NEWS · COMPANIES · #1874
Microsoft announces Maia 300 AI inference chip, to roll out in 2027
Microsoft announced its third‑generation in‑house AI chip, Maia 300, which the company says will become its primary AI inference system and begin rolling out to production workloads in 2027. Built on TSMC’s N2P node with a chiplet architecture, Microsoft claims Maia 300 delivers ~40% more tokens per megawatt vs leading GPU systems and up to 7× rack‑scale token throughput vs Maia 200 in a 96‑accelerator configuration.
KEY POINTS
- Microsoft announced its third‑generation in‑house AI chip, Maia 300, which the company says will become its primary AI inference system and begin rolling out to production workloads in 2027.
- Built on TSMC’s N2P node with a chiplet architecture, Microsoft claims Maia 300 delivers ~40% more tokens per megawatt vs leading GPU systems and up to 7× rack‑scale token throughput vs Maia 200 in a 96‑accelerator configuration.
- If Microsoft scales Maia 300 across multi‑gigawatt Azure deployments it could materially reduce reliance on Nvidia/AMD GPUs and lower cloud AI inference costs, affecting cloud economics and vendor competition.
DECISION BRIEF
CONFIDENCE
WHAT CHANGED
Microsoft announced Maia 300, its third‑generation in‑house AI inference chip, and said it will begin rolling Maia 300 into production AI workloads in 2027. Maia 300 uses a chiplet architecture on TSMC N2P and is described by Microsoft as its future primary AI inference system, with efficiency and rack‑scale throughput improvements over Maia 200 and leading GPU systems.
WHY NOW
Microsoft states it will scale the Azure Maia family to multi‑gigawatt deployments starting in 2027. If realized, the company says Maia 300’s claimed efficiency gains would change the basis of Microsoft’s inference infrastructure and could reduce reliance on third‑party GPUs.
WHO IS AFFECTED
Directly affected parties named in the sources: Microsoft’s Azure infrastructure and Microsoft products/services (examples given: Microsoft Copilot, GitHub Copilot, Microsoft Foundry, Microsoft’s own AI models); GPU suppliers Microsoft currently uses (Nvidia and AMD) are identified as potential sources of reduced reliance.
CONFIRMED
All items below are explicitly stated in the supplied excerpts: - Microsoft announced Maia 300, described as its third‑generation in‑house AI inference chip. (2209, 2213) - Microsoft says Maia 300 will start rolling out to production AI workloads in 2027. (2209, 2213) - Microsoft says it will scale the Azure Maia family to multi‑gigawatt deployments starting in 2027. (2209, 2213) - Microsoft claims Maia 300 delivers ~40% more tokens per megawatt versus leading GPU‑based systems. (2209, 2213) - Microsoft claims a 96‑accelerator (96‑chip) rack with Maia 300 delivers up to 7× the rack‑scale token throughput of a Maia 200 rack. (2209, 2213) - Maia 300 is built on TSMC’s N2P process node and uses a chiplet architecture that separates compute, memory controller, and I/O. (2209) - Maia 200 is already running on production Microsoft AI workloads; sources name Microsoft Copilot, GitHub Copilot, Microsoft Foundry, and Microsoft’s own AI models, and one source specifies Copilot is running in the US, Europe, and Australia. (2209, 2213) - A technical report is referenced and more details were said to be coming at Ignite 2026 (November). (2213)
UNCERTAIN
Missing, ambiguous, or conflicting evidence in the provided excerpts: - Scale and schedule specifics: Microsoft states multi‑gigawatt scaling 'starting in 2027', but excerpts do not provide a detailed deployment timeline, per‑region schedule, or measured rollout pace. - Independent verification: performance and efficiency claims (40% more tokens per megawatt; up to 7× rack throughput) are Microsoft’s statements in the excerpts; no independent benchmarks or third‑party measurements are provided. - Conflicting or context‑dependent metrics: one excerpt reports Microsoft saying Maia (context unclear) 'pushes up to 1.7× more tokens per watt in Teams' while other excerpts state '≈40% more tokens per megawatt' for Maia 300; the excerpts do not reconcile these different metrics or contexts. (2209, 2213) - Customer and model scope: excerpts name Microsoft products running Maia 200 and say Maia 300 will be the primary inference system for Microsoft products/services, but do not list which external or third‑party customers, partner models, or specific commercial offerings will move to Maia 300 or when. - Operational impact on third‑party GPU suppliers (Nvidia/AMD): sources state Microsoft may rely more on its own chips, but do not quantify reductions in GPU procurement or contract changes.
