E-Paper
Anthropic’s IPO filing opens the books on what it costs to build frontier AI and who pays for it
Amazon and Google invest in Anthropic; with Microsoft, they also sell it computing power, resell Claude and compete with it, the IPO filing shows.
Prefer HTon Google
Anthropic, the company behind the Claude chatbot, made its finances public on Monday, as its filing gave the clearest view yet of how the companies building the most advanced artificial intelligence (AI) models pay for them.
Anthropic filed its prospectus, the document a company must publish before selling shares to the public, with the US Securities and Exchange Commission (SEC), the US market regulator. The 335-page document prepares the way for a listing on the Nasdaq exchange next month under the ticker “ANT”, the Associated Press reported. The company is aiming for the week of November 16, the Wall Street Journal reported, though the timing could change.
The initial public offering (IPO) could raise as much as $100 billion, which could make it the largest ever, ahead of the $86 billion SpaceX raised in June. Reports put Anthropic’s value at listing between $1.5 trillion and $2 trillion. Neither figure is final. Besides the US, the shares will be offered to the public in Australia, Canada, Japan, Switzerland, the UK and parts of Europe, according to the filing.
Anthropic would be the first of the standalone companies building frontier AI to list. OpenAI, the maker of ChatGPT, has pushed its own listing to next year. Bloomberg called the offering a test of investor appetite for AI companies, after weeks of concern over stretched valuations and heavy spending.
The revenue sources
Anthropic earned $16.3 billion in revenue in the first six months of 2026, about 14 times the $1.16 billion it made in the same period of 2025, according to the filing. Its revenue in the April-June quarter was $11.6 billion, more than twice the $4.6 billion it earned in all of 2025, AFP reported.
Much of that growth comes from businesses using Claude to write software and handle office work. Nearly 12,000 organisations now spend at least $100,000 a year with Anthropic, up from fewer than 1,000 a year earlier, AFP reported, and about 250 spend $10 million or more, according to AP. Many customers pay for what they use, metered in tokens, the small units of text, code or images that an AI model reads and writes; individual users can also pay fixed monthly subscriptions.
In the April-June quarter, Anthropic made its first operating profit, close to $1 billion, meaning revenue exceeded the cost of running the business. But, over the full six months, it still made an operating loss of $693 million, down from $2.1 billion a year earlier, according to the filing.
The $80.7 billion loss
Anthropic’s net loss for the half-year, which counts everything including accounting charges, was $80.7 billion. Of that, $72.4 billion came from a change in the value of convertible notes, which are loans investors can swap for shares, and similar instruments.
The filing says the charge was driven mainly by the rise in Anthropic’s own valuation. The more the company is worth, the more those notes are worth to the investors holding them, and accounting rules record that increase as a loss for Anthropic. The charge reflects the company’s rising value rather than cash spent, AFP reported.
The charge has appeared every year as the valuation climbed, at $1 billion in 2023, $5.4 billion in 2024 and $34.3 billion in 2025. Anthropic lost $13 billion in the first half of 2025, AP reported.
Also read: Don’t expect supersize profits from Anthropic any time soon
The cost of compute
Anthropic’s biggest expense is compute: the specialised chips, data centres and power needed to train AI models and to run them for customers. It spent $8.95 billion on compute and infrastructure in the first half of 2026, against $2.33 billion a year earlier.
Revenue is growing faster than that bill. In the April-June quarter, revenue rose 1,368% from a year earlier, while compute and infrastructure costs rose 300%, according to the filing.
Anthropic has also signed up for far more capacity than it uses today. Since the end of 2025, it has committed $417 billion to computing and infrastructure under agreements lasting from one year to 20, Bloomberg reported. Many of these payments must be made whether or not the capacity is used. Anthropic has committed to spend at least $114.5 billion with Google through 2033, $110 billion with Amazon through 2036 and $44.9 billion with Microsoft, regardless of use. It estimates it will pay a further $125 billion to $150 billion for more capacity on tensor processing units, chips Google designed for AI, from Google and other providers, to be delivered from 2028 to 2031.
Against those obligations, Anthropic held $131.7 billion at the end of June, AP reported. Of that, $112.4 billion was cash and cash equivalents, and the rest short-term investments, according to the filing.
Dario Amodei, Anthropic’s chief executive, who co-founded the company in 2021 with other former OpenAI employees, has said that misjudging demand could sink it. Predicting demand for a technology that changes this fast, and buying costly computing capacity years ahead to match, is unusually hard, he had said in a podcast early this year, Bloomberg reported. “If I’m just off by a year in that rate of growth, or if the growth rate is 5x a year instead of 10x a year, then you go bankrupt,” he had said.
Investors, suppliers, sellers and rivals
Amazon, Google and Microsoft each play several parts in Anthropic’s business. The filing describes the cloud providers as “investors, customers, distribution partners and competitors”, Bloomberg reported. One of its risk factors warns that relying on a few such companies, some of which invest in Anthropic and buy from it, “creates complex dynamics that could give rise to conflicts of interest”.
