#Anthropic推进IPO估值最高1.4万亿美元 A $1.4 trillion valuation puts the AI bubble on the table!

Anthropic has launched its IPO roadshow. The price range has a $78 cap, with 825 million shares planned for issuance and a maximum valuation of $1.4 trillion.
That figure already exceeds the annual GDP of many countries.
Even more striking are two other figures: the seven co-founders plan to cash out a combined $1.75 billion in the IPO; the company plans to raise up to $90.4 billion through the IPO and a private placement, potentially surpassing SpaceX’s record; and Goldman Sachs predicts its revenue will exceed $600 billion by 2030.
On one side, massive fundraising; on the other, founders cashing out.
The market’s reaction to these two things happening at once is telling—Nvidia fell more than 9% over two trading days, wiping out about 2.85 trillion yuan in market value. On Baidu’s trending searches, “Seven giants’ $26 billion bet on Anthropic” and “Nvidia loses 2.85 trillion yuan in two days” appeared side by side.

What concerns me isn’t the $1.4 trillion figure itself, but the three calculations behind it.
First: where the money comes from.
Anthropic’s revenue mainly comes from API calls and enterprise subscriptions. Its models are widely integrated into coding tools, office software, and customer service systems. In other words, its revenue doesn’t come from ordinary users paying for monthly memberships, but from businesses paying based on usage. The upside of this model is large bills; the downside is customer concentration—a major client switching providers can make revenue fluctuate. Second: where the money goes.
Training models requires computing power, and computing power requires chips. TSMC has already agreed with customers that chip manufacturing services will become 5% to 10% more expensive starting in early 2027, covering processes below 7 nanometers. The higher cost of chips will ultimately be passed along, layer by layer, to model companies and then to the people using their models. So the hotter AI gets, the higher the bills—a direct chain reaction.
Third: who pays.
If businesses find that the efficiency gains from AI don’t outweigh rising usage costs, they’ll cut back. At that point, the $1.4 trillion valuation will take longer to justify.
Goldman Sachs’ forecast of $600 billion in revenue by 2030 is essentially a bet that businesses will keep increasing their investment.
For ordinary people, there are two practical takeaways.
First, don’t treat “AI company fundraising” as proof that “AI is useful.” The amount raised reflects capital’s bet on the future, not the quality of the product today. Spend ten minutes trying the tools you use, and you’ll know how well they work.
Second, if you use AI services from a small team, watch for price changes. Rising chip prices upstream and mounting fundraising pressure on model companies could ultimately mean higher bills for you or reduced free usage limits. Make key workflows “replaceable” in advance, and don’t tie all your work to a single service.
As for whether to invest in related stocks, I won’t offer advice. What I can say for sure is this: when an industry starts telling stories in units of trillions, the most important thing for ordinary people is to manage their cash flow, not try to guess the next peak.

One more thing is easy to overlook: the list of underwriters for this IPO includes almost every major Wall Street bank.
The more prestigious the underwriting syndicate, the more it suggests the issuer wants to fetch a good price—and that these banks are confident in demand.
But history has repeatedly shown that when underwriters are most optimistic, retail investors often have the most reason to stay calm. If you’re just an ordinary user, the most practical impact is that free trials of AI tools may get shorter and shorter. Companies have to report earnings and answer to shareholders, so subsidies will gradually be rolled back. Try more tools while you can, and turn the workflows that work well into established routines—that’s better than anything else.
Anthropic’s IPO will move the term “AI bubble” from the comments section into the prospectus. Whether there’s a bubble, and when it might burst, nobody knows. But the bills are real, and they’re going up day by day.