Amazon’s latest $4 billion investment brings its total equity stake in Anthropic to $8 billion, a figure that underscores the cloud giant’s deepening reliance on the AI startup. This capital injection is not merely financial; it explicitly ties Anthropic’s growth to Amazon’s hardware roadmap. The agreement designates AWS as Anthropic’s "primary training partner," ensuring that the company’s compute-intensive workloads run on Amazon’s infrastructure. In return, Anthropic will help Annapurna Labs refine Trainium chips, a direct attempt by Amazon to challenge Nvidia’s dominance in the semiconductor market.
The revenue trajectory supporting this partnership is aggressive. Anthropic’s annualized run rate surged from $9 billion at the end of 2025 to $47 billion by May 2026, a 422% jump in five months. By the third quarter, quarterly revenue reached $16.8 billion, up 45% from the second quarter’s $11.6 billion. These figures place Anthropic ahead of OpenAI, which recently cut its 2026 revenue estimate to $50 billion after previously projecting $70 billion. While OpenAI delayed its IPO to 2027, Anthropic filed its S-1 registration statement, aiming for a public offering that could value the company near $1 trillion. The gap between the two rivals has widened significantly, with Anthropic reporting its first operating profit in the second quarter while OpenAI continues to burn cash at a rapid pace.
Revenue Structure and Channel Dependence
The S-1 filing reveals that nearly half of Anthropic’s 2025 revenue came through cloud marketplaces. Specifically, $2.16 billion of the company’s $4.6 billion total revenue was generated via Amazon and Google platforms, accounting for 47% of sales. This proportion rose sharply from 32% in 2024 and 11% in 2023, indicating a growing dependency on these partners for customer acquisition. Anthropic paid approximately $351 million in channel fees, meaning the platforms charged roughly $0.16 for every dollar in revenue they helped generate. This model lowers the barrier for enterprise adoption, as clients can purchase Claude directly through existing AWS or Google Cloud contracts without separate billing processes.
However, this convenience comes with significant structural risks. Amazon and Google are both investors and competitors, maintaining their own AI models while selling the compute resources Anthropic needs. The prospectus acknowledges that these dual relationships create conflicts of interest, potentially influencing compute allocation and commercial terms. Furthermore, the revenue model relies heavily on usage-based pricing rather than stable subscriptions. Of the 2025 revenue, $3.8 billion came from actual model usage, while subscriptions contributed only $789 million, less than 20% of the total. This makes cash flow sensitive to fluctuating enterprise demand rather than locked-in annual contracts.
| Metric | Value |
|---|---|
| Total Amazon Investment | $8 billion |
| 2025 Revenue via Cloud Partners | $2.16 billion |
| Cloud Partner Revenue Share | 47% |
| Q3 2026 Quarterly Revenue | $16.8 billion |
| Annualized Run Rate (May 2026) | $47 billion |
Anthropic’s gross profit margin is projected to reach 50% this year and 77% by 2028, up from negative 94% last year. This improvement is driven by compute costs falling to about half of revenue by mid-2026, down from more than double in early 2025. Yet, the company faces hundreds of billions in committed spending on compute from Amazon, Google, and others. If demand slows, these fixed costs could strain liquidity. The core question remains whether the current velocity of enterprise adoption can sustain the exponential growth required to cover these massive infrastructure commitments. Will the revenue curve flatten before the infrastructure spend peaks?
