OpenAI and Anthropic Form New Partnerships with Microsoft Microsoft is deepening its bets on multiple AI partners at once, tightening ties with Anthropic even as it reshapes its long‑running alliance with OpenAI. The moves underscore both the scale of AI demand and the strategic race around specialized chips and cloud capacity.

In recent weeks, Microsoft and OpenAI reworked their agreement, extending Microsoft’s IP rights to OpenAI models through 2032, including any post‑AGI systems, while relaxing exclusivity so OpenAI can collaborate with third parties and release open‑weight models. Crucially, Microsoft also gave up its right of first refusal to be OpenAI’s compute provider, opening the door to parallel cloud alliances.

With that constraint removed, Microsoft on May 22 announced a strategic partnership with Anthropic that brings the startup’s Claude family of AI models — including Claude Sonnet 4.5, Claude Opus 4.1, and Claude Haiku 4.5 — to Microsoft’s Foundry and Azure AI customers for the first time. Anthropic has agreed to buy $30 billion of Azure compute and can expand up to one gigawatt of capacity, signaling huge anticipated usage.

In parallel, Anthropic is “in talks to use Microsoft’s AI chips too,” with discussions underway to rent Azure servers powered by Microsoft’s in‑house Maia 200 accelerators. These chips are tuned to run existing models like Claude, even if they’re not yet the fastest option for training new ones, and Anthropic has been “steadily increasing its Azure usage.”

The Anthropic deal is explicitly non‑exclusive: Amazon remains the company’s primary cloud and training partner, while Nvidia is investing up to $10 billion and co‑optimizing future Claude models for its Blackwell and Vera Rubin systems. Microsoft, committing $5 billion of its own capital, is already weaving Claude 4 into its Copilot products and, in some scenarios, favoring Claude over OpenAI’s GPT‑5 via an automatic model selector in developer tools.

Together, the evolving OpenAI pact and the new Anthropic alliance point to a more pluralistic, multi‑model AI ecosystem—one where cloud, chip, and model providers are tightly interlinked but no longer locked into single‑track exclusivity.

Continue reading https://foxvector.com