Microsoft made waves this week by unveiling its first ever custom-designed chips at the Ignite conference. The tech giant introduced two new processors: the Maia 100 chip for artificial intelligence workloads and the Cobalt 100 chip for general computing purposes. These new silicon offerings have the potential to shake up the chip industry and cloud computing markets.
The Maia 100 is Microsoft’s answer to the AI accelerators from rivals like Nvidia and Amazon. It is tailored to boost performance for AI tasks like natural language processing. During Ignite, Microsoft demonstrated Maia handling queries for its Bing search engine, powering the Copilot coding assistant, and running large OpenAI language models.
Microsoft has been collaborating closely with OpenAI and is a major investor in the AI research company. OpenAI’s popular ChatGPT was trained on Azure using Nvidia GPUs. By designing its own chip, Microsoft aims to reduce reliance on third-party silicon for AI workloads.
Though performance details remain unclear, Microsoft stated that Maia handles AI tasks with high throughput and low latency. It emphasized efficiency as a key design goal. The chip was engineered in close consultation with Microsoft’s internal AI teams to ensure it fits their requirements.
Microsoft has created novel liquid cooling technology called Sidekicks to work alongside Maia server racks. This advanced thermal management unlocks Maia’s full processing capacity while avoiding the overheating issues that often plague GPU-powered data centers.
When available on Azure, Maia will provide customers access to specialized AI hardware on demand instead of buying dedicated GPUs. Microsoft did not provide a timeline for Maia’s availability or pricing. But offering it as a cloud service instead of a physical product sets Maia apart from AI chips from Intel, Nvidia and others.
The second new chip announced at Ignite was the Cobalt 100 ARM-based processor for general computing. It is expected to deliver a 40% performance boost over existing Azure ARM chips from Ampere.
Microsoft believes Cobalt will provide a compelling alternative to Intel’s server CPUs for cloud workloads. Companies like Amazon have already demonstrated success in cloud data centers by transitioning from Intel to custom ARM chips.
Virtual machines powered by Cobalt will become available on Azure in 2024. Microsoft is currently testing it for key services like Teams and Azure SQL database. More efficient ARM servers can translate to lower costs for cloud customers.
The Cobalt announcement highlights Microsoft’s growing reliance on ARM architecture across its cloud infrastructure. ARM chips are known for power efficiency in mobile devices, but companies like Amazon, Microsoft and Apple now recognize their benefits for data centers too.
By designing its own server-class ARM processor, Microsoft can optimize performance and features specifically for its cloud services. With both Maia and Cobalt, Microsoft aims to give Azure a competitive edge over rivals like AWS and Google Cloud.
Microsoft has lagged behind in cloud infrastructure market share, but introducing unique silicon could help close the gap. Its vertically integrated approach produces chips tailor-made for AI and its cloud platform. With demand for AI compute and cloud services booming, Microsoft’s gambit on custom chips could soon pay dividends.