings. They are the verdict on whether AI infrastructure spending is accelerating or rolling over. Alphabet already gave us a preview last week: Google Cloud revenue surged 82% year-on-year, the cloud backlog reached $514B, and TPU (Google’s custom AI chips) revenue was recognized for the first time. The demand signal was unmistakable. What spooked the market was the spending, not the demand.

If Microsoft’s Azure growth holds at 40%, if Meta’s capex guidance stays disciplined, if Amazon’s AWS reaccelerates, the narrative flips. The meltdown gets reframed as a correction within a bull market, not the end of it. And that is when the second, more powerful force enters the picture.

The Pattern Nobody Talks About

Every four years, American voters go to the polls in November to cast ballots that the financial press treats as a risk event. Volatility rises. Hedging intensifies. Strategists issue cautious notes about political uncertainty.

And then, almost without fail, the market rips higher.

In the three-quarter window covering the midterm election quarter (Q4) and the two quarters that follow (Q1 and Q2 of the following year), the S&P 500 delivered positive returns 90.9% of the time. Not 60%. Not 70%. Nine times out of ten.

The average quarterly return during those three quarters was +7.68%, more than four times the +1.82% average for non-election quarters. When you compound all three quarters together, the gain reaches +24.9%.

To put that in perspective: in a typical non-election three-quarter stretch, the market rises roughly two-thirds of the time. In the post-midterm window, it rises nine times out of ten, and the average drawdown when it does fall is a modest -2.1%, compared to -6.6% in ordinary periods.

Not Presidential. Midterm.

Here is where it gets counterintuitive. You might assume the rally comes after presidential elections, when a new commander-in-chief is chosen and uncertainty is resolved. The data says otherwise.

After presidential elections, the S&P 500 was positive only 70.8% of the time across the same three-quarter window, with an average return of just +2.36% per quarter, barely above the non-election baseline. The compounded gain was +6.3%, less than a third of the midterm effect.

The authors are explicit: “The post-election surge is present after midterm elections and absent after presidential elections.”

Why? Because presidential elections do not fully resolve uncertainty. The new president arrives with an agenda, an opposition party mobilizes, and the policy direction remains contested. Midterms, by contrast, tend to produce divided government.

The voters hand one party the White House and the other party a chamber of Congress. The result is gridlock, and gridlock, it turns out, is the market’s preferred political state.

Not Policy. Not Luck.

A natural objection: perhaps the Federal Reserve eases around midterms, or fiscal spending accelerates, and that is what drives the rally.

Białkowski and Nahavandi tested this directly. They compared fiscal policy proxies (changes in federal spending and tax receipts) and monetary policy proxies (Fed discount rate changes, the federal funds premium) across midterm and non-midterm periods. The t-statistics showed no statistically significant difference. The policies were essentially identical.

In plain English: the midterm rally is not manufactured by the Fed or by fiscal stimulus. It is not compensation for higher volatility either; in fact, post-midterm volatility was lower (12.7% annualised) than in non-election periods (15.4%).

The Economic Policy Uncertainty Index runs notably higher around presidential elections than around midterms. Once the ballots are cast and the new balance of power is known, the uncertainty premium embedded in equity prices compresses, and the market rerates upward.

The Convergence

This is where the two threads meet.

The AI infrastructure trade has been decimated not because demand is faltering but because the market lost confidence in the spending trajectory.

Alphabet’s results last week were the first crack in that thesis: cloud revenue +82%, backlog $514B, capex raised to $205B. The demand is accelerating, not peaking. If Microsoft, Meta, and Amazon confirm the same pattern this week, the correction in semiconductors and memory gets reclassified from “cycle peak” to “buying opportunity.”

And that is precisely when the midterm tailwind arrives.

The S&P 500 currently sits at 7,412, about 2.6% below its 2026 high and roughly 17% above its February low of 6,344. With 100 days to go before the November midterms, the forecasters at Decision Desk HQ give Democrats a roughly 60% chance of retaking the House, while Republicans hold a similar edge to keep the Senate (The Hill).

If the historical pattern holds, the three-quarter window from Q4 2026 through Q2 2027 would deliver, on average, a compounded gain of roughly +24.9%. Applied to today’s S&P 500 level, that would imply a path toward the 9,200 range by mid-2027.

