Microsoft is preparing to take another step in its effort to build more of the artificial intelligence stack in-house, with reports suggesting it could introduce its next Maia chip as soon as September. The move would mark another push by the software giant to reduce its reliance on Nvidia, which still dominates the market for AI accelerators.
According to industry reports, Microsoft has spent years trying to turn its own silicon program into a meaningful part of Azure, but the harder challenge has been moving from prototype to scale. That is now starting to change. The company is said to be discussing manufacturing capacity with Taiwan Semiconductor Manufacturing Co. for more than 300,000 Maia chips that would be delivered in 2027, a sign that Microsoft wants the program to become more than an internal experiment.
Microsoft’s recent Maia 200 chip offers a preview of what that strategy could look like. The company has said the accelerator is built on TSMC’s 3-nanometer process and is designed for inference workloads, the kind of day-to-day processing used by AI services such as chatbots and productivity tools. Microsoft has also claimed performance gains over its earlier Maia hardware, along with better efficiency metrics, while positioning the chip as part of a broader effort to make large-scale AI infrastructure cheaper and less energy intensive.
The strategic ambition goes beyond saving money on internal computing. Reports say Microsoft also wants major cloud customers, including Anthropic, to run workloads on its chips. If that effort succeeds, Azure could gain more control over one of the costliest layers in AI infrastructure and strengthen its position against rivals building their own custom silicon, including Amazon and Google. Meta is also moving ahead with its own chip plans, underscoring how major technology companies are trying to loosen Nvidia’s grip on the AI market.
Even so, Nvidia remains the centerpiece of the current ecosystem, and dislodging it will be difficult. But Microsoft’s scale gives it room to chip away at that dependence, and the expected September unveiling will be watched closely as a test of how serious the company is about turning its custom silicon effort into a lasting competitive advantage.
Disclaimer: This article may have been created with AI assistance and reviewed by our editorial team. It is provided for general informational purposes only. Readers should verify information independently before relying on this content.
Sources by paragraph:
- Paragraph 1: [2], [4]
- Paragraph 2: [2], [6], [7]
- Paragraph 3: [2], [3], [4]
- Paragraph 4: [5], [6], [7]
- Paragraph 5: [1], [2], [4]
Bibliography
- https://finance.yahoo.com/technology/ai/articles/microsofts-nvidia-alternative-getting-muc... - Please view link - unable to able to access data
- https://www.tomshardware.com/pc-components/cpus/microsoft-introduces-newest-in-house-ai-ch... -
Microsoft has unveiled its latest in-house AI accelerator, the Azure
Maia 200, designed to outperform custom chips from competitors like
Amazon and Google. Fabricated using TSMC’s 3nm process and featuring 140
billion transistors, the chip supports up to 10.14 petaflops of FP4
compute power, surpassing Amazon’s Trainium3 and matching some aspects
of Nvidia’s Blackwell B300 Ultra, while using far less power. It
includes 216GB of HBM3e memory with 7TB/s bandwidth and 272MB of
advanced on-die SRAM, optimized for efficient data processing. Though
not commercially available, the Maia 200 is positioned within
Microsoft’s Azure infrastructure for AI inferencing workloads. Microsoft
claims a 30% improvement in performance-per-dollar over its
predecessor, the Maia 100, despite a higher thermal design power (TDP).
The chip was initially delayed but is now in operation at the US Central
Azure data center, with more deployments expected. Microsoft's
messaging emphasizes environmental efficiency and social responsibility
amid growing scrutiny over AI’s ecological impact. Despite the Maia
200’s limited release scope, it represents a significant step in
Microsoft’s pursuit of high-performance, environmentally-conscious AI
infrastructure. - https://www.techradar.com/pro/microsoft-unveils-maia-200-its-powerhouse-accelerator-lookin... -
Microsoft has revealed Maia 200, its 'next major milestone' in
supporting the next generation of AI and inference technology. The
company's new hardware, the successor to the Maia 100, will
'dramatically shift the economics of largescale AI,' offering a
significant upgrade in terms of performance and efficiency as it to
stake a claim in the market. The launch will also look to push Microsoft
Azure as a great place to run AI models ... . Microsoft says Maia 200
contains over ... . This all contributes to the ability to ... . - https://www.windowscentral.com/microsoft/microsofts-new-ai-silicon-is-here-is-this-what-wi... -
Microsoft has unveiled the Maia 200, a custom AI chip aimed at enhancing
large-scale AI operations and competing with similar offerings from
Amazon and Google. Built using TSMC’s advanced 3nm process, Maia 200
reportedly outperforms Amazon's Trainium and Google’s TPU in
low-precision AI tasks while also delivering greater efficiency.
Microsoft claims the chip offers 30% better performance per dollar than
previous hardware, making it the most efficient inference system it has
deployed. Maia 200 is already operational in Microsoft’s Iowa data
center, with expansion plans set for Arizona. The chip uses water
cooling to boost sustainability, aligning with Microsoft’s
'Community-First' AI infrastructure goals, although previous
environmental impact criticisms persist. This development could
significantly lower the high operational costs of running large language
models (LLMs) like OpenAI’s GPT-5.2. Microsoft says Maia 200 will be
used for inference tasks—processing user queries—which represent ongoing
daily costs for AI services like ChatGPT and Microsoft 365 Copilot. As
OpenAI faces projected losses of up to $14 billion in 2026, Maia 200's
potential to reduce per-token costs may play a crucial role in making AI
services more economically viable by 2027. - https://www.tomshardware.com/tech-industry/semiconductors/open-ai-building-its-own-chip-st... -
OpenAI's long-rumored $10 billion partnership with Broadcom is already
showing cracks. The company is widely understood to be developing a
custom chip designed specifically for OpenAI's inference workloads, but
according to individuals familiar with the matter, the project has "hit
snags": OpenAI wanted more power, sooner, than Broadcom could deliver,
and an internal push to roll the chip out in Q2 2026 has already slipped
to Q3 at the earliest, according to a report from The Information. The
project, which has been kept deliberately quiet, is set to have
manufacturing run through TSMC. Once live, the chip could handle
inference jobs across OpenAI’s growing fleet of data centers, cutting
its exposure to GPU bottlenecks and potentially lowering costs. - https://www.datacenterdynamics.com/en/news/meta-could-start-production-of-iris-ai-chip-in-... -
Meta is planning to start production of its forthcoming AI chip, dubbed
Iris, in September, according to a report from Reuters. Citing details
from an internal memo reviewed by the outlet, Iris is one of the chips
unveiled by the company in March this year, when it detailed the next
four generations of its Meta Training and Inference Accelerator (MTIA)
chip, at the time named the MTIA 300, 400, 450, and 500. At the time,
Meta said all four chips had either already been deployed or were
scheduled for deployment in the next 18 months and would primarily be
used to support generative AI inferencing workloads. Per the memo, the
chip took six weeks to test, and no major issues were found. - https://www.straitstimes.com/business/companies-markets/meta-to-put-its-own-ai-chip-into-p... -
Meta Platforms plans to start manufacturing its own artificial
intelligence chip from September as part of its plan to boost overall
computing power to 14GW in 2027, according to an internal memo reviewed
by Reuters. The tech firm’s data centre chip, code-named “Iris”, is part
of a four-generation project for Meta Training and Inference
Accelerators that it will design in-house. The plan is to use
custom-built silicon to improve the AI that powers its Facebook and
Instagram social media platforms.
