At 6:40 on a Tuesday morning in Miami, a number crossed my feed and moved money it could not possibly have touched.
The number was $965 billion. The headline attached to it concerned Anthropic — the AI lab behind Claude — and Rishi Sunak, the former British prime minister, who had reportedly agreed to join the company as an adviser. By the time I finished my second coffee, four AI-adjacent tokens across two chains had repriced between nine and nineteen percent. None of them has any contractual relationship to Anthropic. None of them licenses Claude. None of them appears anywhere in an Anthropic filing, or a term sheet, or a cap table I could ever hope to see.
What happened in that window was not investing. It was not even speculation in the ordinary sense. It was a private valuation — generated by paperwork that no public investor will ever read — propagating through an information system optimised for speed and indifferent to provenance, and finding at the other end a market that will price anything in ninety seconds flat.
A transaction is just a promise frozen in time. There was no transaction here. There was a promise, unfrozen, drifting out of a press release and onto a pool of liquidity like a leaf falling onto still water — and the water rippled in exactly the shape of the leaf.
I have watched this pattern before. In 2017, while finishing a master's in economics, I hand-audited fifteen ICO whitepapers for a Miami fintech, and the thing that struck me was never the code. It was how quickly a well-set table of tokenomics could stand in for a business. The market priced the story first and audited it later. Sometimes never. The mechanism has not changed. Only the collateral has.
So let me be precise about what I actually know here, and what I am inferring, because with this particular story the difference matters more than usual.
Context: what is documented, and where my knowledge runs out
Anthropic's valuation arc is reasonably well documented up to the edge of what I can verify. Roughly $5 billion in 2022. Around $18 billion in 2023. $61.5 billion in March 2025, attached to a $3.5 billion raise. Later reporting through 2025 pointed to an implied range somewhere between $170 billion and $350 billion on the private market. Past that point, I have to stop pretending. My working knowledge largely ends around mid-2025, and anything after it I will flag rather than guess.
$965 billion sits a full order of magnitude above the top of that trajectory. For the figure to be real, Anthropic would have had to more than double or triple its implied worth inside a handful of months, which would place it above the market capitalisation of the overwhelming majority of listed technology companies — on annualised revenue that, even on a generous read, sits in the low tens of billions at best. That is not impossible. Private markets have produced impossible arithmetic before and will again. But it is extraordinary, and extraordinary numbers carry an extraordinary burden of proof.
The most likely explanations, in my own order of probability: a transcription error, where $96.5 billion or $9.65 billion lost a decimal point somewhere in the pipeline; a conflation of a raise size with a post-money valuation; or a figure lifted from an unconfirmed secondary-market rumour or an analyst's addressable-market scribble. Any of these happens in a newsroom at four in the morning. None of them is a scandal on its own. All of them become one the moment they hit a liquid market, because liquid markets do not wait for verification. They price the first version.
Now, the part of the story that is solid. Sunak. A former prime minister with a Goldman Sachs and hedge-fund background who, while in office, convened the Bletchley Park AI Safety Summit and midwifed the UK's AI Safety Institute into existence — joining an AI lab whose entire brand is built on being the careful one. That is not a technology hire. It is a policy hire, and a geopolitical one, and the distinction is where the actual signal lives.
In my day job I work at the intersection of central bank digital currency design and stablecoin infrastructure, and I have spent the last two years watching compliance requirements get absorbed into protocol architecture rather than bolted on top of it. When a former head of government joins a lab, the thing being purchased is not insight into model architecture. It is proximity to the institutions that decide what is legal to build, plus the credibility to be invited into the room where those decisions are drafted. Governments buy AI the way governments buy anything — slowly, through procurement, on the recommendation of people who are personally exposed if the vendor embarrasses them. A former prime minister's name is a risk transfer. That is a design decision, not a public relations one.
Core: the anatomy of an unsoracled price
There is a version of this article that is simply a complaint about a bad headline, and that version is boring. The structural question is sharper: why does a private company's estimated worth have price impact in public, permissionless markets at all?
Start with what a private valuation actually is. It is not a measurement. It is the residue of a negotiation between a lead investor and a founder, expressed as a post-money figure, frequently accompanied by terms that make the headline number look rosier than the underlying economics. It is not marked to market, because there is no market to mark to. It is not audited in any sense a public equity investor would recognise. It carries no obligation to be restated when it turns out to be wrong, and no mechanism by which anyone is penalised for repeating it.
In systems terms: it is a data point with no oracle, no dispute window, no slashing condition, and no finality. If you tried to settle a smart contract against it, you would be laughed out of the room. And yet it settles trades constantly — just not on-chain. It settles them inside the heads of everyone who reads the headline, and then it settles them again when those heads place orders.
A valuation is a promise nobody is required to keep. That asymmetry is the whole problem. A transaction is a promise frozen in time. A valuation is the same promise, still liquid, still deniable, still unfalsifiable.
The AI-token complex is where this gets interesting. The tokens that repriced that morning pitch themselves as exposure to something real — decentralised compute markets, agent frameworks, inference networks. Their fundamental connection to Anthropic is approximately nil. But their narrative connection is total. They trade as levered beta on the AI capital cycle, and the AI capital cycle's most visible signals arrive as headlines rather than as filings. So the price of a token becomes a function of the media cycle of a private company, transmitted through an information channel with no verification layer anywhere in it.
