Cloud AI arms race intensifies

Microsoft’s restructured deal with OpenAI gives Microsoft a 27% equity stake in the new OpenAI Group PBC and ties OpenAI to an incremental $250 billion of Azure cloud purchases, according to company filings and Microsoft’s October 28, 2025 announcement. (geekwire.com) The joint OpenAI–Microsoft statement published in February 2026 specifies that stateless API calls to OpenAI models from third parties will be hosted on Azure and that OpenAI’s first‑party Frontier products will continue to run on Azure. (openai.com) AWS reports that more than half of its Amazon Bedrock service now runs on Amazon’s Trainium family of chips, a shift AWS says reduces reliance on third‑party GPUs and lowers per‑token hosting costs for customers. (tradingview.com) To counter Microsoft’s Azure‑OpenAI position, AWS struck a multiyear deal to add Cerebras WSE‑3 wafer‑scale engines to Bedrock, promising up to 5x more token capacity for inference and planned availability of Cerebras hardware on Bedrock later in 2026. (winbuzzer.com) Analyst forecasts and market data show the commercial stakes: Gartner put worldwide AI spending at roughly $2.52 trillion in 2026 with AI infrastructure driving a $401 billion component of that spend, while cloud infrastructure revenues concentrated among AWS and Azure crossed the $100 billion quarterly threshold in 2025. (gartner.com) Both hyperscalers are locking down chip and partner ecosystems — Microsoft reaffirmed deep Azure infrastructure commitments with OpenAI and AWS expanded partnerships with NVIDIA and custom‑chip vendors — intensifying procurement choices for enterprises evaluating model hosting, latency SLAs, and vendor IP terms. (blogs.microsoft.com)