Observed Signal · Oct 12, 2026 · Technical Release · Source: CNBC Technology · Impact: 4/5 · Sentiment: Positive
Microsoft to shift AI inference to homegrown Maia chips
Microsoft announced its third-generation Maia 300 AI accelerator, designed for inference workloads like long-context reasoning and AI agents, to become the primary architecture for AI inference, reducing reliance on Nvidia GPUs. Built with TSMC's 2nm process, each chip features 288GB of HBM4 memory and 2.5x the bandwidth of the Maia 200. A rack of 96 Maia 300 accelerators consumes 167kW, provides 27.6TB of HBM memory, and delivers up to seven times more tokens than a Maia 200 rack, with 40% more tokens per megawatt than GPU systems, though independent benchmarks are pending. Microsoft plans to deploy about one million Maia 300 accelerators to power its AI services, with multi-gigawatt deployments starting in 2027, targeting 300,000+ chips from TSMC. Maia 200 already powers Microsoft Copilot in Iowa, Europe, and Australia. Nvidia remains crucial for training and specific devices, with analysts estimating Maia at 10% of chip spending by Q1 2027. This mirrors hyperscaler trends toward custom silicon.
Microsoft's strategic shift to its own Maia AI chips signals a major move away from Nvidia dominance, impacting the AI infrastructure ecosystem and potentially influencing cost structures and supply chain dynamics across the industry.
MARKET SIGNALS
Track Microsoft Signals & Strategic Shifts
Polaris7 connects regulatory filings, primary sources, leadership changes, and M&A activity into structured intelligence. Set up your Polaris7 workspace to monitor relevant companies in your executive briefing.
- Verified primary sources
- Automated briefings
- Instant watchlist setup
Key Takeaways & Evidence Grounding
- Microsoft will make Maia 300 the primary architecture for AI inference and plans to deploy about one million of them.
- Maia 300 is a 2nm chip with 288GB HBM4 memory and 2.5x bandwidth of Maia 200.
- Maia 300 racks deliver up to 7x more tokens than Maia 200 racks and 40% more tokens per megawatt than GPUs (independent benchmarks pending).
- Microsoft plans multi-gigawatt deployments starting in 2027 and has asked TSMC for capacity for over 300,000 Maia 300 chips.
- Maia 200 powers Microsoft Copilot and is deployed in Iowa, Europe, and Australia.
Connected Companies & Entities
“Microsoft said it will eventually have multiple gigawatts worth of its Maia AI chips and intends to make Maia its primary architecture for r...”
“Microsoft will ramp up use of its homegrown AI chips after years of reliance on Nvidia's graphics processing units.”
“Microsoft's greatest strength in AI has been its Azure cloud... [and] the company develops in-house models, though they’ve yet to make a big...”
“Nvidia's GPUs have been the standard for AI workloads and models, and they fill Microsoft’s existing data centers, while also serving to tra...”
“As demand keeps growing, the model leaders have embraced a variety of silicon from companies including Amazon, Advanced Micro Devices, Cereb...”
“As demand keeps growing, the model leaders have embraced a variety of silicon from companies including Amazon, Advanced Micro Devices, Cereb...”
“As demand keeps growing, the model leaders have embraced a variety of silicon from companies including Amazon, Advanced Micro Devices, Cereb...”
“For general-purpose use, Microsoft already taps its own central processing units as well as silicon from Intel and AMD.”
“The compute chiplets are made with TSMC’s 2-nanometer N2P process.”
“In a note distributed to clients on Wednesday, Evercore ISI analysts estimated that Maia will represent about 10% of Microsoft’s spending on...”
“OpenAI is bringing its first inference chip, Jalapeño, developed with Broadcom.”
“Nvidia's $20 billion deal with the chip start-up Groq.”
Ontology Mapping & Concepts
Related Market Signals & Shifts
Recent verified developments and strategic activity across this market segment.
Microsoft Unveils Maia 200 Inference Accelerator
ChipStrat interviewed Saurabh Dighe, CVP of Azure Systems and Architecture at Microsoft, about Maia 200 — Microsoft’s second-generation AI accelerator designed specifically to optimize inference economics. Maia 200 targets performance-per-dollar and performance-per-watt for inference, with architectural trade-offs that favor inference workloads over training. Key technical choices discussed include a much larger on-die SRAM, a memory hierarchy balancing SRAM, HBM and system DRAM, and a preference for a large Ethernet-based scale-up domain with a custom transport layer. Microsoft positions Maia 200 as complementary to merchant GPUs within a heterogeneous fleet, exposing capacity through Azure services rather than as a standalone product. The interview also emphasizes KV cache management for long-context workloads and the importance of software investments (compilers, kernels, pre-silicon tooling) ahead of silicon.
AI Hardware Stack Rebuilt from the Wafer Up
The article explains that modern AI accelerators rely on a constrained hardware stack beginning at wafer fabrication and advanced EUV lithography. TSMC (72% share) and ASML (EUV machines) are central bottlenecks, but the immediate chokepoint is CoWoS packaging for stacking HBM, capacity for which is sold out through 2026. TSMC plans $52–56 billion capex in 2026, yet wafer demand for AI accelerators is projected to rise 11x from 2022–2026. The piece argues GPUs (e.g., NVIDIA H100/B200) are optimized for training and often over-provisioned for latency-sensitive inference. It highlights Cerebras’ wafer-scale WSE-3 (trillions of transistors, massive on-die bandwidth) and cited benchmarks showing material inference throughput and cost advantages versus NVIDIA B200. The article notes OpenAI signed a $20B+ agreement with Cerebras for large-scale inference capacity and recommends builders benchmark their own workloads on emerging inference hardware.
Nvidia Invests $3.5B in MediaTek for Custom AI Chips
Nvidia has announced a $3.5 billion investment in convertible bonds issued by Taiwanese chipmaker MediaTek, a major rival to Qualcomm. Under the partnership, MediaTek will integrate Nvidia's NVLink Fusion ecosystem into its custom AI chip business, enabling third-party chips to connect directly to Nvidia-based data center architectures. This move counters the trend of hyperscalers like Amazon, Google, and Microsoft developing proprietary silicon, helping Nvidia maintain its dominance in data centers. The collaboration also extends to local AI computing for PCs and cars, building on existing projects like DGX Spark and RTX Spark. MediaTek shares closed 10% higher after the announcement. The deal strengthens MediaTek's push into data centers, positioning it as a potential challenger to Broadcom. MediaTek expects its custom AI chip business to generate $2 billion in revenue this year, with a total addressable market of up to $80 billion by 2027.
WATCHLIST & BRIEFINGS
Receive future signals for Microsoft & competitors in your briefing
Stay ahead of material M&A deals, leadership changes, account wins, and product rollouts. Polaris7 connects primary filings, company newsrooms, and trade reporting directly into your executive briefing.
- Regular primary-source checks
- Structured Watchlist briefings
- Preserves current context
Market Signal Monitoring
Follow affected companies in Polaris7 with regular source checks and structured Watchlist briefings.
Monitor Microsoft
Follow affected companies in Polaris7 with regular source checks and structured Watchlist briefings.
