The Claude maker swung to an operating loss in the third quarter as computing costs grew faster than revenue, according to an updated filing ahead of its stock market debut.
Anthropic, the company behind the AI assistant Claude, has committed to spend $895.4bn (about £667bn) on computing power and infrastructure in the coming years, according to an updated stock market filing published on Monday.
The company said it also expected to spend up to $175bn on further contracts that have not yet been priced, taking its potential commitments above $1tn.
Of the agreed commitments, $447.5bn is expected to be spent before the end of 2030, including about $70bn in the next year. The rest falls due after 2030.
The figures were disclosed in an update to Anthropic’s S-1 filing, ahead of a planned listing on the Nasdaq exchange this month.
Anthropic’s revenue rose 46% to $17bn in the third quarter, compared with the previous three months. That was well below the 144% quarter-on-quarter growth it recorded before that.
Its spending on chips, power and infrastructure grew faster, climbing 77% from $5.6bn to $9.9bn. Data centre costs reached 58% of revenue in the quarter, up from 48% in the second quarter.
As a result, the company swung from a $974m operating profit in the second quarter to a $567m operating loss in the third, mainly because of $4.3bn in additional computing costs.
Anthropic said expensive short-term computing contracts, taken on to meet demand for Claude and its other products, had hit its margins. It spent about $5bn on short-notice contracts, and a key measure of how much revenue it earns relative to its computing spending fell for the first time.
The company has also committed to spend more than $3bn a month with Elon Musk’s SpaceX.
Anthropic reported a net loss of $23.3bn for the three months to the end of September, down from $66.2bn in the second quarter. That brings its losses for the first nine months of the year to $104bn, four times the figure for the same period last year.
Most of the net loss reflects accounting charges linked to shares issued to staff, whose value has soared, rather than cash spent running the business.
The company earned $1.2bn in interest income in the quarter. It did not include an updated annualised revenue run rate, a closely watched projection of yearly sales based on recent results.
Anthropic’s backers expect it to list at a valuation of between $1.5tn and $2tn. Its share sale could be the largest stock market listing in history, overtaking SpaceX’s $86bn raise in June.
The flotation is expected to make billionaires of the company’s seven founders, including chief executive Dario Amodei, who will keep majority voting control after the listing.
The company was founded in 2021 by former employees of OpenAI, the maker of ChatGPT.
The filing comes at a difficult time for technology stocks. The Nasdaq 100 fell 3.5% on Friday, capping its worst month in more than a year, while the Philadelphia Semiconductor index, which tracks major US chipmakers, dropped 7.2%. Stocks recovered some of those losses on Monday.
Investors have grown increasingly concerned about the cost of the AI boom, the debt being used to fund it, and whether businesses will keep paying for AI services. Expectations for Anthropic’s valuation have been trimmed as a result.
Anthropic moved ahead of OpenAI in selling AI services to large businesses, but faces renewed competition from its rival and from cheaper “open weights” models, many of them developed in China.
The updated filing also includes new warnings about the pace of progress in Anthropic’s own AI systems.
The company said its latest models “may have already advanced” at a rate that requires new internal safety measures under its Responsible Scaling Policy, a set of rules introduced in 2023 to guard against “catastrophic risks”.
It said future models “may imminently meet our Responsible Scaling Policy’s definition of ‘dramatically accelerated progress attributable to AI'”. Under its policy, reaching that point would require extra safety measures and could delay the release of new systems.
“We have in the past not identified, and may in the future not identify, a significant acceleration in our rate of development until after it has occurred,” the company said.
The language did not appear in its original filing a week earlier, which warned that its products could pose “existential risks to humanity”.