That is illustrative, not a forecast; the range of historical outcomes is wide, from flat to +44%. But the base rate is unambiguous: 9 out of 10 times, the market was green.

What Could Break It

No pattern is ironclad.

The 1954 to 2017 sample includes only 16 midterm cycles, and one of them (Q4 2008) produced a negative quarter. The few misses were associated with exogenous shocks: the 2008 financial crisis, the 1973 oil embargo. A severe recession, a geopolitical escalation, or a policy error from the Fed could overwhelm the midterm effect.

As shown in the chart below, the S&P 500 enters this midterm window with a forward 12-month P/E of around 20.0, above the 10-year average of 19.0 but roughly in line with the 5-year average of 19.9. Valuations are stretched relative to the longer arc, but not extreme by post-pandemic standards.

The real risk is not the headline multiple but the concentration beneath it: a handful of AI-driven names account for an outsized share of the index’s earnings and market cap.

The paper’s sample period (1954 to 2017) spans a wide range of starting valuations, but none quite like the current AI-driven concentration. A multiples de-rating would need to be offset by earnings growth for the pattern to hold.

And there is the paradox of consensus: if enough investors front-run the midterm effect, the rally could be partially pulled forward into Q3, leaving less upside for the actual post-election window. We saw a version of this dynamic in 2022, when the “midterm rally” started in October rather than November.

The Setup

The takeaway is not to blindly buy the dip on election night. It is to recognise that two forces are converging.

On the fundamental side, the AI infrastructure demand thesis is being confirmed, not contradicted, by the very earnings that spooked the market. Alphabet’s cloud growth, capex raise, and TPU monetization were demand signals, not warnings.

If Microsoft, Meta, and Amazon echo that this week, the semiconductor and memory complex is priced for a scenario that the data does not support.

On the cyclical side, the market’s most statistically reliable rally is now fewer than 100 days away. Nine times out of ten, the S&P 500 has been positive in the post-midterm window.

The average gain was nearly 25% in three quarters. And the effect is not explained by policy, luck, or volatility. It is the market’s response to the resolution of uncertainty.

A demand-accelerating secular story, corrected by 20 to 40% across the AI complex, meeting the most reliable cyclical tailwind in the modern market record. That is the setup.

The market’s “unintentional gift” from American politicians is coming. The question is whether you will be positioned to receive it.

Below is the Disruptive Investor selection as it stands today.

The next ten positions are where the AI infrastructure thesis gets concentrated: memory, optoelectronics, power delivery, and test equipment. Several are still down 35 to 57% from entry, with 200 to 950% upside to target. These are the names I am holding into the midterm window.

Free subscribers: upgrade to unlock the full portfolio, real-time entry alerts, and the analysis behind each position.

Read more

It is a timely question, and the answer speaks to something much broader about where we are in this market right now.

Nothing has changed on the fundamentals

Both names remain solid convictions. What has changed is the environment in which they are trading, and that distinction matters enormously.

Sentiment is in the driver’s seat, for now

We are navigating a market where fear and geopolitics are dominating the tape far more than earnings or business fundamentals.

The data points are striking : 82% of fund managers surveyed by Bank of America now consider the AI trade overcrowded. Meanwhile, the implied volatility of the Nasdaq 100 relative to the VIX is trading at its highest level since 2002.

These are not the hallmarks of a rational, fundamentals-driven market. They are the hallmarks of a sentiment-driven one.

Add to that the geopolitical backdrop, with fresh tensions in the Middle East pushing Brent crude back toward $88 and the WTI hovering around $82, and it becomes clear why even historically strong earnings prints are being met with indifference or outright selling.

We saw this last week with the major US banks : JPMorgan posted the best quarter in its history with $21.2 billion in net income, Goldman Sachs saw profits surge 78% to $6.6 billion, yet Citigroup and Wells Fargo closed lower on the day of their results.

The market is not trading the numbers. It is trading the noise.

The real catalyst is coming

This is precisely why the next two weeks are so critical.

Alphabet reports Q2 results on July 22, Intel on July 23, followed by Microsoft, Meta, Apple and Amazon in the final week of July. These are not peripheral names. They sit at the heart of the AI infrastructure buildout.

