The Forced Absorber

/ 01Cembalest Named the Beast, Then Fed It

Every June, Michael Cembalest — Chairman of Market and Investment Strategy at J.P. Morgan Asset & Wealth Management, and genuinely one of the most data-literate people the sell side has ever produced — publishes a special edition of Eye on the Market. The 2026 edition is called “Semiquincententacles”, a pun on the 250th anniversary of the Declaration of Independence and the octopus-eagle hybrid he puts on the cover to represent America's grip on global markets. The Aquilaceph, he calls it. Half bald eagle, half octopus. It is a forty-two page document and it is, in its own institutional way, magnificent.

It is also a confession. Cembalest lays out — in his own charts, with his own footnotes — nearly every component of the machine I have been describing in this series. The mechanical bid of passive money. The float-weighting rule that turns insider selling into forced retail buying. The hyperscaler profits that are increasingly just paper markups on private bets. The single Taiwanese chokepoint underneath the whole edifice. The dismantling of the institutions that are supposed to see trouble coming.

He documents all of it. And then, because his job is to keep clients invested, he files it under “signs of investor exuberance in AI” — a phrase that does to the situation roughly what “some weather” does to a hurricane — and reassures you that, in global context, things are basically fine.

Cembalest even prints the disclaimer. On the page where he begins the AI section, there is a literal box he titles “Warning label,” and it is worth reading slowly, because it is the sound of a smart man telling you he is about to narrate a fire while standing very still:

“This section includes technical AI jargon, conjecture on AI products and services that are rapidly changing, views on non-public companies whose disclosures can be opaque or incomplete, and thoughts about the future which may be wildly off the mark… take a deep breath and remember that I'm trying my best to cover what is arguably the most technically complex market catalyst in the last 40 years.” — Cembalest, “Semiquincententacles,” §5, June 2026

Here is the difference between what Cembalest does and what I am about to do. Cembalest is the heir to the great explanatory tradition of finance: he tells you what the flows were, beautifully, after they have already happened, in a register calibrated so that no client ever does anything as rude as sell. That tradition has a function, and the function is not to warn you. It is to make the absorption feel like analysis.

My job is the opposite one. It is to look at the exact same data and tell you the thing the institution structurally cannot: that the machine is not malfunctioning. It is running perfectly. It is just running on you.

/ 02The Smoking Gun Is in the Appendix

Most people who worry about index funds worry about the wrong thing. They worry that passive investing is “dumb money” that distorts prices. Fine, true, boring. The actual mechanism that should frighten you is more specific, and Cembalest documents it in Appendix I of his report, where he thinks no one excitable is reading.

Indexes like the S&P 500 are weighted by float-adjusted market cap. In Cembalest's own words: weighted by “shares available for trading by the public in the secondary market, excluding restricted and locked-up shares.”² Read that twice, because the entire trap is in it. A passive fund does not own a company in proportion to how big it is. It owns it in proportion to how many shares insiders have already let go of.

Now watch what happens at an IPO lockup expiration. Cembalest studied the ten largest IPOs since 2010. They went public with an average free float of 45%. When the lockups expired, that float jumped to 75%.² Why? Because — again, his words — “insiders and pre-IPO investors sell shares in the secondary market.” And the instant those shares become “available for trading,” they get counted into the float. The index weight mechanically rises. Every passive fund tracking that index is now required to buy more of the stock.

Sit with the choreography. Insiders sell. The act of insiders selling increases the index's required holding. The passive fund buys exactly what the insider is unloading, in the amount the insider unloads, because a rule said so. The seller picked the moment. The buyer was conscripted.

