New Constructs Questions Reported Anthropic IPO Valuation

Whether Anthropic is worth $150 billion or $2 trillion is the least interesting question here. The more useful one is who ends up holding the asset when the market decides.

Start with the numbers, and with their limits. New Constructs is a research shop with a bearish thesis, and its figures rest on leaked documents it says show $4.6 billion in 2025 revenue and a $42 billion loss. Anthropic says its annualized run rate hit $65 billion by late July. Those two figures are hard to reconcile. Either the company is growing at a speed with no real precedent, or "run rate" is carrying a lot of rhetorical weight. I don't know which, and neither does anyone outside the company's books. That is the point. Private markets let the people selling the story control the evidence, and an IPO is the moment that arrangement is supposed to end.

The optimist's case deserves a fair hearing. Great companies have looked absurd on trailing earnings, and Amazon lost money for years while skeptics were wrong. Going public can also be a democratizing act, letting ordinary savers into a gain that otherwise stays with venture funds and sovereign wealth. If the growth is real, the bears will look foolish.

But the structure of this particular offering should make anyone pause. New Constructs argues the listing would mainly give Wall Street backers liquidity. That claim is plausible whatever one thinks of its valuation model, because that is the usual function of a mega-IPO. Early investors have marked their stakes to a private price that no public buyer has tested. The IPO converts those marks into cash. Public investors supply the cash, and they do so at the price the underwriters, insiders, and a hungry market agree on. The democratization story is real in one sense: the public gets to take over the risk after the people with the best information have priced it.

At $2 trillion the risk spreads far beyond people who choose to gamble on AI. A company of that size would flow into major indexes, and index funds must buy whatever the index holds. The buyers include the retirement accounts of teachers, nurses, and warehouse workers who never made a decision about Anthropic and have no vote on its strategy. Passive investing was sold as a way to avoid single-stock bets. A handful of AI firms priced for perfection turns it into a concentrated bet on one technology, made on workers' behalf without their consent.

Then there is the mission question. Anthropic has presented itself as a safety-minded public benefit corporation. That is an interesting experiment in whether a company can hold commercial pressure at bay. Public markets are where that experiment gets stress-tested. A firm priced at several multiples of what its earnings can justify faces quarterly pressure to grow into the price, which means more compute, more deployment, more speed, and more reasons to cut corners on caution. If the valuation is built on a promise nobody can meet, the shortcuts will eventually serve the share price. I'm not claiming that outcome is certain. I'm saying the incentive points that way, and the public gets no say in it.

So the demand is not to cheer or jeer at the number. It is to insist on transparency before any share trades. A prospectus should reconcile the gap between recognized revenue and run-rate claims. It should disclose the full cost of the compute commitments, the loss trajectory, the real competitive threat from open-source models, and exactly who is selling and when. Lockups on insiders should be long and meaningful. Regulators should treat "annualized run rate" with the skepticism they would give any metric that conveniently flatters a number.

AI may well be transformative. That doesn't mean its financing has to follow the old pattern, in which the winners cash out and the losses are socialized across everyone's pension. The test of this IPO, if it happens, is whether the public gets the facts before it gets the bill.