The numbers coming out of Anthropic are the kind that would once have silenced any doubt. The company’s annualized revenue hit $65 billion in July, and its backers expect the figure to exceed $120 billion by the end of the year, a trajectory that would make it the fastest-growing company in history. Yet the same investors who recite those figures are now arguing about whether they can be believed, and their disagreement is showing up in the price they would put on an initial public offering.
The gap is striking. Anthropic is currently valued at $965 billion, according to people familiar with the company’s finances, but estimates of what it might trade at after a listing range from $1.5 trillion to $4 trillion. The spread between the low and high ends is roughly the size of Amazon’s entire market value, a range so wide that it amounts to two different companies wearing the same name.
What drives the doubt is not the past but the future. Anthropic’s growth has been extraordinary, but it rests on customers who can switch models with a few keystrokes and who have shown little loyalty beyond price and performance. Competition from OpenAI, Google, Meta, and a crowd of well-funded startups is intensifying, and the models they release now leapfrog one another on a quarterly schedule.
There is also a risk that has nothing to do with quarterly revenue. Anthropic has built its identity around the idea that artificial intelligence could one day threaten humanity, and around the claim that it will build the technology more safely than its rivals. Investors who buy the stock are being asked to accept that the company’s caution will not become a competitive handicap, even as others race ahead.
Joey Brookhartz, an analyst at SemiAnalysis, the research firm that tracks AI labs, put the uncertainty plainly. “It’s stunning,” he said of the growth, “but it’s too early to say how this ends.” The comment reflects a view that has taken hold among investors who once treated Anthropic’s trajectory as inexorable: the company’s rise is real, but its destination is not yet legible.
The valuation spread is not just academic. It will determine how much money the company raises, how its shares price, and whether early investors and employees see the returns they have been promised. A company that prices at the low end would still be one of the most valuable in the world; one that prices at the high end would reset every record for a technology debut.
Anthropic’s own posture has not helped settle the question. The company has said it will prioritize safety and alignment over speed, a stance that wins it trust among cautious customers but leaves open the question of whether it can keep pace with rivals who have no such reservations. In a market where the best model tends to win the marginal dollar, restraint is a strategy that has to be explained over and over.
The competition for enterprise customers is the sharpest point of pressure. OpenAI’s GPT-6 line has been making inroads among corporate buyers, and the leading models have converged to the point that differences in capability are measured in percentage points rather than categories. When the products are nearly interchangeable, price becomes the weapon, and price is exactly where Anthropic’s competitors are willing to bleed.
The existential-risk framing adds a final layer of complexity. Anthropic has argued that the very technology it sells could be dangerous, and that argument is now a feature of its corporate story. Some investors find the candor reassuring; others worry that a company which tells the world its product might be dangerous is not the one they want to defend in a downturn.
None of this means the IPO is in doubt. The company is proceeding toward a listing, and demand for shares in the most prominent AI companies remains strong. What has changed is the temper of the conversation around it, from one of certainty to one of open disagreement about what the company is worth and whether its growth can last.
SemiAnalysis has become one of the most closely read voices on the economics of the AI buildout, and its analysts have spent two years mapping the gap between the industry’s revenue claims and the cost of the computing that produces them. Brookhartz’s caution draws on that work, which has repeatedly warned that the companies racing to sell AI are spending faster than they earn and that the final shape of the market is not yet visible.
The people familiar with the company’s finances describe a business that is still accelerating, and the revenue figures support that view. But the market is now doing what markets do with a company moving this fast: it is trying to decide whether the speed is sustainable, and pricing the answer into a range wide enough to hold almost any outcome.