This is why the SpaceX calendar is not a metaphor. SpaceX came public in 2026 as the single largest IPO on record — $75 billion in proceeds.³ At the IPO, its float was about 5% of its own market cap and roughly 0.1% of the entire US equity market. By Cembalest's estimate, as lockups roll off, that figure climbs toward 2.9% of the total equity market by July 2027.³ Musk alone holds about 42%, on a one-year lockup. Between mid-August and mid-December 2026, JPM estimates 40–50% of SpaceX shares unlock — “amounting to roughly US$900 bn. If the stock price rises, the figure could exceed US$1 trn.”³

That is the supply. Here is the demand that was built to meet it: passively managed funds now hold roughly 55% of US equity fund assets, something on the order of twenty trillion dollars that buys what the index tells it to buy.⁴ And in May 2026, Nasdaq rewrote its entry rules to fast-track the largest companies into its index in fifteen trading days with the float requirement waived⁵ — pre-positioning the conscript army at exactly the gate the insiders are about to walk through. (To their modest credit, S&P's committee rejected a parallel profitability waiver in June.⁵)

None of these numbers are mine. They are Cembalest's, plus a public ICI tally and a Nasdaq rule change. I have added exactly one thing he did not: the sentence that connects them.

/ 03The Profits Are Eating Themselves

A reasonable person might say: fine, but these are the most profitable companies in history. Surely the absorber is at least buying real earnings. This is where Cembalest hands me the single best paragraph in the report, in a text box on the concentration page that I would frame on a wall if I were a more sentimental man:

“In Q1 2026, other income was 60%, 51% and 27% of Google, Amazon and NVIDIA profits; it includes marked up positions on frontier labs like OpenAI and Anthropic.” — Cembalest, “Semiquincententacles,” §5, June 2026

Stop and metabolize that. A majority of Alphabet's first-quarter profit — sixty percent — did not come from selling ads or cloud or anything you could ship. It came from marking up the value of its own private stakes in OpenAI and Anthropic. Half of Amazon's. More than a quarter of Nvidia's. These are unrealized gains on illiquid holdings in companies that file no public statements, priced by the same funding rounds the markups are meant to validate.

It is a circle, and the circle is the point. Nvidia sells chips to the hyperscalers and to the labs. The hyperscalers and sovereign funds pour capital into the labs. The labs spend that capital buying Nvidia chips and renting hyperscaler compute. Everyone's revenue is someone else's capex; everyone's profit is partly a markup on everyone else's valuation. Cembalest notes, evenly, that cloud providers are sitting on roughly $2 trillion in revenue backlogs and that frontier-lab compute commitments are about half of that.⁶ Anthropic has raised over $95 billion in equity this year alone.⁶ OpenAI spent $34 billion last year.⁷

And the cash? On the labs' own claimed timelines for when they turn cash-flow positive — OpenAI 2029–2030, Anthropic 2028 — Cembalest renders the only honest verdict in the building:

“At this point, I view frontier lab projections of when they will be cash flow positive… as speculative, uncertain and subject to revision.” — Cembalest, “Semiquincententacles,” §5, June 2026

Meanwhile the hyperscalers are pouring something like $725 billion of combined capex into AI in 2026.⁸ Meta's capital spending is running between 109% and 127% of its revenue, depending on the quarter you catch it.⁸ Anthropic is burning around $1.25 billion a month on a single Tennessee data-center campus against a $30 billion revenue run-rate.⁸ This is the engine the absorber is being filled with: profits that are partly self-referential markups, financing a capex cycle whose payback even its own banker calls speculative.

And the one genuinely external thing in the whole circle — Nvidia's pricing power — is leaking. By Cembalest's own chart, Nvidia's share of accelerator revenue slips from 85% in 2023 to an estimated 75% in 2026 as Google, Amazon and Meta's in-house ASICs and a wave of cheap Chinese open-weight models eat into the moat.⁹ The moat the valuations assume is permanent is visibly draining, in the same report that prices it as permanent.

“Sixty percent of Alphabet's profit was a number it wrote down about something it owns. The index bought it at par.”