Two of the three are among Anthropic’s largest backers. Amazon has invested about $18 billion and committed up to $15 billion more, subject to conditions. Google has invested about $23 billion and committed $20 billion more, tied to performance targets, Bloomberg reported from the filing. Anthropic rents computing power from all three, under the commitments above.
The three also sell Claude. Businesses can buy it through their existing Amazon, Google or Microsoft cloud accounts; the cloud company bills the customer, keeps a fee and passes the rest to Anthropic. These cloud marketplaces brought in 48% of Anthropic’s revenue in the first half of 2026, up from 11% in 2023.
The fees add up. Anthropic’s sales, marketing and partnership costs rose to $3.2 billion from $120 million a year earlier, mostly because of a $2.7 billion increase in fees paid to partners, the cloud companies among them, Bloomberg reported.
All three also build or sell AI models that compete with Claude. The filing warns that the cloud companies can see Anthropic’s pricing and commercial terms and could favour their own models, and that the ties have drawn antitrust scrutiny. Critics have previously called such arrangements “circular deals”, with the same companies putting cash into Anthropic and taking it back out, Bloomberg noted. Amazon and Google did not immediately respond to Bloomberg’s requests for comment.
Borrowing to rent chips
Lenders supply some of the money too, in arrangements in which Anthropic is not the borrower. In these deals, a company set up for the purpose raises debt, buys chips with it and rents them to Anthropic, with the chipmaker Broadcom backstopping the debt, Bloomberg reported.
The first such deal, called Big Sky, raised $35 billion, most of it backed by Broadcom. On Monday, Broadcom announced another, worth $63 billion. Earlier this month, Bloomberg reported that banks were preparing a $42 billion senior portion of a $60 billion chip-financing package meant to benefit Anthropic and other companies, while Blackstone, a US investment firm, was leading an $18 billion junior portion, which is repaid only after the senior debt. Broadcom and Blackstone declined to comment to Bloomberg.
Anthropic borrows directly as well. As of June 30, it owed Amazon $6.2 billion in convertible notes. It has a $15 billion credit line from banks led by Morgan Stanley, and can draw up to another $77 billion in convertible notes from Amazon, Broadcom and Google, Bloomberg reported. Leases it has signed that have not yet begun, mostly for compute and infrastructure, rose to $321 billion at the end of June from about $50 billion in December. The filing warns that Anthropic may need to raise money sooner than it had expected. Anthropic declined to comment to Bloomberg.
Other technology companies were also raising large sums this week. Meta was seeking to raise as much as $95 billion, and SpaceX signed an equipment-lease deal worth about $40 billion, Bloomberg reported.
Also read: Anthropic’s IPO filing warns of catastrophic AI risks to humanity
The risks
Anthropic told investors its growth is likely to slow, possibly sharply, AFP reported. Revenue grew more than 1,000% in 2025. Its quarter-on-quarter growth then slowed from about 140% in April-June to about 45% in July-September, according to earlier reports.
The company will publish its July-September results before the listing.
Its revenue is also concentrated. The top five customers accounted for about 25% of revenue this year, AP reported, and the cloud marketplaces of three companies that compete with it for nearly half.
Its relationship with the US government is another risk. In February, the Trump administration accused Anthropic of endangering national security and designated it a supply chain risk, in a dispute over limits the company placed on military use of its technology. Anthropic sued the Pentagon in March. A California judge ruled in Anthropic’s favour in that case, while a Washington, DC, judge sided with the government in a separate, narrower one. Both sides are appealing the orders.
The filing’s risk factors run to 93 pages, according to AP. They range from losing key staff to the possibility that AI models could “pose catastrophic or existential risks to humanity”.
The market test
The filing lifted chip stocks across Asia on Tuesday, after concerns over valuations and spending had driven a rout in the sector, Bloomberg reported. Samsung Electronics and SK Hynix, the South Korean chipmakers, each rose more than 4%. Japan’s Advantest, which makes chip-testing equipment, rose about 10% during the session, Reuters reported. Results due this week from Alphabet, Microsoft and Amazon, which are Anthropic’s partners as well as three of the biggest spenders on AI, will be watched for signs that the spending is paying off.
A successful listing is expected to encourage other AI companies to go public. “Never have IPOs represented so much to the venture capital industry writ large,” Ravi Viswanathan, founder and managing partner of NewView Capital, a US venture capital firm, told Bloomberg.
The case for the shares rests on AI revenue in the coming years covering what is being spent now. Aswath Damodaran, a professor of finance at New York University’s Stern School of Business who specialises in valuing companies, told Bloomberg that no one, Amodei included, knows for sure what those sales will be in five years. “They are just as lost as you or me,” he said.
