At 2:14 a.m. Pacific, I was reading a filing I did not need to read, next to a headline I could not put down. The headline came from a crypto vertical I have followed since the 2017 cycle, and it reported that Bill Ackman was rotating capital out of Big Tech and into “AI contenders.” There was no token in it. No chain, no gas fee, no validator set, no governance vote. The only cryptographic thing in the entire piece was the domain name.
Nineteen years in this industry teaches you to read a certain kind of story the way you read a smart contract: not for what it claims, but for what it fails to constrain. This story constrained almost nothing. It named no company, no dollar amount, no position size, no filing date. Two of its handful of information points were pure filler — “strategic shift,” “sustainable growth potential” — the linguistic fingerprints of a text generator filling space around a title. And yet it arrived in my feed as news, framed with the confidence of a verified fact.
That gap — between the certainty of the framing and the vacuum of the content — is the real event here. Not what Ackman may or may not have done, but that an outlet could publish essentially nothing and have the market treat it as something.
What a 13F Actually Constrains
Pershing Square is a concentrated activist fund, and its public equity positions in US-listed companies are disclosed quarterly on Form 13F, which the SEC requires within 45 days of each quarter’s close. That lag is the first constraint nobody mentioned. By the time a reader sees “Ackman bought X,” the position is at least six weeks old, in a market that reprices within days.
The second constraint is structural, and it is the one that matters most. A 13F captures long positions in US exchange-listed equities. It does not capture shorts, options, swaps, foreign listings, or private placements. So the phrase “AI contenders” is carrying an enormous amount of unexamined weight. If those contenders are public — the listed compute and data names — a future filing can confirm the story. If they are private — OpenAI, Anthropic, xAI — then the position lives entirely outside the disclosure regime, and the claim becomes unverifiable by construction. You cannot check it, because there is nothing to check.
The third piece of context is the biggest, and it gets buried every time. The Magnificent Seven are the largest buyers of AI infrastructure on the planet. Microsoft, Alphabet, Amazon, and Meta committed hundreds of billions of dollars in capital expenditure across 2024 and 2025, and every quarterly guidance revision pushes that number higher. When someone trims “Big Tech” to buy “AI contenders,” they are not leaving AI. They are changing where in the AI stack they are willing to eat risk — from the layer that sells compute to the layer that consumes it.
Which brings us to why a crypto outlet was publishing this at all. Crypto media has spent two years quietly rebranding: less on-chain coverage, more AI coverage, because attention migrated and the algorithm followed. That migration is itself a signal — not of capital, but of where the industry believes its own growth story now lives.
Crypto’s Own “AI Contenders”
Take the phrase seriously on crypto’s own terms, because there is a domestic equivalent. The decentralized-compute sector — render networks, GPU marketplaces, storage layers, inference markets — is crypto’s version of the same thesis. The pitch is clean: AI compute demand is inelastic and rising, hyperscalers are capacity-constrained and expensive, and a permissionless marketplace of idle GPUs can undercut them.
On paper, beautiful. In practice, three things have to be solved at once, and each one is a knife.
Utilization is the first and cruelest. A hyperscaler runs at high sustained load because it owns the demand curve — its own models, its own enterprise contracts. A decentralized network aggregates fragmented supply and then hunts fragmented demand, and the gap between the two shows up as idle hardware that still has to be paid for. Most networks pay for it with token emissions. Emissions are not revenue. They are a subsidy with a vesting cliff.
Latency eliminates a whole class of work. Training a frontier model across a heterogeneous, geographically scattered, intermittently available GPU set is not a solved problem, whatever the decks imply. Inference is more forgiving — and also where price competition is most vicious and margins thinnest.
Verification is where the crypto-native part lives, and where my skepticism sharpens. How do you prove a remote machine actually ran the computation you paid for? The honest answer is that most networks do not, fully. They use reputation, staking, redundant execution against a sample, or trusted hardware attestation. Each is a trade. Redundant execution multiplies cost. Trusted hardware reintroduces a vendor. Sampling leaves a statistical hole. This is the same wall I hit during my long, wet Vancouver winter of 2022, studying ZK proving costs: the cryptography is elegant and the invoice is real. Proving a computation on-chain frequently costs more than performing it, which is why so much “verifiable compute” quietly degrades into “somebody we trust says it happened.”
I learned the shape of this failure in 2020, when I launched EquiSwap around the theory that perfectly balanced liquidity pools would outperform. My curiosity carried me into exotic yield strategies, and the launch broke the moment conditions shifted. The lesson was not about pools. It was that a mechanism tuned for a calm regime becomes a liability in a volatile one. Decentralized compute has the same exposure: its economics work at high utilization and low inflation, and both assumptions snap at exactly the moment the narrative is hottest.
So when crypto AI tokens rally on an AI-capital headline, I ask three questions the narrative never asks. What fraction of revenue is real versus emitted? What share of supply unlocks next, and to whom? And which specific workload is this network actually cheaper at — at today’s gas prices and today’s token prices? In a bull market, the answer to all three is “higher.” That is not an answer.
The Vote Nobody Audits
Here is where my DAO scars itch.
In 2017 I co-founded a community fund called LibertyDAO. We had a treasury, a multisig, and a beautiful theory of decentralized governance. We lost the treasury to a flawed multisig, and the autopsy taught me the sentence I have repeated ever since: the failure was not technical, it was philosophical. We had built a mechanism without a model of who was accountable when the mechanism failed. Code is law, but people are the soul — and we had optimized for the law and forgotten the soul.