I have modelled a version of this reflexivity before. In 2026, working on how autonomous AI agents interact with on-chain liquidity pools, the thing that kept surfacing was that these markets do not price information. They price the speed of information, and the shape of whichever story arrives first. A wrong number arriving first beats a right number arriving sixty minutes later, every time. That is not a market failure. That is a market working exactly as designed, on inputs it was never designed to receive.
Now layer the liquidity environment on top of it. There are dozens of Layer 2s now, and the same small population of active users circulating between them. When news hits, it does not hit one deep, unified order book. It hits twenty shallow books that all reference each other for price. The move becomes bigger and uglier than the underlying information deserves, because everything that would normally absorb a shock has been sliced into fragments. This is not scaling. It is dilution wearing the costume of scaling — and the costume only comes off during exactly the kind of event I spent that morning watching.
So who benefits? The accounts that saw the headline first. The market makers quoting both sides of a widening spread. Anyone with a Telegram bot and a latency advantage over the rest of the room. Not the retail buyer who read "$965 billion" and concluded that the company had been validated by someone. That reader was handed not information but an aesthetic — a very large number set in a confident typeface, implying a certainty that nothing in the causal chain supports.
Strip all of that away, though, and something substantive remains. Every serious AI lab is converging on a similar capability frontier. The benchmark gaps between the top three or four are now measured in single-digit percentage points and are routinely swamped by prompt engineering. When the product converges, differentiation retreats to the non-technical layer: procurement, jurisdiction, liability, trust.
Anthropic chose, years ago, to make safety its brand rather than its constraint — constitutional training, interpretability research, cautious release schedules. In an ordinary market that reads as a cost centre. In a market where the buyers are hospitals, banks, defence ministries and revenue authorities, it reads as a distribution channel. Safety is not the thing slowing the company down. Safety is the thing that unlocks the customer segment with the longest contracts and the lowest churn.
And there is an irony here that my own industry has spent a decade failing to appreciate. For years, DeFi's self-image held that regulatory engagement was a betrayal of the ethos — build something that cannot be switched off, and the rules will be forced to adapt. Then MiCA arrived, and the protocols that survived were not the purist ones. They were the ones that had built hooks, allowlists, transfer restrictions and reporting layers into their architecture from the beginning. In 2025 I wrote a thirty-page report on eight protocols that redesigned their contracts without abandoning their core value proposition, and every one that came through did so because it treated compliance as a design surface rather than a wall. Anthropic is running the same play at a larger scale, with a former prime minister where a KYC module would otherwise sit. Same logic. Different ledger.
Contrarian: the decoupling thesis is backwards
The prevailing story in my corner of the market is that AI-crypto has decoupled — from Bitcoin, from the halving cycle, from the liquidity tides that used to govern everything. AI tokens are supposed to be their own asset class now, driven by technological progress rather than macro.
That Tuesday morning said the opposite, and it said it loudly.
These assets have not decoupled from correlation. They have decoupled from observation. They are the most tightly coupled instruments in crypto — coupled not to chain activity, not to fee revenue, not to anything they can measure, but to the capital cycle of a private company whose financials are invisible and whose news arrives through a media pipeline with no oracle attached to it. That is the most fragile coupling available in any market. A genuine decoupling would mean these assets trade on something they can verify. Instead they trade on something they cannot even see.
I would rather have my beta attached to a Federal Reserve statement — which at least arrives on a schedule, with a document, and a press conference where someone can be asked a follow-up question — than to a headline about a number that may well be a typo.
There is a second blind spot here, quieter than the first. The revolving-door framing writes itself: former rule-maker joins regulated entity, ACOBA review, cooling-off period, regulatory capture. It is legitimate, it deserves attention, and I suspect it is the least consequential thing in the story. The more interesting question is not whether Sunak's arrival poses an ethics problem. It is whether the rest of us can do anything with the fact that the AI industry has now formalised the purchase of institutional trust as a line item — while crypto is still holding internal debates about whether trust is a product at all.
Anthropic is not being accused of anything. It is being priced. And a substantial portion of that price is the accumulated credibility of a state, converted into something that trades.
Takeaway: watch the mandate, the correction, and the tokens
So where does that leave a reader who wants to act rather than react?
Watch the mandate before you watch the number. If what Sunak has actually been retained to do touches procurement, defence or export controls, this was a structural hire and the consequences will be measurable inside a quarter. If it is ambassadorial — a name on a letterhead — then it is cosmetic dressing, and the entire news cycle has been about a business card.
Watch, too, whether the $965 billion figure is corrected, and how. A silent edit teaches you more about a newsroom's standards than a formal retraction ever would. Whether Anthropic itself pushes back tells you which story the company wants told: the one where it is worth $965 billion, or the one where it is disciplined enough to say so out loud.
And watch the tokens. If the AI-crypto complex keeps pricing itself off unverifiable figures, the next shock in this corner of the market will not arrive from a rate decision or a regulatory filing. It will arrive from a journalist at a desk at four in the morning, typing a number that has no settlement layer, into a market that has no oracle for the truth.
A transaction is just a promise frozen in time. A valuation is a promise nobody has to keep. The market that learns to tell the two apart will be the one still standing when the difference finally costs something.