HomeArticle

Anthropic: Google, Amazon and Microsoft, "frenemies"

36氪的朋友们2026-10-27 14:37
Anthropic discloses its IPO documents, and its deep ties with cloud giants imply multiple underlying risks.

Latest filings show that Anthropic is increasingly relying on Google, Amazon and Microsoft to sell its artificial intelligence technology to customers.

According to the IPO registration document released by Anthropic on Monday, in the first half of 2026, sales generated through the cloud marketplaces of these three "hyperscalers" accounted for 48% of Anthropic's total revenue, compared with only 11% in 2023.

These cloud service providers are responsible for collecting payments from customers, then transferring the funds to Anthropic after deducting their own fees. The fees Anthropic paid to channel partners — a category that includes the above cloud service providers — increased by 2.7 billion US dollars in the first half of this year compared with the same period last year.

This cooperation model has driven Anthropic's rapid growth. The company's sales in the second quarter reached 115.55 billion US dollars, while recording an operating profit of 974 million US dollars.

Anthropic's IPO will be a key event to test public market investors' sentiment toward the artificial intelligence industry. Nowadays, investors seem to have begun to pay more attention to the high cost of large-scale AI infrastructure construction.

Earlier this month, reports said Anthropic's revenue growth rate in the third quarter was slowing down, which further exacerbated the broader sell-off in technology stocks, as the market worried that the growth rate of AI spending might have outpaced actual demand.

Anthropic also explicitly warned of these risks in its S-1 filing, describing these cloud service providers as playing four roles at the same time: "investors, customers, distribution partners and competitors." The company warned that these cloud service providers "are able to access our pricing and business terms."

The document also pointed out that these cloud companies may give priority to promoting their own AI models; in addition, Anthropic's close relationship with these companies has drawn the attention of antitrust regulators, and more scrutiny may be faced in the future.

"Circular Deal" Model

Previously, Anthropic was criticized for the so-called "circular deals" in the technology industry: the same companies that inject funds into Anthropic's balance sheet also obtain revenue from Anthropic.

Anthropic is an important customer of Amazon AWS. It uses Amazon's data centers and chips to train and run the Claude model, and Amazon in turn provides Claude to customers through its Bedrock platform.

Similarly, Anthropic uses Google's custom AI chips, while Google sells Claude through its own cloud platform. In other words, these companies are not only Anthropic's investors and suppliers, but also its sales channels and competitors.

Anthropic also relies on these companies in terms of computing power. The company has promised to pay at least 114.5 billion US dollars to Google by 2033, at least 110 billion US dollars to Amazon by 2036, and 44.9 billion US dollars to Microsoft.

Moreover, these minimum commitments remain valid regardless of whether Anthropic actually uses these computing power capacities in the end.

The "Frenemy" Relationship With Cloud Giants

These figures further reflect the complex "frenemy" relationship between Anthropic and cloud giants.

In April this year, when Anthropic reached an agreement with Amazon, Anthropic called AWS its "primary training and cloud service provider for mission-critical workloads". However, judging from the contract amount, Anthropic's minimum procurement commitment to Google is larger, and the term is shorter.

In addition, Anthropic expects to pay another 125 billion to 150 billion US dollars for additional TPU (Tensor Processing Unit) computing power. This additional computing power will come from Google and other suppliers, and is scheduled to be delivered between 2028 and 2031.

Amazon and Google Are Both Major Shareholders of Anthropic

Amazon and Google are both important shareholders of Anthropic, but the IPO document does not disclose how many shares the two companies hold specifically.

One of the reasons is that the shares they hold have no voting rights, so they are not included in the list of shareholders holding more than 5% of Anthropic's voting securities.

However, Anthropic stated in other parts of the document that after the conversion of preferred shares and certain convertible notes during the IPO issuance, the shares held by Amazon and Google each will exceed 5% of the company.

Since agreeing to invest up to 40 billion US dollars in 2023, Amazon has established a very deep investment and business relationship with Anthropic.

According to the IPO document: Amazon has invested about 18 billion US dollars in Anthropic so far; and has promised to invest an additional up to 150 billion US dollars when specific conditions are met.

In addition, Amazon also holds convertible notes from the purchase agreements in 2023 and 2024. These notes can be converted into up to 1.84 billion Anthropic shares, equivalent to about 10.9% of Anthropic's fully diluted shares as of June this year.

This does not include other equity interests that Amazon currently holds. However, there is a restriction on these convertible notes: if the conversion would cause Amazon's total shareholding ratio to exceed 12%, the notes cannot be converted unless Amazon voluntarily waives this restriction.

The part exceeding the 12% upper limit will be automatically converted into new notes. These new notes can be converted at the end of the subsequent quarter, but are still subject to the same 12% shareholding upper limit constraint.

Google has invested about 230 billion US dollars in Anthropic through multiple rounds of transactions, and has promised to invest an additional 200 billion US dollars after reaching specific performance targets. According to the latest round of financing agreement, the upper limit of Google's shareholding ratio is 14.99%, and Google can also waive this restriction.

Anthropic Is Deeply Tied to Cloud Giants

The IPO document details how Anthropic has tied its own success to these cloud computing giants.

Anthropic said that if it moves workloads to other cloud service providers, the process will be "costly and time-consuming", and may damage the company's computing efficiency.

This means that Anthropic's reliance on Google, Amazon and Microsoft is not just a simple supplier relationship, but has penetrated deep into the company's infrastructure and business model.

However, Anthropic's reliance does not stop at the three major cloud computing giants. It also relies on other suppliers. For example: Anthropic has a 161.2 billion US dollar equipment lease agreement related to Broadcom; Anthropic has agreed to spend up to 120.5 billion US dollars to obtain computing power from SpaceXAI.

In other words, Anthropic is building an extremely large-scale AI infrastructure system, but at the same time, this also means that the company is increasingly relying on a small number of large technology giants and chip suppliers.

Judging from the IPO document, this dependency relationship is not only an important driving force for Anthropic's rapid expansion, but also an important source of potential future costs, supply chain, competition and antitrust risks.

This article is from the WeChat official account "CLS.cn", author: Zhao Hao, published with authorization from 36Kr.