That lesson is arriving in the AI-crypto corner in a specific shape. DAOs increasingly hold AI-adjacent tokens and increasingly vote on deploying capital into compute, models, and data. Those votes are where the real allocation decisions happen. Almost nobody audits them. I have watched a mid-sized DAO approve a six-figure compute grant in a forum thread with eleven replies, two of which were the proposer talking to himself. Quorum was met by a single delegated whale wallet. That is not governance. That is a signature with extra steps.
The uncomfortable symmetry is that Ackman’s reported rotation and a DAO compute vote are the same act — a concentrated actor moving capital toward AI. Only one of them is disclosed. Only one of them pretends to be democratic. The hedge fund is at least honest about being a hedge fund. The DAO can concentrate control and still call the output a community decision, because the distribution was skewed from genesis and everyone tacitly agreed not to look.
If there is a bull-market thesis with genuine information gain in it, it is not “contenders beat platforms.” It is that AI-adjacent treasuries are the least scrutinized pools of capital in crypto, and the governance around them is the most brittle. Decentralization is a verb, not a noun — something you have to keep doing, every quarter, against your own incentives. Most structures calling themselves decentralized stopped doing it the day the token launched.
The Compliance Moat
Add the regulatory layer, because it decides who survives.
Europe’s MiCA framework is usually sold as clarity. I read it less generously. For a large, well-capitalized exchange, MiCA is a moat: it raises the fixed cost of operating, and fixed costs are trivially absorbed at scale. For a small team building an AI-adjacent protocol that touches custody, payments, or token issuance for EU users, the compliance bill — legal, audit, reporting, the CASP authorization process — can exceed the entire seed round it is meant to protect.
Watch what that does to the contender set. Well-funded survivors hire the compliance team and call it maturity. The long tail either geo-fences Europe, shrinking its market, or leans on offshore structures, which raises the exact counterparty risk MiCA claimed to eliminate. Neither outcome expands decentralization. Both expand the advantage of entities that were never decentralized to begin with.
There is a subtler distortion. Compliance has become a marketing asset. A project can pass an audit, print “regulated” on its homepage, and still run its validator set behind three addresses in one jurisdiction. The badge substitutes for the property. I watched the same pattern in the reserve-reporting debates: the label outlives the substance, and the market keeps paying for the label because checking the substance is work.
For anyone genuinely building in the AI and crypto overlap, the operative question is not “are we compliant.” It is “is our decentralization a legal posture or a physical fact?” Those are different questions, and only one shows up on the marketing page.
What You Can Actually Verify
Which returns the whole thing to verification — the one thing crypto was supposed to be good at.
Trust isn’t verified on-chain. Not the trust that matters here. I can verify a transfer, a contract state, a pool’s reserves, a proposal’s tally. I cannot verify that a fund manager believes his own press release, and no primitive bridges that gap. A 13F signature is a mark on a legal form. A wallet signature is a mark on a state transition. They live in different trust architectures, and the industry has spent a decade pretending they are the same object.
The honest version of “verify, don’t trust” is narrower: verify what is on-chain, and be explicit about everything else. When someone claims capital is rotating into AI, the verifiable fragments are always identical. Exchange net flows show whether coins are moving toward selling venues. Stablecoin mints show whether dry powder is entering. DEX and derivatives volume show where speculative energy sits. Funding rates show how leveraged each side is. None of that confirms a narrative. All of it confirms positioning. In a bull market, positioning is the only thing that moves price — which is why the noise is loudest exactly when the stakes are largest.
A real fix exists, and it is slow: signed fund disclosures, on-chain attestations bound to legal entities, standardized DAO treasury reporting. Some of it is being built. Much of it is being marketed as though it were finished. Until it ships, the correct posture is the one I learned the hard way in 2017 — assume you are missing the line item that changes the conclusion, and go find it before you size anything.
The Counter-Intuitive Read
The angle I keep returning to is that the market is on the wrong side of this story.
Everyone is reading it as a technology thesis — platforms versus challengers, incumbents versus disruptors, a bet on who wins AI. I think it is a liquidity thesis wearing a technology costume. In a bull market, the marginal dollar does not flow to the best idea. It flows to the least efficiently priced asset, because that is where the variance lives, and variance is what people are actually buying when they are afraid of missing out.
Big Tech is the most efficiently priced large-cap complex on Earth. “AI contenders” are less liquid, less covered, less arbitraged. Crypto AI tokens are less liquid still. That gradient from efficient to inefficient is the real shape of the trade, and it is not insight. It is the mechanical signature of late-cycle euphoria.
Which means the celebrated signal may be the opposite of what it appears. A famous fund moving from the most liquid corner of the market to the least liquid one is not proof the destination is cheap. It may be evidence the source is fully valued, and that whatever alpha remains lives where you cannot easily exit. That is a liquidity preference, not a thesis — and it reverses the instant sentiment does. Conviction you can stress-test. Mechanics you can only survive.
What to Do With a Headline That Told You Nothing
If you take one thing from a story that said almost nothing, take this: the burden of proof has moved, and most of the market has not noticed. Next time a headline promises that capital is rotating into the future, open the underlying form, find the line item, and check whether the future has a ticker or merely a narrative. In a cycle where attention is the scarcest asset, the ability to recognize a low-information signal and refuse to act on it may be the most valuable position you hold. Decentralization, after all, is a verb. So is verifying.