If these results confirm the solidity of AI capital expenditure and end demand, fundamentals should reassert themselves and cut through the geopolitical noise.

Patience is the edge

Markets driven by sentiment rather than fundamentals create exactly the kind of dislocations that long-term investors should welcome. When the underlying business is sound and the thesis intact, short-term volatility is an opportunity, not a threat.

Bloom Energy : a valuation reset worth noting

For Bloom Energy, the long-term case around distributed power generation and energy security has only strengthened in the current geopolitical context.

Importantly, the valuation picture has shifted dramatically in favor of new entrants : the 12-month forward P/E now stands at 84x, down from 191x in January 2026.

The market has already done a significant part of the de-rating work. The fundamental thesis, however, remains intact for now.

The following section is reserved for paid subscribers.

Read more

The move originated overnight in Asia. Samsung Electronics fell close to 7% and dragged the Kospi down nearly 5%, and it did so despite a sharp jump in second-quarter profit.

A reminder that in a de-risking tape the print matters far less than the positioning behind it.

Compounding the tone, a Reuters report that DeepSeek is developing its own AI accelerator revived the familiar fear of reduced dependence on the incumbent silicon supply chain, Nvidia and Samsung included.

By the time the wave reached the U.S., the entire semiconductor complex was on the back foot.

As of mid-session, the sequence is textbook. The VanEck Semiconductor ETF is off -5.4%. Nvidia is relatively defended at -1.2%; the higher-beta complex is not, with Broadcom -2.5%, AMD -7.4%, KLA -8.3% and Marvell -9.1%.

The catalyst is real but marginal. The mechanism is de-grossing: a crowded, high-conviction trade being reduced under duress, not repriced on fundamentals.

Sentiment is the variable; the process is the constant

The distinction matters because it defines where the edge sits.

When a selloff is driven by fundamental reassessment, whether a demand air-pocket, a pricing collapse or a structural share loss, the correct response is to cut.

When it is driven by positioning, whether forced de-risking, headline-triggered stop-outs or cross-asset contagion out of a single Asian session, the correct response is the opposite: lean in where the crowd is capitulating.

Share

Distinguishing the two in real time is the hard part, and it is where discipline beats narrative.

For eighteen months, the AI trade has repeated the same cycle: an exogenous shock out of Asia or a China-competition headline, a sharp multi-day drawdown across the semiconductor complex, and a recovery that begins precisely when sentiment indicators reach their extremes.

Subscribe now

The tape does not remember the headlines. It remembers the levels.

Two names in particular, which I identify below, are printing that condition today with unusual clarity.

Read more

• Nvidia slipped 1.4%, closing at $194.83 and shedding $66 billion in market value on the day.

• Micron Technology erased an early gain to close down 5.5% at $975.56, a day after a 10.6% drop.

• SanDisk was among the ones hit hardest, down over 14% to $1,745.

Many other AI issues like Lam Research also fell double digits. The tape read like a funeral for the AI infrastructure trade.

Three charts, one uncomfortable question

Attach a valuation multiple to any of these three names and the panic starts to look strange.

Nvidia trades at 19.6 times 12-month forward earnings, near the bottom of its three-year range and well under the 45x it commanded in late 2024.

Micron Technology, a stock that has gone from an afterthought to a trillion-dollar company, trades at just 6.8 times forward earnings.

SanDisk, up several thousand percent over the past year, trades at 10 times the earnings Wall Street expects twelve months out.

Look again at the twelve-month forward price-to-earnings charts above for all three names.

In each case the price line has gone parabolic while the multiple line has gone the opposite direction, grinding down toward levels last seen when these were unloved, cyclical, forgotten stocks.

That divergence only has two honest explanations. Either the analysts behind those earnings estimates are wrong, badly and repeatedly wrong, or the market itself is still pricing these businesses as though the AI cycle they are powering could vanish overnight.

I think it is the second explanation, and I think that fear is the trade.

That is where the free preview ends. Paid subscribers get the rest: the real story behind Thursday’s selloff that has nothing to do with softening end markets, why Nvidia is priced for a stall it is not showing, and the one chart that should make every bear rethink their thesis.

Read more

Optical and memory names took the worst of it. Applied Optoelectronics is down 4.5% mid-session to $141.43, alongside Coherent and Lightwave Logic, all suppliers of the hardware layer rather than finished cloud services.

