Microsoft Unveils Maia 200 AI Chip to Challenge Nvidia in Cloud Computing Race
Microsoft has taken a major step in strengthening its artificial intelligence infrastructure with the rollout of its second-generation custom AI processor, the Maia 200 chip. The new accelerator is aimed at reducing the company’s dependence on Nvidia while offering cloud customers a more cost-effective option for running large-scale AI workloads.
The Maia 200 marks Microsoft’s most ambitious attempt yet to build in-house hardware for artificial intelligence. It follows the earlier Maia 100 chip, announced in 2023, which was used internally and never made available to external cloud users.
This time, Microsoft says the strategy is different.
Wider Availability for Cloud Customers
Scott Guthrie, Executive Vice President for Cloud and AI at Microsoft, said the company plans to make Maia 200 accessible to a broader group of customers in the coming months.
In a blog post, Guthrie described the chip as the most efficient inference system the company has ever deployed. Developers, research institutions, and AI laboratories will be able to apply for early access through a preview software development kit.
Microsoft CEO Satya Nadella also confirmed that the new processor is already live on Azure.
“Our newest AI accelerator Maia 200 is now online in Azure. Designed for industry-leading inference efficiency, it delivers 30% better performance per dollar than current systems,” Nadella wrote on X.
Built on Advanced Manufacturing Technology
The Maia 200 is manufactured using Taiwan Semiconductor Manufacturing Company’s advanced 3-nanometer process, placing it among the most technologically advanced chips currently in production.
Each server houses four interconnected Maia 200 chips. Unlike Nvidia’s systems, which rely heavily on InfiniBand networking, Microsoft’s design uses Ethernet connections. This helps lower infrastructure costs and simplifies deployment inside data centres.
The chip features:
- Over 10 petaflops of FP4 computing power
- Around 5 petaflops of FP8 performance
- 216GB of HBM3e high-bandwidth memory
- Memory bandwidth of nearly 7 terabytes per second
These specifications make it suitable for training and running complex AI models at scale.
Powering Copilot and Future AI Models
Some of the first Maia 200 units will be used by Microsoft’s Superintelligence team, led by Microsoft AI CEO Mustafa Suleyman.
The chips will also support key products such as Microsoft Copilot and cloud-hosted models, including OpenAI’s latest systems.
“It’s a big day. Our Superintelligence team will be the first to use Maia 200 as we develop our frontier AI models,” Suleyman said.
Microsoft claims the new chip delivers significantly higher performance than comparable offerings from Amazon Web Services and Google Cloud.
According to Guthrie, Maia 200 offers nearly three times the FP4 performance of Amazon’s latest Trainium processors and outperforms Google’s seventh-generation TPU in FP8 tasks.
Competing in the Hyperscaler Chip Race
Microsoft joins Amazon and Google in developing custom chips to control costs and improve efficiency.
All three companies, known as hyperscalers, operate massive cloud platforms and invest heavily in data centre infrastructure. By designing their own silicon, they aim to reduce reliance on third-party suppliers and better optimise hardware for AI workloads.
The move is also driven by market pressures. Nvidia’s advanced AI chips remain expensive and are often in short supply, creating challenges for cloud providers and startups alike.
Custom processors like Maia 200 allow Microsoft to secure computing capacity while offering more predictable pricing to customers.
Our Thoughts
Microsoft’s launch of the Maia 200 reflects a broader shift in the technology industry, where control over hardware is becoming as important as software innovation.
As artificial intelligence continues to reshape business, healthcare, education, and research, access to affordable and powerful computing will be a key competitive advantage. With Maia 200, Microsoft is signalling that it intends to be a long-term leader not just in AI software, but also in the infrastructure that powers it.
The success of this strategy will depend on how quickly the chip scales across Azure and how well it performs under real-world workload.