Microsoft's Maia 300 Chip Cuts Reliance on Nvidia GPUs — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Microsoft announced Maia 300, its third-generation in-house AI chip, with production rollout to AI workloads beginning in 2027 and plans to make it the "primary AI inference system" for Microsoft products and services.
- Maia 300 delivers 40% more tokens per megawatt than leading GPU-based systems, and in a 96-accelerator configuration achieves up to 7x the rack-scale token throughput of its predecessor Maia 200, according to Microsoft cloud EVP Scott Guthrie.
- Microsoft will scale its Azure Maia family to multi-gigawatt deployments starting in 2027, shifting the bulk of its AI workloads off Nvidia and AMD silicon.
- Maia 300 uses a chiplet architecture built on TSMC's N2P process node, separating compute, memory controller, and I/O functions so each can use the best-suited silicon technology, per Microsoft's Azure hardware and infrastructure team.
- The prior-generation Maia 200, only announced in January, is already running production AI workloads across Microsoft Copilot, GitHub Copilot, Microsoft Foundry, and Microsoft's own AI models.
- Microsoft AI CEO Mustafa Suleyman said running Microsoft AI models on top of Maia chips is "a critical focus" because competing with Google DeepMind, OpenAI, and Anthropic requires co-design from models down to the silicon.
Why it matters: Microsoft is making a multi-gigawatt, multi-year bet on its own silicon that pulls AI inference workloads off Nvidia and AMD starting in 2027, with the 40% efficiency gain and 7x rack-scale throughput claims over Maia 200 setting the bar Microsoft must clear to credibly run its own models on homegrown chips.
Ask SkimNews
