Why Amazon and Google Have Become the Main Threat to Anthropic

Anthropic has released an investor prospectus ahead of its upcoming initial public offering (IPO). In the document, the developer of the Claude AI model explicitly stated the risks associated with its deep dependence on tech giants, as well as the threats posed by autonomous AI agents.

Partners, Clients, and Competitors All in One

Anthropic identified its relationship with Amazon and Google as its primary risk factor. The tech giants simultaneously act as investors in the startup, provide it with server capacity, sell its services, and develop their own competing AI solutions.

Furthermore, these key partners have access to confidential information regarding the startup's pricing and marketing strategies.

Revenue and Commitments

By the end of last year, 47% of sales of Anthropic services ($2.16 billion out of $4.6 billion in total revenue) flowed through Amazon (AWS) and Alphabet (Google) infrastructure. However, cloud platforms take a significant cut for mediation: an average of about 16 cents for every dollar Anthropic earns (commissions totaled $351 million).

Anthropic's financial performance demonstrates rapid scaling: over the past year, the startup's revenue grew 12-fold, reaching 4.6 billion (3.8 billion generated by Claude enterprise clients, and another $789 million from individual subscribers). 

Securing AI infrastructure also requires substantial expenses from Anthropic: for instance, by early 2026, long-term data center contracts for computing power surged to $417 billion.

The Risk of Autonomous AI and a Legal Vacuum

A dedicated section of the prospectus focuses on the risks tied to the development of autonomous AI agents. Anthropic warns that agents’ ability to make independent decisions increases the likelihood of system failures, errors, and cyberattacks resulting in real-world damage.

The situation is further complicated by the lack of a regulatory framework:

• The legal status of AI agents has not yet been established in legislation.
• The extent of users' legal liability for the actions of algorithms toward third parties remains undefined.
• Liability limits outlined in contracts may prove insufficient to cover potential losses in the event of major incidents.

The company acknowledges that current safety measures and internal constraints may be unable to fully prevent harm to both clients and Anthropic's own business.