No startup has ever grown so fast. In five years, Anthropic has become an AI giant valued at $1 trillion, and it is now aiming for a record-breaking stock market debut, even if its original ambitions have been watered down somewhat.
The San Francisco-based firm faces three threats: rivals that are catching up, prices that are collapsing, and bills that keep climbing.
Anthropic spelled out the risks in its Wall Street prospectus, the document that lays out its vision to potential investors.
The first is commoditization: “the degree of differentiation among advanced AI models may narrow,” the company warns.
Additionally, Anthropic says that “substantially all” of its revenue comes from its AI tools. Unlike Google or Meta, it has no advertising empire to fall back on if business takes a hit, and no cloud business like Microsoft’s.
Investors have already grown nervous.
In mid-August, shareholders were floating a valuation of $2 trillion. But quarter-on-quarter growth slowed from 145 percent in the second quarter to 45 percent in the third, cooling expectations to around $1.3 trillion.
So-called open AI models, which anyone can download and modify, are approaching the capabilities of closed models such as Claude or ChatGPT, which are available only through their owners’ services.
Open models lag their closed counterparts by only about four months on average, according to research institute Epoch AI.
They include Kimi K3.1 Pro from Chinese company Moonshot, which costs half as much as Claude Opus 5.5, Anthropic’s current flagship.
And among paid models from Anthropic, OpenAI and Google, none has stayed on top for long.
Anthropic also acknowledges it is holding itself back: its safety measures “could slow our capability gains relative to competitors that do not adopt similar constraints.”
Still, the company says each new version of Claude has boosted sales. More than 80 percent of its revenue last quarter came from models less than six months old.
When products look alike, prices fall. According to Morningstar, the price of having AI process the same amount of text at the same level of performance has dropped 97 percent, from $17 at the end of 2024 to 50 cents today.
Anthropic itself cut the price of Claude Opus by 20 percent in September, and OpenAI cut the price of one of its flagship models by a factor of five in July.
The effect is already visible at Vercel, a developer platform that routes requests across AI models.
In August, more than half of the text it handled was processed by open models, up from 13 percent in April.
