Microsoft says its Maia 200 AI chip is now running OpenAI's GPT-5.2 models, Microsoft's own MAI workloads and supporting infrastructure for AI services as the company pushes to lower AI costs in both software and hardware. The TSMC-built 3-nanometer inference accelerator has been deployed in Iowa, with a second rollout planned near Phoenix, and Microsoft says it can deliver operating costs 30% to 40% lower than comparable Nvidia hardware for certain models, with about 30% better performance per dollar. The company also introduced MAI Code 1.1 Flash, a cheaper and faster coding model rolling out on GitHub Copilot, with pricing reduced by about 73%, faster token generation and improved CLI and .NET performance. Microsoft is preparing a Maia 300 launch and has discussed securing more than 300,000 chips of TSMC manufacturing capacity by 2027, though that timeline and scale remain subject to supply constraints and have not been fully confirmed by the company.