Microsoft unveils Maia 300 AI chip to power internal inference workloads
Microsoft has revealed the Maia 300, its third-generation custom AI processor, aiming to make it the primary inference system across its data centers and reduce reliance on Nvidia hardware.
Microsoft has introduced the Azure Maia 300, marking what Chief Executive Satya Nadella described as the company's biggest bet yet on custom silicon. Designed to power internal products like Copilot as well as partner services, the third-generation chip is intended to serve as Microsoft's primary inference system for handling incoming artificial intelligence requests.[1][8][11][13][14]
The custom hardware push represents a calculated attempt by Microsoft to curb its long-standing dependence on Nvidia GPUs and save on operational costs. Nadella claimed that the Maia 300 delivers 40% more tokens per watt compared to leading GPU-based systems, an efficiency metric aimed at maximizing inference output within power-constrained data centers. The accelerator is built on TSMC's N2P process node and tailored for AI models with heavy memory demands and long contexts, with investor Lark Davis reporting the chip features 288GB of HBM4 memory.[2][3][4][7][8][10][11]
Deployment of the new silicon is planned at substantial scale. According to a Reuters report cited by Techmeme, Maia 300 server racks will house 96 chips and consume roughly 167 kilowatts of power, with multi-gigawatt installations beginning next calendar year. While initial infrastructure scaling starts next year, reports from The Verge and Coin Bureau state that production AI workloads will begin running on Maia 300 in 2027.[4][6][8][9][12]
Key facts
- Microsoft unveiled the Maia 300, its third-generation homegrown AI chip.
- Microsoft plans to scale the Maia line to multi-gigawatt deployments starting next calendar year, per CEO Satya Nadella.
- Nadella claimed the Maia 300 delivers 40% more tokens per watt compared to leading GPU-based systems.
- Microsoft intends for the Maia 300 to serve as its primary inference engine across internal products, including Copilot, reducing dependence on Nvidia GPUs.
- The chip is built on TSMC's N2P process node and includes 288GB of HBM4 memory, according to reports.
- Reuters reported that Maia 300 server racks will each hold 96 chips and use approximately 167 kW of power.
- The Verge reported that the Maia 300 will begin rolling out to production AI workloads in 2027.
Sources · 12 sources
- QU
Quartz@qzPost on X ·
Microsoft is making its homegrown Maia 300 chip the primary engine for AI inference: Starting with the third-generation Maia 300, Microsoft says it will rely on its own silicon to handle incoming AI model requests dlvr.it/TVtPg5 https://t.co/6SjPLF6BrY
Open source - LD
Lark Davis@LarkDavisPost on X ·
Microsoft just dropped details on the Azure Maia 300, and the message is pretty clear: they are tired of paying the Nvidia tax. 2nm tech, 288GB of HBM4, and 40% more tokens per megawatt. Big tech is not just buying GPUs anymore. They are building their own silicon from scratch. MSFT ended +1.14% after the announcement.
Open source - TE
Techmeme@TechmemePost on X ·
Microsoft unveils its third-gen in-house AI chip, the Maia 300, built on TSMC's N2P process node and aimed at serving as its "primary AI inference system" (@tomwarren / The Verge) (Visit Techmeme dot com for the link and full context!)
Open source - CB
Coin Bureau@coinbureauPost on X ·
🚨BREAKING: Microsoft unveils Maia 300, its third-gen AI chip, in a landmark move reshaping the company’s long-standing dependence on Nvidia. Production deployments begin in 2027, with Microsoft planning to scale its Maia infrastructure to MULTI-GIGAWATT capacity. The chip is designed for running AI models, particularly workloads involving long contexts and heavy memory use. Microsoft wants Maia to become the primary inference system powering its products and services, bringing more of its massive AI computing needs onto its own silicon.
Open source - QU
Quartz@qzPost on X ·
Microsoft is making its own chip the heart of its AI infrastructure: Microsoft's new Maia 300 accelerator will become the primary engine for AI inference across its data centers — a major step toward breaking its dependence on Nvidia. The chip… dlvr.it/TVtNRF #Microsoft #AI https://t.co/nkI2B5tOp6
Open source - TV
The Verge@vergePost on X ·
The new Maia 300 chip will start rolling out to production AI workloads in 2027. https://t.co/MFSRfAc5H8
Open source - RE
Reuters@ReutersPost on X ·
Microsoft plans huge deployment of its own AI chips, with an eye on saving costs https://t.co/ByDOafYboO https://t.co/ByDOafYboO
Open source - EV
Evan@StockMKTNewzPost on X ·
Microsoft $MSFT CEO Satya Nadella just posted this: “Today we are announcing our biggest bet yet on our own silicon. Starting next calendar year, we will scale our Maia family to multi-gigawatt deployments, making it the primary inference system for our own product and services. And we’re introducing Maia 300, which delivers 40% more tokens per watt compared to leading GPU-based systems.”
Open source - TW
Tom Warren@tomwarrenPost on X ·
Microsoft’s new AI chip makes it less reliant on Nvidia’s GPUs. The new Maia 300 chip will start rolling out to production AI workloads in 2027 https://t.co/ouirRn7IAK
Open source - CN
CNBC@CNBCPost on X ·
Microsoft will ramp up use of its homegrown artificial intelligence chips after years of reliance on Nvidia’s graphics processing units, as the software giant attempts to bolster its AI stack. Starting with its third-generation Maia AI accelerator, unveiled on Monday, Microsoft said it will mainly rely on its own silicon to handle incoming requests. Learn more: https://t.co/V1AnCqQhjl
Open source - SB
Shay Boloor@StockSavvyShayPost on X ·
$MSFT CEO Satya Nadella says Maia its “biggest bet yet on our own silicon” with multi-GW deployments starting next year and Maia 300 delivering 40% more tokens per watt than leading GPU systems. That efficiency lets Microsoft serve more inference from same power capacity as electricity becomes one of biggest constraints on AI growth.
Open source - TE
Techmeme@TechmemePost on X ·
Microsoft says Maia 300 racks will hold 96 chips and use ~167 kW of power and plans to roll out "multi-gigawatt deployments" of the chip starting next year (Reuters) (Visit Techmeme dot com for the link and full context!)
Open source - NE
NeowinFeed@NeowinFeedPost on X ·
Microsoft has officially unveiled Azure Maia 300, its next-generation custom silicon designed to power internal and partner AI workloads at multi-gigawatt scale. #Microsoft #Azure #MAIA300 https://t.co/oKaG9xbuM3
Open source - UW
unusual_whales@unusual_whalesPost on X ·
Microsoft, $MSFT, is cutting back on Nvidia. The company just unveiled Maia 300, its third-generation homegrown AI chip, and says Maia will now be its primary system for running AI across its products, including Copilot. Microsoft claims Maia 300 delivers 40% more AI output for the same power as leading GPU systems. It plans to deploy the chips at multi-gigawatt scale starting next year.
Open source