WHAT TO WATCH
Concrete, observable signals to monitor (as referenced or implied in the excerpts): - Publication of the referenced technical report and the Ignite 2026 disclosures (the Decoder excerpt says more details will come at Ignite in November). (2213) - Microsoft statements or Azure notices that provide specifics about the timing, geographic scope, and capacity targets for the 'multi‑gigawatt' Maia deployments starting in 2027. (2209, 2213) - Microsoft‑published or third‑party benchmark data that compare Maia 300 performance and energy efficiency to leading GPU systems and to Maia 200 (to assess the 40% and 7× claims). (2209, 2213) - Product/service migration announcements or case studies showing Microsoft Copilot, GitHub Copilot, Foundry, or other services moving from Maia 200/GPU infrastructure to Maia 300 in production. (2209, 2213) - Any public statements from Nvidia, AMD, or cloud customers referencing Microsoft’s shift to Maia chips or changes in GPU procurement (sources indicate potential reduced reliance on third‑party GPUs but give no procurement details). (2209)
WHY IT MATTERS
If Microsoft scales Maia 300 across multi‑gigawatt Azure deployments it could materially reduce reliance on Nvidia/AMD GPUs and lower cloud AI inference costs, affecting cloud economics and vendor competition.
EVIDENCE MAP
4Editorial claims linked to specific sources, with support, contradiction and context shown separately.
Microsoft announced the Maia 300 third‑generation in‑house AI chip and said it will start rolling out to production AI workloads in 2027.
SUPPORTEDChecked 2026-10-12 · 2 supporting
The Verge excerpt states Maia 300 is Microsoft’s third‑generation in‑house AI chip and will start rolling out to production in 2027.
SUPPORTS The DecoderMEDIA · 2026-10-12The Decoder excerpt also states Maia 300 ships in 2027 and targets production workloads.
Microsoft said it expects Maia 300 to become the company’s primary AI inference system and to be scaled to multi‑gigawatt (multi‑GW) Azure deployments starting in 2027.
SUPPORTEDChecked 2026-10-12 · 2 supporting
The Verge quotes Scott Guthrie saying Maia 300 is designed to be the 'primary AI inference system' and that Microsoft will scale Azure Maia to multi‑gigawatt deployments starting in 2027.
SUPPORTS The DecoderMEDIA · 2026-10-12The Decoder excerpt states Microsoft plans to deploy Azure Maia chips at multi‑gigawatt scale starting in 2027 and that they will become the company's main inference system.
Microsoft described Maia 300 as using a chiplet architecture built on TSMC’s N2P process node, separating compute, memory controller, and I/O functions.
SUPPORTEDChecked 2026-10-12 · 1 supporting
Microsoft claims Maia 300 delivers about 40% more tokens per megawatt than leading GPU‑based systems.
SUPPORTEDChecked 2026-10-12 · 2 supporting
The Verge quotes Scott Guthrie saying Maia 300 is expected to deliver 40% more tokens per megawatt versus leading GPU‑based systems.
SUPPORTS The DecoderMEDIA · 2026-10-12The Decoder excerpt repeats Microsoft’s claim of 40% more tokens per megawatt versus leading GPU systems.
SOURCES & TIMELINE
2The new Maia 300 chip will start rolling out to production AI workloads in 2027. Microsoft is announcing its third-generation Maia in-house AI chip today, designed to be the “primary AI inference system” for Microsoft’s products and services. Maia 300 not only improves the cost and efficiency of running AI workloads in the cloud, it’s also more capable of handling the latest reasoning models and agentic tasks. Maia…
Microsoft plans to deploy its Azure Maia AI chips at multi-gigawatt scale starting in 2027. They'll become the company's main inference system across its products, executive Scott Guthrie writes . Maia 200 already runs Microsoft Copilot in the US, Europe, and Australia. GitHub Copilot and Foundry follow by year's end. CEO Satya Nadella calls it "our biggest bet yet on our own silicon." The chip handles OpenAI, MAI, a…