About $72 billion of the loss came from a non-cash charge. It reflects the rising fair value of convertible notes, derivative liabilities and SAFEs, which grew as the company's valuation climbed. Without that charge, Anthropic lost $693 million from operations in the six months ended June 30, an operating margin of about minus 4 percent. On an adjusted basis it reported $1.6 billion in operating income and $2.7 billion in net income.

Reported figures for revenue in the same period a year earlier range from $1.1 billion to $1.2 billion, which puts first-half growth at roughly 14 times. In 2023, revenue was $41 million. More than 11,900 customers now generate at least $100,000 in revenue each, up from 950 a year ago. Net revenue retention was about 900 percent as of September 30, which means long-standing customers are spending about nine times what they spent a year earlier. Computing and infrastructure costs fell from $2.01 per dollar of revenue a year ago to 55 cents.

Anthropic held $131.7 billion in cash, cash equivalents and short-term investments at the end of June. Against that, it disclosed $245 billion in future minimum payments on infrastructure. These include up to $114.5 billion with Google through 2033, at least $110 billion with Amazon through 2036, and up to $120.5 billion with SpaceXAI and Nvidia. An earlier draft of the filing reportedly put total long-term compute commitments at $518 billion. The final total has not been made public. The company warned that it may never become profitable, that it expects costs to rise and that it anticipates continued losses.

The filing has more than 90 pages of risk factors. Among them are AI safety risks, including model behaviors that resemble self-preservation, code sabotage and manipulation. The company lists these as possible risks and does not say they routinely occur. Chief Executive Dario Amodei has called for a slowdown in releasing new capabilities to address such risks.

Anthropic is incorporated as a public benefit corporation and has pledged to keep its Claude model free of advertising. Its seven co-founders will keep a voting majority through a single share that carries 50.1 percent of the vote on most corporate matters.

Anthropic was valued at $965 billion in May. The Wall Street Journal reported that the offering could raise $100 billion or more at a valuation between $1.5 trillion and $2 trillion. After the financial details came out, investors were valuing the company at about $1.5 trillion, down from about $2 trillion a week earlier. Jared Carmel, co-founder of Manhattan Venture Partners, said his firm wants the shares "priced to sell." The offering price and third-quarter results have not been disclosed.

Reports of the timing differ. The Journal reported a target listing in the week of November 16, and other estimates range from mid-November to before the end of the month. Rival OpenAI filed for an IPO in June and is expected to list by early 2027.