At its annual developer conference, Ignite, Microsoft on Wednesday unveiled the long-anticipated custom cloud computing chip for its Azure cloud service, called Azure Maia 100, which it said is optimized for tasks such as generative AI.
The Maia 100 is the first in a series of Maia accelerators for AI, the company said. With 105 billion transistors, it is "one of the largest chips on 5-nanometer process technology," said Microsoft, referring to the size of the smallest features of the chip, five billionths of a meter.
In addition, the company introduced its first microprocessor built in-house for cloud computing, the Azure Cobalt 100. Like Maia, the processor is the first in a planned series of microprocessors. It is based on the ARM instruction-set architecture from ARM Holdings that is licensed for use by numerous companies including Nvidia and Apple.
Microsoft said Cobalt 100 is a 64-bit processor that has 128 computing cores on die, and that it achieves a 40% reduction in power consumption compared to other ARM-based chips that Azure has been using. The Cobalt part is already powering programs including Microsoft Teams and Azure SQL, said the company.
The two chips, Maia 100 and Cobalt 100, are fed by 200 gigabit-per-second networking, said Microsoft, and can deliver 12.5 gigabytes per second of data throughput.
Microsoft is the last of the Big Three cloud vendors to offer custom silicon for cloud and AI. Google pioneered the race to custom silicon with its Tensor Processing Unit, or TPU, in 2016. Amazon followed suit with a slew of chips including Graviton, Trainium, and Inferentia.
Rumors of Microsoft's efforts have circulated for years, fed by occasional disclosures such as last summer's leak of a planning document from the company.
Microsoft made a point of noting that it continues to partner with both Nvidia and AMD for chips for Azure. It plans to add Nvidia's latest "Hopper" GPU chip, the H200, next year, as well as AMD's competing GPU, the MI300.
Microsoft's chips will assist with programs such as GitHub Copilot, but they will also be used to run generative AI from AI startup OpenAI, into which Microsoft has poured $11 billion in investment to secure exclusive rights to programs such as ChatGPT and GPT-4.
At OpenAI's developer conference last week in San Francisco, Microsoft CEO Satya Nadella pledged to build "the best compute" for OpenAI "as you aggressively push forward on your roadmap."
Microsoft and OpenAI are both trying simultaneously to lure enterprises to use generative AI. Microsoft is seeing big growth in the generative AI business, Nadella told Wall Street last month. The company's paying customers for its GitHub Copilot software rose by 40% in the September quarter from the prior quarter.
"We have over 1 million paid Copilot users in more than 37,000 organizations that subscribe to Copilot for business," said Nadella, "with significant traction outside the United States."
Also at Ignite, Microsoft announced it is extending Copilot to Azure with a public preview of Copilot for Azure, a tool it said will give system administrators an "AI companion" that will "help generate deep insights instantly."
In addition to the chip innovations, Microsoft announced general availability of Oracle's database programs running on Oracle hardware in the US East Azure region. Microsoft is the only cloud operator to offer Oracle database on Oracle's own computer systems infrastructure, it said.
Other partner news included the general availability of Microsoft's edge computing service, Arc, for VMware's vSphere virtualization suite.