/ 04Eleven Days of Gas

Underneath the circular profits and the float mechanics there is a physical object, and the object is an island. Cembalest is admirably blunt about it. Eight of the ten largest companies on Earth by market cap depend on TSMC; the only exceptions are TSMC itself and Saudi Aramco. More than a third of those eight companies' combined $2 trillion in revenue runs through hardware built on TSMC products. Global trade in semiconductors has now overtaken global trade in crude oil as a share of GDP.¹⁰ In his words: “without it the world economy would sputter.”

So how robust is the chokepoint that the entire Computedollar is balanced on? Cembalest lists the numbers, and they are the numbers of a hostage. Taiwan imports 90% of its primary energy. It let its nuclear generation fall from 50% of electricity in the 1980s to 5% today, replacing it with imported LNG — and it keeps, on hand, ten to eleven days of natural gas.¹¹ He writes that he “almost didn't believe these figures” when he first read them. Taiwan imports 60% of its food and 67% of its calories. LNG suppliers in Singapore now write act-of-war clauses into Taiwan delivery contracts. Cembalest's own conclusion: Taiwan “may be the most blockade-sensitive advanced economy in the world.”

A reminder of what the index fund is actually holding when it holds the Nasdaq 100: a leveraged bet that an island with eleven days of fuel, encircled on a standard day now by Chinese destroyers and frigates, remains perfectly accessible forever. That is not diversification. That is a single point of failure wearing the costume of a broad market.

And the bet is decaying from the other side too. China's GPU self-sufficiency, Cembalest reports, has gone from 10% in 2021 to 40% today, with a path to roughly 80% by 2030.¹² Huawei announced in May 2026 a route to 1.4-nanometer-class chips by 2031 via vertical logic density, sidestepping the EUV machines it has been embargoed from.¹² Commerce Secretary Lutnick wants 40% of US chip demand onshored by the end of 2028; Cembalest estimates the US might, generously, reach 30–35% of advanced-node production by 2028–2030.¹⁰ Which is a quiet way of admitting that for the rest of this decade, the answer to “what if Taiwan?” is: there is no answer.

/ 05They Are Firing the Smoke Detectors

Here is the part Cembalest flags, in his own introduction, as one of the two biggest medium-term threats to US assets — alongside the debt — and then, having flagged it, cannot do anything with, because doing something with it would mean naming the project:

“The biggest medium-term concerns for investors in US assets…: the increased unpredictability in the rule of law, and government defunding of science and sidelining of scientific expertise.” — Cembalest, “Semiquincententacles,” Introduction, June 2026

Section 10 of his report is a careful accounting of the United States dismantling its own scientific base. Competitive NIH grants down sharply against the prior decade's average. The entire National Science Board fired in April 2026, days before it could publish a report on China closing the gap. American scientists submitting 32% more applications for jobs abroad. He quotes Shirley Tilghman, the former president of Princeton, and the quote is the whole thing:

“The intention was to punish elite universities, it was not to destroy the scientific capacity of the United States, but that's what they're doing… It's one thing to destroy something. It is quite another to destroy it and have nothing to replace it with.” — Shirley Tilghman, quoted in “Semiquincententacles,” §10

Then he gets to the OMB rule, and this is where the financial story and the political story turn out to be the same story. In May 2026, the Office of Management and Budget proposed a rewrite of the rules governing every federal grant, binding on every agency by October 1, 2026.¹³ Political appointees — not scientists — would conduct a “pre-issuance review” of every discretionary grant, and are explicitly forbidden from deferring to peer reviewers. Peer review becomes, in the rule's own language, “advisory” and not “binding.” Active grants can be terminated mid-stream for being “inconsistent with… agency priorities.” Applicants can be denied for their affiliations.¹³

Understand what this is, in market terms. The institutions that fund and validate independent inquiry — the people whose entire job is to notice, early and without permission, that something is wrong — are being converted into instruments that can only say what the executive wants said. The economy is firing its smoke detectors at the exact moment it has loaded the building with the most flammable capital structure in forty years. A market with 55% of its money on autopilot was already a market that had outsourced its judgment. Now the public institutions that might have supplied judgment from outside are being lobotomized on a published timeline too.

