Read the numbers in the Anthropic IPO again, slowly. AMD, Broadcom, Google, Micron, Microsoft, Nvidia and TSMC are putting $26 billion into the offering. Anthropic, meanwhile, has more than $900 billion in commitments for cloud capacity, chips and data centers. Those are not two separate facts. They are one loop. The companies selling Anthropic its computing are buying its stock, and Anthropic's computing bills are what justify the valuation that makes the stock worth buying.
We have seen this before. In the late-1990s telecom boom, equipment makers financed the customers who bought their gear, and everyone booked growth until the customers couldn't pay. Nobody has to be a fraud for this to end badly. They only have to be certain that the music keeps playing.
The defense will be that demand reportedly exceeded the available shares, that compute shortages are real, and that raising capital in public markets is what companies do. All true. But oversubscription proves appetite, not value. And Anthropic's own prospectus concedes the mechanism: it signed enormous long-term deals after its earlier estimates of computing needs left it capacity-constrained. A company burned by guessing too low has now guessed very high, with nearly a trillion dollars of obligations riding on the guess. If the guess is wrong, who absorbs it?
Not the insiders who sold at the top, and not the suppliers who booked the revenue. It will be retail investors, pension funds and index holders who bought in after the private fortunes were made. Public markets are being asked to underwrite a private bet on a scale that would make a sovereign wealth fund flinch.
Then there is the character of this company. Anthropic built its brand on being the adult in the room: the safety-first lab, a public benefit corporation with a governance structure meant to resist exactly the pressures an IPO creates. Its own chief executive has warned that AI could wipe out a large share of entry-level white-collar jobs. That is an extraordinary thing to say while racing to build the machine. Now the company will answer to quarterly earnings, analysts and shareholders who did not buy in for the sake of caution. Good intentions are not a governance model. A promise to be responsible, made by a firm with a trillion-dollar appetite for capital, is a promise with an expiration date.
This is the moment for the Democratic Party, and anyone on the left who claims to care about technology and power, to stop treating AI as a vibes debate about chatbots and start treating it as industrial policy. The questions are concrete. What do public companies of this size have to disclose about their energy and water demand, and who pays for the grid upgrades? What happens to workers displaced by systems whose makers openly forecast the displacement? What binding, enforceable safety obligations apply when the voluntary ones collide with the share price? And if a $900 billion web of commitments starts to fray, will we be told once again that it is too connected to fail?
The Trump administration's answer is to cheer the boom and stay out of the way. That is not neutrality. It is a decision that the gains stay private and the risks get socialized later. The Left's answer should be to say now, before the bell rings, that a technology built on public data, public infrastructure and public trust owes the public something more than a chance to buy shares at a premium.
Congratulations to the founders. Now let's see who is left holding the bag.
How it may affect me
If you own a broad index fund or a 401(k), you may end up with exposure to Anthropic and the wider AI buildout, especially if the stock is added to major indexes, as large listings often are. Check how concentrated your holdings already are in a few tech giants. Data-center growth has also been tied to rising power-demand costs in some regions, so watch utility rate proceedings in your state.
