Anthropic has publicly filed its S-1 prospectus with the SEC, with the stock ticker "ANT," underwritten by Goldman Sachs and Morgan Stanley. This will be the largest IPO in human history. The document reveals a lot of interesting details, and one thing to be thankful for: we don't see an AI bubble 👀
Entering 2026, Anthropic's revenue growth accelerated further, reaching $16.286 billion in the first half of the year, up about 1,300% from $1.163 billion in the same period last year. Half a year's revenue alone is about 3.6 times its full-year revenue for 2025.
However, the net loss over the same period also widened to $80.748 billion, compared with $12.969 billion a year earlier. About $72.388 billion of that came from fair value changes in convertible debt, derivative liabilities, and SAFEs (simple agreements for future equity), mainly related to the rise in Anthropic's valuation. This is a non-cash accounting loss; it did not involve actually paying out cash of a similar scale.
From an operating perspective, Anthropic's operating loss in the first half narrowed from $2.135 billion in the same period last year to $693 million, a decline of about 68%. In the second quarter, it posted GAAP operating profit of about $974 million.
Claude has expanded from "the engineers' programming tool" to "a work agent for the whole company," and it has begun to be embedded in products that customers sell to their own end users. Agents that can actually get work done have entered real enterprise production environments, operating like bearings in a business. And all of this is just beginning!
In addition, Anthropic is putting more compute into the model's self-improvement to secure its leading position. But the two most noteworthy risks in the S-1 are:
First, customers and cloud providers are too concentrated. A considerable portion of Anthropic's revenue comes from a few large customers and sales channels. If one or two of them reduce their spending, its results will be noticeably affected. At the same time, the compute needed to train and run its models depends mainly on a few cloud providers. These cloud providers have complicated roles: they are suppliers, investors, and customers all at once, and their interests are not fully aligned. If the relationship changes, compute supply and revenue could be hit at the same time.
Second, shareholders have almost no control. Voting power is concentrated in the Long-Term Benefit Trust and the founders, and far exceeds their actual economic stake. Ordinary Class A shareholders mainly buy economic returns and have essentially no say over the board, major decisions, or even a sale of the company. Moreover, Anthropic is a public benefit corporation, so the board must balance shareholder returns against the company's mission, and shareholder interests are not necessarily the top priority.
Personally, I feel that after this IPO, Anthropic will mark a short-term peak for US stocks, and the market will soon undergo sharp pullbacks due to various unexpected news. For example, customers using various APIs more economically, customer churn, and slowing growth could all be factors causing market volatility.
But in the medium to long term, enterprise AI adoption has only just begun, and there is plenty of room for AI's real penetration and for leading companies' revenue and valuation growth. The real AI bubble lies in the hot early-stage companies right now 🤔