A number arrived on my feed this week wearing the confidence of scripture. SpaceX, Anthropic, and OpenAI — five point two trillion dollars. More than every United States technology IPO of the past forty-five years, combined. No timestamp. No source. No funding round. No methodology. Just a figure large enough to stop the scroll.
I have spent fifteen years reading crypto claims, and I have learned to distrust any number that arrives without its scaffolding. Our industry gets mocked for opacity — anonymous teams, unstated treasuries, oracle feeds nobody can inspect. Some of that criticism lands. But here was a claim about three of the most consequential private companies on the planet, and it was harder to verify than a token contract dropped by a pseudonymous developer in 2021. The figure was never the story. The absence around it was.
How a private mark gets born
Private valuations are constructed, not observed. When a startup raises, the round has a price, and from that price we extrapolate a post-money figure across the entire company. That number is then repeated as if it were a market cap. It is not. It is the price of a thin, privileged slice — preferred shares, usually, carrying liquidation preferences, participation rights, sometimes ratchets and anti-dilution clauses. A headline valuation tells you what one negotiated buyer paid for a fraction of the future. It tells you nothing about what a seller of the whole could collect today.
Every daily public mark is a compromise between greed and fear, but at least it is a mark. Private marks appear when someone chooses to make one, and they vanish when someone chooses not to.
Now examine the comparison itself. "More than every US tech IPO of the past 45 years combined." That sentence has two possible readings, and it fails both. If it means the combined current market capitalization of every company that went public in that window — Apple, Microsoft, Nvidia, Alphabet, Amazon, Meta, to name the loudest — then the total runs many multiples beyond five trillion, and the claim collapses. If it means the cumulative capital raised across those IPOs, then we are weighing a stock of present valuations against a flow of historical fundraising. That is not a comparison. It is a category error wearing arithmetic.
Then there is the mixing of concepts. Two of the three companies build AI. SpaceX builds rockets and operates a satellite broadband network. Folding it into a claim about the AI economy is not analysis; it is set decoration around a thesis that should stand or fall on its own merits.
Reconstructing a number that was never built
Let me do what I do with any unaudited figure. Let me try to rebuild it from the ground up.
A post-money valuation is the product of two inputs: the capital raised and the ownership sold. If you know the round size and the dilution, you can estimate the output. What you cannot do — what the headline refuses to do — is compare that output to any measure of business quality. There is no ARR. No gross margin. No customer concentration, no renewal rate, no inference cost per query, no burn multiple, no cash runway. We are handed the denominator of a ratio whose numerator has been politely withheld.
In 2017, as a software engineering student in Washington DC, I read through more than 150 early project whitepapers — not to trade them, but to test whether their stated covenants matched the code beneath. The projects that died were rarely the ones with weak cryptography. They were the ones that could not be audited. The pattern repeated in 2020 during DeFi Summer, when I watched yield-farming schemes wrap themselves in opaque incentive structures and call the opacity "innovation." I resigned from the analytics firm I was working at after six months because I could no longer tell the difference between a mechanism design and a magic trick.
The five point two trillion figure belongs to that same family of artifacts. It is not a lie, necessarily. It may even be defensible if you accept a very specific, very generous set of assumptions. But an unauditable number offered without assumptions is not data. It is a vibe with commas.
The multi-sig behind the curtain
Here is where my governance work keeps circling back. We in crypto spend enormous energy fighting about upgrade rights. A DAO claims to be decentralized, then you read the contract and discover that a four-of-seven multi-sig can swap the implementation address at will. "Code is law" becomes a slogan for whoever holds the keys. The community votes; the admins execute; the two are not the same thing.
Private AI valuations run on the identical architecture, just with better tailoring. The headline number belongs to the class of shares with the strongest protections. The employee common stock — the shares that funded a decade of eighty-hour weeks — sits behind liquidation preferences, sometimes stacked twice over. On paper, a company is worth five hundred billion. In a down exit, the people who built it can walk away with a rounding error while the lead investor recovers principal first. The markup is real for the people who negotiated it, and theoretical for everyone else.
So when someone cites a private valuation as proof of an economic fact, I ask the same question I ask of a protocol upgrade: who holds the keys, and what happens if the price they set is wrong?
What crypto does better, and refuses to admit
There is an irony here worth sitting with. Crypto gets called opaque. And yet every DEX swap is timestamped, every LP position is inspectable, every governance proposal is archived, every liquidation is public. We argue endlessly about whether the data means what it says. We almost never argue about whether the data exists.
Even our flaws are legible. Layer2 rollups fragmented liquidity across dozens of chains that share one small user base — a real problem, and one we can measure. I can pull the TVL, watch the bridges drain over seven days, and see exactly which sequencer is bleeding. When the oracle feeds misprice because a handful of node operators are run by the same three cloud providers, the concentration is visible to anyone who looks. We criticized Chainlink for solving decentralization with a curated node list, and we were right to. At least the list is public.
Now set that against the AI valuation story, where the "feed" is a few leaked term sheets, a couple of secondary brokers, and a media ecosystem rewarded for escalation. We call it the market. It is closer to an oracle whose reporters we cannot name.
The pragmatism test
Here is the counter-intuitive part, and I want to say it plainly. The problem is not that five point two trillion might be wrong. The problem is that my industry keeps auditioning for the same opacity we claim to reject.
We envy private valuations. We build token treasuries we decline to disclose. We raise rounds we do not announce. We choreograph unlock cliffs to obscure the float rather than reveal it. We hold the multi-sig and call it decentralization, and then we point at Wall Street and laugh.
In a bull market, transparency is a liability — it invites scrutiny you can afford to avoid. In a bear market, it becomes the only asset you have left. When liquidity is abundant, nobody audits. When liquidity dries up, only the verifiable survive. Bulls react. Bears reflect. We build.
A headline from a funding cycle that has already cooled, circulating through a risk-off tape, is a bull-market fossil. The rational response in a survival market is not to argue about whether the number is true. It is to discount unverifiable inputs to zero — not because they are false, but because an input you cannot check is not an input at all.
Where this goes
The next cycle will not be won by whoever tells the largest story. It will be won by whoever can prove the smallest true thing. I expect the first serious AI companies to publish attested financials on cryptographic rails — not out of virtue, but because institutional capital will eventually demand a mark it can independently verify. When that day arrives, the five point two trillion headline becomes a relic of an era when we trusted numbers we could not check. Verify the code, trust the community. Tech changes. Values remain.