Anthropic has revealed plans to spend almost $900bn (£670bn) on computing power in the coming years in a move that risks stoking investor concerns about the cost of the AI boom .
The tech giant, which is behind the AI assistant Claude, said it had made spending commitments worth $895.4bn, around half of which is due in the next four years.
Investors have grown increasingly nervous about the cost of AI, amid concerns that large businesses who pay for the services are looking to trim their own spending. The Nasdaq suffered its worst month in more than 18 months in October.
Anthropic is preparing for a Wall Street flotation in the coming weeks at a valuation of around $1.5tn. Shareholders have cut expectations for the float amid a widespread sell-off in AI stocks owing to concerns about debt and the technology's high costs.
Documents show that Anthropic has committed to $70bn of spending in the next year, and $448bn by the end of 2030, with the remainder due after then.
It also revealed that it had racked up losses totalling $104bn in the first nine months of the year.
Anthropic lost $23.3bn in the three months to the end of September, it said in an updated IPO filing on Monday, bringing its deficit for the year to date to $104bn, four times what it lost in the same period last year.
The figure makes Anthropic one of the most loss-making companies in history.
Anthropic's giant net loss is largely an accounting quirk, affected by the cost of issuing shares to its workforce and its soaring value.
But the company also revealed that it had swung to a loss on an operating basis, which ignores such costs, and that revenue growth had slowed.
It made a $567m operating loss in the three months to the end of September, compared to a $974m profit three months earlier.
Sales were 46pc higher than three months earlier at $17bn, but they had risen by 144pc before then.
Anthropic raced ahead of ChatGPT-maker OpenAI in selling AI services to large businesses, helping them automate tasks such as software production.
However, in recent months has seen resurgent competition from OpenAI, as well as the continued rise of cheaper "open weights" models, many of which come from China.
Anthropic revealed that it had spent $5bn on short-notice contracts to secure enough computing power to meet demand.
This meant a crucial metric – the amount of revenue it makes compared with how much it spends on computing power – declined for the first time.
The company's Nasdaq flotation is expected to turn its seven founders , including Dario Amodei, the chief executive, into billionaires. The group, which split off from OpenAI in 2021, will exercise majority voting control over the company when it goes public.
Anthropic's IPO filing has warned that its product may pose "existential risks to humanity". Its updated document on Monday said the company was approaching "dramatically accelerated progress attributable to AI", a reference to a much-anticipated point at which AI can start building itself.
Anthropic's policies say that if it hits this point, it must introduce extra safety measures and potentially delay the release of new systems.
