SemiAnalysis estimates ~90% of Anthropic's business comes from agentic AI, while sources say nearly 25% of its revenue in 2025 came from just two clients

Your AI assistant costs may fluctuate unpredictably as providers shift to usage-based pricing.

SemiAnalysis estimates that approximately 90% of Anthropic's business revenue is generated from agentic AI applications. Furthermore, internal sources indicate that in 2025, nearly 25% of Anthropic's total revenue was derived from just two of its clients. This highlights a significant reliance on a specific AI technology and a concentrated customer base for the company. The broader context involves AI agents rapidly rewriting computing economics, with companies like Uber experiencing significant budget overruns due to the unpredictable and high token consumption of these autonomous tools. This has led to a shift towards usage-based pricing by AI labs, transferring cost risks to customers, who are responding with spending caps. There is a growing industry-wide discussion about moving pricing models from token-based to value- or outcome-based systems, a critical consideration for AI companies like Anthropic and OpenAI as they approach potential IPOs.

Anthropic's business model appears heavily skewed towards agentic AI, which, while potentially lucrative, carries inherent risks due to its high computational demands and unpredictable token usage. The concentration of nearly a quarter of its 2025 revenue from just two clients points to a significant dependency that could impact its long-term stability and valuation, especially as the company eyes an IPO. This client concentration, coupled with the general industry trend of unpredictable AI costs, suggests potential vulnerability for Anthropic.

The economics of AI agents are clearly shifting, with companies facing ballooning costs and labs attempting to manage risk through new pricing structures. The industry's move from token-based to outcome-based pricing is a crucial development for both providers and users. For customers, this means a potential future where AI costs are more aligned with delivered value, but also a need for careful monitoring and budgeting as the transition occurs.