With a potential commitment of roughly $33 billion to Anthropic, Amazon signs one of the largest financial backings to a scale-up in generative AI, securing up to 5 gigawatts (GW) of current and future Trainium chip generations to train and power its models.
In pursuing a stance as the indispensable AI pipeline rather than the creator of dominant models, its strategy differs from the two main competitors.
Comparison with Microsoft/OpenAI
Microsoft has structured an even more vertical relationship by investing about $13.8 billion in OpenAI. A 27% stake in the for-profit entity, valued at roughly $135 billion to over $200 billion, implies a potential double-digit return on its invested capital.
At the same time, OpenAI has signed a $250 billion Azure services contract over the decade, strengthening Microsoft’s position while giving it direct leverage over the models.
Amazon stands out with a more open approach: it backs Anthropic, but without the same level of financial visibility into the proprietary model. Concretely, Amazon pays a premium to access Claude, while Microsoft benefits from exposure to the value of the models and a direct capture of cloud revenues via OpenAI.
Comparison with Google and its internal ecosystem
Google, for its part, pursues a more integrated strategy and invests directly in its own stack: roughly $75 billion earmarked for AI infrastructure (chips, data centers, servers) to support Gemini and other internal models. Google supplements this with its AI Futures Fund, which provides access to advanced models, Cloud credits, and minority stakes in startups.
Put differently, Amazon sits in an intermediate zone. It does not possess Google’s full vertical architecture (models + chips + data centers), nor the stock-market valuation tied to the Microsoft/OpenAI investment, but in return it secures a contract worth more than $100 billion to be paid to AWS over ten years by Anthropic, locking in recurring revenues and strengthening its cloud dominance.
Strategic risks behind the numbers
Behind these dizzying sums lie several structural risks. First, the cross‑dependence: the more Anthropic spends with AWS, the tighter the lock‑in for Amazon, yet this also increases Amazon’s financial exposure if the models fail to generate the expected cash flows or if compute costs explode.
At the same time, Amazon has been slow to roll out a large, in‑house model, leaving it exposed to the value created by Anthropic itself, while Microsoft and Google capture more directly the value of their own assets.
Finally, the strategy is both open and biased: Amazon positions itself as a multi‑model platform via Bedrock, but by backing Anthropic with billions, it creates an implicit structural advantage for its own models on the platform, risking a challenge to its own marketing neutrality. This tension is all the more acute as Microsoft begins to diversify its internal models within Microsoft 365 Copilot, reducing its reliance on OpenAI.
*$8 billion already injected + $5 billion announced immediately and up to $20 billion conditional
