Microsoft unveils third-generation Maia 300 AI chip, with production rollout from 2027
Microsoft announced Maia 300, its third-generation in-house AI accelerator, which it says will become the primary inference system for its products from 2027. It claims 40% more tokens per megawatt than leading GPU-based systems.
1 / 2
The story, neutrally told
Mixed · 2Microsoft announced Maia 300 on Monday 12 October 2026, the third generation of its in-house AI accelerator, designed to be the "primary AI inference system" for its products and services. The VergeLC “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.” Read at The Verge ↗ Anadolu AgencyN “Microsoft on Monday unveiled its third-generation Maia AI accelerator” Read at Anadolu Agency ↗ Mixed · 2Production deployments of Maia 300 are due to begin in 2027, when Microsoft says it will scale its Azure Maia family to multi-gigawatt deployments. Anadolu AgencyN “Production deployments of Maia 300 will begin in 2027.” Read at Anadolu Agency ↗ The VergeLC “We will scale our Azure Maia family to multi-gigawatt deployments starting in 2027” Read at The Verge ↗ Mixed · 2Scott Guthrie, Microsoft's executive vice president of cloud, said the company expects Maia 300 to deliver 40% more tokens per megawatt than leading GPU-based systems, and that a full 96-accelerator configuration delivers up to 7x the rack-scale token throughput of Maia 200. The VergeLC “We expect Maia 300 to deliver 40 percent more tokens per megawatt compared to leading GPU-based systems” Read at The Verge ↗ Anadolu AgencyN “A configuration containing 96 accelerators can deliver up to seven times the rack-level token throughput of Maia 200, according to the company.” Read at Anadolu Agency ↗
Mixed · 2The chip is aimed at workloads such as long-context reasoning, latency-sensitive conversational experiences and sustained multi-step agentic tasks. The VergeLC “Long-context reasoning, latency-sensitive conversational experiences, and sustained, multi-step agentic tasks.” Read at The Verge ↗ Anadolu AgencyN “long-context reasoning, time-sensitive conversations and multistep AI agent tasks” Read at Anadolu Agency ↗ Left · 1The Verge reports that Maia 300 uses a chiplet architecture built on TSMC's N2P process node, separating compute, memory controller and I/O functions, which Microsoft's Azure hardware team says increases silicon area dedicated to compute. The VergeLC “Built on TSMC’s N2P process node, the latest Maia 300 architecture separates compute, memory controller, and I/O functions.” Read at The Verge ↗ Mixed · 2The Verge says the efficiency gains will let Microsoft rely mainly on its own chips instead of those from Nvidia and AMD; Anadolu's report does not mention the rival chipmakers. The VergeLC “These efficiency improvements will allow Microsoft to rely mainly on its in-house AI chips soon, instead of ones from Nvidia and AMD.” Read at The Verge ↗ Anadolu AgencyN “Inference is the process through which trained artificial intelligence models generate responses to requests.” Read at Anadolu Agency ↗
Mixed · 2Maia 200, announced in January, already runs production workloads including Microsoft Copilot; the outlets differ on whether GitHub Copilot and Microsoft Foundry already run on it or will follow by year-end. The VergeLC “Maia 200 is already running on production AI workloads across Microsoft, including Microsoft Copilot, GitHub Copilot, Microsoft Foundry, and Microsoft’s own AI models.” Read at The Verge ↗ Anadolu AgencyN “Its predecessor, Maia 200, already supports production workloads including Microsoft Copilot, with GitHub Copilot and Microsoft Foundry set to follow by year-end.” Read at Anadolu Agency ↗ Left · 1Microsoft AI CEO Mustafa Suleyman said the company needs "a platform that we own" and co-designs with its models, and that running Microsoft AI models on Maia is a critical focus; The Verge links the chip to Microsoft's goal of catching up with Google DeepMind, OpenAI and Anthropic. The VergeLC “At that scale we need a platform that we own, and that we co-design with our models.”“could help the company achieve its goal of catching up with Google DeepMind, OpenAI, and Anthropic” Read at The Verge ↗
Every sentence links to the reporting it rests on. The pill in front of each says where its sources sit: Left, Centre or Right when one side supplies at least half of them, Mixed when they are evenly split. The number is how many outlets it cites.
Left1 outlet
- Framing
- Frames the launch as a competitive move that reduces Microsoft's dependence on Nvidia and AMD and supports its AI model ambitions.
- Emphasis
- Reliance on Nvidia, chip architecture detail (TSMC N2P chiplets), and Mustafa Suleyman's comments on owning the platform and catching up with rivals.
- Leaves out or plays down
- Does not explain what inference is; efficiency figures are presented as Microsoft's claims without outside assessment.
- Charged language
- “less reliant on Nvidia’s GPUs”
- For example
-
“These efficiency improvements will allow Microsoft to rely mainly on its in-house AI chips soon, instead of ones from Nvidia and AMD.” — The Verge
Centre1 outlet
- Framing
- Short, product-announcement style report centred on the rollout timeline and performance figures.
- Emphasis
- 2027 rollout, 40% tokens-per-megawatt claim, 7x throughput versus Maia 200, and a plain definition of inference.
- Leaves out or plays down
- Omits Nvidia and AMD, the chip's manufacturing process and the Microsoft AI model strategy.
- For example
-
“Microsoft expects Maia 300 to deliver 40% more tokens per megawatt than leading graphics processing unit-based systems.” — Anadolu Agency
Right0 outlets
No right outlet in our sources has covered this story yet.
What every side reports
- Microsoft unveiled the third-generation Maia 300 AI accelerator on 12 October 2026.
- Production deployments begin in 2027, with Maia meant to become the primary inference system for Microsoft products.
- Microsoft expects 40% more tokens per megawatt than leading GPU-based systems.
- A 96-accelerator configuration is said to deliver up to 7x the rack-level token throughput of Maia 200.
- Figures are Microsoft's own projections; no independent verification is reported.
Where accounts differ
-
Whether GitHub Copilot and Microsoft Foundry already run on Maia 200
- Left
- The Verge says Maia 200 already runs Microsoft Copilot, GitHub Copilot, Foundry and Microsoft's own models.
- Centre
- Anadolu says Maia 200 supports Microsoft Copilot, with GitHub Copilot and Foundry set to follow by year-end.
Microsoft organisation
Presents Maia 300 as a complete AI system for reasoning and agentic workloads, to be the primary inference system for its products from 2027, backed by its own efficiency projections.
“Maia 300 is a complete AI system designed for the new types of AI workloads emerging today” — The Verge
“The company plans to scale its Azure Maia family to multi-gigawatt deployments” — Anadolu Agency
Left1 article
-
Microsoft’s new AI chip makes it less reliant on Nvidia’s GPUs
Neutral Tech-industry analysis stressing Microsoft's reduced dependence on Nvidia and its AI model ambitions, with technical detail and executive quotes.
Centre1 article
-
Microsoft unveils Maia 300 AI chip, plans expanded rollout from 2027
Neutral Concise summary of the announcement, rollout timing and Microsoft's performance claims.
Right0 articles
No coverage yet.
- 12 Oct 14:00 First The VergeLC Microsoft’s new AI chip makes it less reliant on Nvidia’s GPUs
- 12 Oct 18:03 +4h 3m Anadolu AgencyN Microsoft unveils Maia 300 AI chip, plans expanded rollout from 2027
Times are when each article was published, or when we first saw it if the outlet gave no time.
Ask about this story
Loading…