Strategic positioning — Microsoft frames AI as an application-layer revolution, embedding Copilot into Office, GitHub, and Windows to capture users where they already work. Amazon positions AI as an infrastructure and model-agnostic utility, with Bedrock and SageMaker designed to let enterprises swap models (Anthropic, Meta, Cohere) without lock-in. This is not accidental: Microsoft’s $13B OpenAI bet ties its fate to one model family, while Amazon’s $4B Anthropic investment is diversified across multiple providers and its own Trainium chips. Microsoft wins on distribution density; Amazon wins on optionality.

Product and ecosystem — Microsoft’s moat is installed base: 365 Copilot has 400K+ paying customers, GitHub Copilot 1.8M developers. Azure OpenAI Service is the default for enterprises that want GPT-4o with compliance. Amazon’s moat is infrastructure breadth: Bedrock supports 10+ model providers, Nova models undercut on price, and Trainium2 offers 40% cost-per-inference savings versus Nvidia A100. Developer adoption favors Microsoft for rapid prototyping, but Amazon leads in production-scale inference cost, especially for high-throughput agent workloads.

Recent momentum — Google’s $920M/month compute commitment signals a three-way infrastructure arms race. Microsoft’s 141 mentions versus Amazon’s 84 in our dataset suggests stronger mindshare, but Amazon’s focus on custom silicon (Trainium3 tape-out in 2025) points to a long-term cost advantage. The arXiv data shows Anthropic’s Claude Opus 4.6 papers appearing at 3x OpenAI’s rate — Amazon’s Anthropic bet is gaining research credibility, while Microsoft remains tied to OpenAI’s deployment speed.

The critical question — Can Microsoft’s distribution advantage survive if model commoditization makes Copilot replaceable? Amazon’s bet is that enterprises will demand multi-model orchestration, not single-vendor lock-in. The winner is not the best AI model, but the platform that makes switching costs lowest for buyers.