Anthropic PBC is relying more and more on Alphabet Inc.’s Google, Amazon.com Inc. and Microsoft Corp. to sell its artificial intelligence technology to customers, despite the three companies being rivals, investors and key suppliers of computing power.
Amazon, Google and Microsoft customers can buy access to Anthropic’s Claude AI models through their existing cloud accounts. The cloud providers bill the customer, keep a fee and pass the rest to Anthropic. That business accounted for 48% of the AI startup’s revenue in the first half of 2026, up from 11% in 2023, according to Anthropic’s initial public offering filing published Monday.
The setup has fueled Anthropic’s fast growth. Sales hit $11.6 billion in the second quarter, when the firm posted a $974 million operating profit.
But the fees Anthropic pays the three cloud companies add up. The AI lab’s sales, marketing and partnership costs jumped to $3.2 billion in the first half of the year from $120 million a year earlier — driven mostly by a $2.7 billion increase in fees paid to channel partners, a group that includes the cloud providers.
Also read: Anthropic files for IPO, eyes one of history’s biggest stock market debuts
At the same time, the S-1 filing warns of risks. Anthropic describes the cloud providers as “investors, customers, distribution partners and competitors” all at once. It cautions that the companies have “visibility into our pricing and commercial terms.” The filing also says the cloud companies could favor their own AI models, and that Anthropic’s ties to them have attracted antitrust scrutiny and could draw more.
Anthropic’s IPO will be a key test of AI sentiment among public market investors, who already appear more sensitive to the high costs of AI buildouts. Reports of Anthropic’s third-quarter revenue growth slowing compounded a broader tech selloff earlier this month amid concerns that AI spending may be outpacing demand.
Anthropic has previously been criticized for “circular deals” in the tech ecosystem, with the same companies both putting cash into its balance sheet and taking it out. Anthropic is a significant Amazon Web Services customer, using Amazon data centers and chips to train and run its Claude models, while Amazon offers Claude through its Bedrock platform. Similarly, Anthropic uses Google’s custom AI chips, and Google sells Claude through its own cloud platform.
Anthropic depends on the same companies for computing power. It 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, whether or not it uses the capacity.
The figures add to the frenemy dynamics. When Anthropic struck its Amazon deal in April, it called AWS its “primary training and cloud provider for mission-critical workloads.” But its minimum commitment to Google is larger and runs over a shorter period. Anthropic also estimates it will pay another $125 billion to $150 billion for additional Tensor Processing Unit computing capacity from Google and other providers, to be delivered from 2028 to 2031.
Major Shareholders
Amazon has forged a deep investment and commercial relationship with Anthropic since agreeing in 2023 to invest as much as $4 billion. It’s since invested about $18 billion in Anthropic and committed up to an additional $15 billion, subject to certain conditions, according to the filing.
It currently holds an economic stake of roughly 9% on an as-converted, fully diluted basis, according to Bloomberg calculations. It also holds about $6.2 billion of outstanding convertible notes issued under financing agreements signed in 2023 and 2024, which could convert into 1.84 billion shares — equivalent to an additional 11% of Anthropic. The company’s ownership stake is limited to a maximum of 12%, but Amazon has the ability to waive that.
Also read: Don’t expect supersize profits from Anthropic any time soon
Google, meanwhile, has invested about $23 billion in Anthropic over multiple deals and committed to invest an additional $20 billion tied to performance targets. It currently holds an economic stake of at least 6.6% on an as-converted, fully diluted basis, according to Bloomberg calculations, which exclude a separate $300 million investment by the company in 2002 that was reported by the Financial Times and isn’t detailed in Monday’s filing.
It also has $20 billion of remaining financing capacity under a 2026 agreement that could result in additional equity or convertible-note issuance if drawn. Its stake is capped at 14.99% under its latest financing deal, though Google can also waive that limit.
In addition, Google has an option to invest up to $10 billion in stock in a private placement concurrent with Anthropic’s IPO at the public offering price. Any amount invested through that option would reduce the remaining capacity under the financing agreement.
Also read: Anthropic IPO filing: Claude maker targets Nasdaq listing under ticker ANT amid AI market jitters
Amazon and Google didn’t immediately respond to requests for comment.
The filing spells out how Anthropic has entwined its own success with the cloud companies. Moving its workloads to other providers would be “costly, time-consuming” and could hurt its compute efficiency, the company says.
Anthropic’s dependence doesn’t end with the three cloud giants, however. It also leans on other suppliers. It has $161.2 billion in equipment leases with Broadcom Inc. and has agreed to spend up to $120.5 billion on computing power from SpaceXAI. The firm also says chipmaker Taiwan Semiconductor Manufacturing Co. has at times been unable to meet demand.