Bloom Energy barely moved, up just 0.21% to $303.33.

That calm hides real news from the night before. Brookfield Asset Management expanded its AI power financing partnership with Bloom from $5 billion to $25 billion, a five times increase, sending shares up roughly 12% after hours. Market cap now sits at $86.28 billion. The money funding AI electricity just exploded, on the same day chips and memory were sold as if AI demand were ending.

The rest of this analysis, with the full data on cloud backlogs, GPU lead times, and the six data points that contradict today’s narrative, is for paid subscribers.

Read more

The table below has one purpose and one purpose only: accountability.

Every position was entered at a documented price, with a stated price target, before the market had a chance to render its verdict.

The record is public. Every entry is timestamped. The sequence is exactly as it unfolded, unedited, in real time.

No survivorship bias. No retroactive additions. Just the ledger.

This week tested both the portfolio and the conviction behind it. Here is what happened, and why the thesis remains intact.

Act I: Micron lights the fuse

On Wednesday June 24, Micron reported fiscal Q3 FY2026 results that belong in a different category from anything the memory chip industry has ever produced. Not a beat. An annihilation.

➡️ Revenue: $41.46B vs. $35.91B expected, beat by $5.54B
➡️ Year-over-year growth: +346% (from $9.3B a year ago)
➡️ Adjusted EPS: $25.11 vs. ~$20.78 expected
➡️ Gross margin: 84.9%, a record, up from 39% twelve months prior
➡️ Q4 guidance: $50B revenue midpoint vs. $43.58B expected; EPS guidance $31.00 vs. ~$25.50 expected

A company that earned $9.3B in a quarter a year ago just reported $41.46B.

That is not a product cycle. That is a structural reordering of the economics of memory. The gross margin expansion from 39% to 84.9% in a single year tells you everything about pricing power when AI clusters need high-bandwidth memory and cannot get enough of it.

MU stock surged +16% on Thursday June 25, dragging the entire AI and semiconductor complex with it.

The Q4 guidance matters as much as the Q3 print. When a company guides $50B on the heels of a $41B quarter, it is telling you the demand curve is still pointing up. That is not a blip. That is a signal.

Act II: Friday’s cold shower

By Friday June 26, that enthusiasm had been cut with cold water, twice.

The first bucket arrived Thursday evening.

The New York Times reported that OpenAI is leaning toward delaying its IPO to 2027. Sam Altman wants a $1 trillion valuation or he won't bother. Reportedly, he was not content with the $750 billion valuation his bankers were prepared to defend.

Bankers, freshly sobered by SpaceX's underperformance after its listing, pushed back. SoftBank, one of OpenAI's largest backers, fell 13% in Tokyo when the market opened on the news. The AI trade, which had been euphoric 24 hours earlier, repriced itself in a matter of hours.

The second blow hit Bloom Energy. BE fell roughly 18%, from $309 to around $252, in what amounted to a perfect storm of bad headlines arriving simultaneously. The Crusoe/Project Jade Wyoming data center pause put roughly $2.65B in potential revenue (900MW of Bloom fuel cells) into question.

FuelCell Energy, a direct competitor, announced a new data center power deal that sent its stock up 24%. And to complete the picture, Jim Chanos, one of the most prominent short-sellers still active, publicly called the AI energy build-out a bubble. Markets took all of this at face value. They usually do, on Fridays.

What has actually changed?

Read more

And then the guidance dropped.

$50 billion next quarter. The Street was at $43.58 billion. Micron didn’t beat expectations. It r…

Read more

The KOSPI crashed -10%, the Nasdaq tumbled, and the semiconductor complex got obliterated: SOX -7.87%, Micron -13.2%, ARM -10%, AMD, Intel and Qualcomm each down -5 to -8%. Weeks of gains in the AI trade, gone in hours.

It was noise. Not signal.

The real cause had nothing to do with AI demand:

1. SK Hynix, the world’s largest supplier of the specialized memory chips that power every Nvidia AI accelerator, published a report announcing it was slowing the production ramp of its next-generation AI memory chip. The market read the headline as a warning sign on AI demand and hit the sell button, without reading the detail.

2. A Korean proposal to tax unrealized capital gains on stocks and real estate triggered panic selling among retail investors.