Cembalest sees this clearly enough to rank it as a top-two risk. What he cannot say, from inside JPMorgan, is that it is not a bug in the AI trade. It is the same move. Hollow out the thing that can say no — the active manager, the peer reviewer, the civil servant — so that capital and power flow without friction to the people writing the rules. The forced absorber and the captured grant office are the same machine pointed at two different public goods.

/ 06Who Is the Absorber

Let me now say the sentence the report is built never to say. The absorber is you. Specifically: it is the 55% of US equity money that moves by rule, concentrated by that rule into the Magnificent Seven, who now make up roughly 32.7% of the entire S&P 500⁴ — a basket of 42 AI-linked companies that Cembalest reports have driven 65–80% of the index's returns, profits and capital spending since ChatGPT launched.¹⁴ It is the pension fund whose funding ratio looks healthy only because that basket went up. It is the target-date fund inside your 401(k) that bought SpaceX on its lockup schedule because the float said to.

When Cembalest reaches the concentration number, watch the institutional reflex fire. The top ten stocks are now about 40% of the S&P 500, up from 17% in 2015 — a genuinely alarming figure — and his immediate next move is to soothe: “40% concentration still ranks among the three lowest equity concentration figures in the world; only Japan and India have less.”¹⁴ Which is true, and which is exactly the kind of true thing you say to keep someone from selling. Your retirement is less concentrated than Switzerland's. Sleep well.

The tell is everywhere once you see it. The leveraged-ETF complex that now amplifies every semiconductor move — its rebalancing market impact up fivefold since early 2024, by JPM's chart.⁹ Margin debt at 5.74% of M2 as of April 2026, the second-highest reading in recorded history, behind only the dot-com peak.¹⁵ The semiconductor index stretched far above its 200-day average at levels Cembalest's own desk labels an “acid flashback” to 2000. A new Fed chair, Kevin Warsh, confirmed in May, and the 30-year Treasury yield breaking above 5% the same week.¹⁶ Every one of these is in the report. Every one is presented as an observation, never as an instruction.

Because that is the deal. Cembalest's role — the honorable, lucrative, structurally compromised role of the great market explainer — is to describe the absorption so well that you mistake the description for a warning, and stay. He is not lying to you. That is the genuinely unsettling part. Every number checks out. He has simply been hired to narrate the machine in a tone that keeps you standing under it.

“The insider picks the moment. The rule picks the buyer. The buyer is you.”

/ 07The Calendar Doesn't Care

I want to be careful here, because the speculative future is where this kind of writing goes to die. I'm not going to tell you when it breaks. The forced absorber is grotesquely durable precisely because trillions are trapped inside it — pensions can't exit without crystallizing the loss, retail is anchored by inertia and fees, and the active managers who might have constituted an alternative were fired a decade ago to cut costs. A machine this large doesn't crash on a schedule. It grinds.

But I will tell you what is not speculative, because JPMorgan printed it. The lockups are dated. The insider sale plans are filed. The float will rise on those dates, and the rule will buy on those dates, whether or not the price deserves it. The hyperscaler profits will keep including markups on private companies their own banker calls speculative. The island will keep holding eleven days of gas. The grant offices will keep being captured on the October timeline. None of that is a forecast. All of it is a calendar.

Cembalest closes his AI section with a warning label and a deep breath. I'll close mine differently. The most dangerous document in finance is not the one that lies to you. It's the one that tells you the exact truth in a voice engineered to make you do nothing about it. “Semiquincententacles” is that document, and it is a good one, and you should read it — not for the reassurance it's selling, but for the confession it can't help making. The beast on the cover isn't a metaphor for American strength. It's a picture of what's standing on top of your retirement. Cembalest drew it himself. He just labeled it majestic.