3. Overleveraged single-stock ETFs on Samsung and SK Hynix, already flagged as a systemic risk by regulators, unwound violently.

4. MSCI confirmed South Korea stays in Emerging Markets, crushing speculators who had bet on a long-awaited upgrade to Developed Market status.

In hindsight, this kind of dislocation, where price disconnects from reality for mechanical and not fundamental reasons, is precisely what long-term investors look back on as the entry point they wish they hadn’t missed.

Here’s what the detail actually said. SK Hynix has already sold 100% of its AI memory allocation for all of 2026. Fully contracted. In June. The company is simply slowing the ramp of its next-generation AI memory chip while redirecting production capacity toward standard DRAM chips, where margins are currently approaching 90%. That is product mix optimization. That’s good management, not a demand collapse.

There are exactly 3 producers of high-bandwidth memory in the world. Supply isn’t coming from nowhere.

On June 7th in Seoul, Nvidia’s CEO Jensen Huang said it plainly:

“Everything from wafers to silicon photonics, everything’s in short supply. It is going to persist for several years.”

He called AI stocks “very cheap” and said the AI buildout “will take more than 10 years.” The SK Hynix chairman sees the memory shortage running through 2030.

Goldman Sachs puts the macro numbers on the table:

• AI infrastructure investment set to exceed $765 billion in 2026

• Forecast to more than double to $1.6 trillion by 2030, at a +20% CAGR

• Cumulative spending reaching $6 trillion between 2026 and 2031 (compute + data centers + power)

• No supply/demand equilibrium in memory before H2 2027

The structural bid is not going away.

Every data point says the same thing. Every CEO. Every analyst. Every supply chain signal. The shortage is real, the demand is structural, and the capex behind it is the largest coordinated infrastructure buildout in modern history.

And the market just handed you a discount. The question is whether you act on it before everyone else figures that out.

🔒 This content is for paid subscribers

Read more

AI infrastructure investment is set to exceed $700bn in 2026 and is forecast by Goldman Sachs to more than double to $1.6tn by 2030.

The datacenters being built today are precision environments housing thousands of interconnected processors, each producing heat, each demanding bandwidth, each dependent on the absolute reliability of every component in the chain.

Inside those environments, the transmission medium has shifted from copper to light.

Silicon photonics — the technology that encodes and decodes data as optical pulses through glass-based waveguides — is the only viable solution at the bandwidth densities required by next-generation AI accelerator clusters.

Hyperscalers are deploying silicon photonics transceivers by the millions. The technology is proven, deployed, and in production at scale.

The silicon photonics market is expanding at a brisk pace. Mordor Intelligence measures global silicon photonics market revenues at $4.0b in 2026 and projects them to reach $13.2b by 2031 at a compound annual growth rate of 27%, driven by the accelerating deployment of optical transceivers inside AI datacenter infrastructure.

But optical devices carry a failure profile that copper components do not.

Infant mortality — the elevated failure rate that manifests in the first hours of device operation — represents the primary reliability risk in silicon photonics manufacturing.

A single failed transceiver in a production datacenter rack can disrupt a compute cluster worth hundreds of millions of dollars to operate. The solution is burn-in: deliberately aging devices under thermal and electrical stress before shipment, to screen out early-life failures before they reach a live environment.

Wafer-level burn-in extends that quality gate to the earliest point in the manufacturing process.

Within the broader silicon photonics market, the equipment layer responsible for qualifying those devices before production shipment commands a distinct and growing share.

Share

William Blair, one of the few sell-side analysts with deep coverage of this niche, estimates the total addressable market for AI burn-in tools at $1.5b to $2.3b by 2030 — implying a compound annual growth rate in excess of 130% from the nascent base emerging today.

That trajectory defines the growth envelope available to the sole qualified supplier of wafer-level burn-in for silicon photonics devices in production.

The company that has built the first commercially deployed system capable of wafer-level burn-in for high-power silicon photonics devices remains largely below the radar of institutional coverage.

That gap between structural relevance and market recognition is the opportunity.

Upside: +215% | Sector: Semiconductor Equipment

PAID SUBSCRIBERS — THE COMPLETE ANALYSIS FOLLOWS

PART 2 — THE COMPLETE ANALYSIS

Read more