Microsoft and OpenAI are reportedly working on a massive datacenter to house an AI-focused supercomputer featuring millions of GPUs. The Information reports that the project could cost "in excess of $115 billion" and that the supercomputer, currently dubbed "Stargate" inside OpenAI, would be U.S.-based.
The report says that Microsoft would foot the bill for the datacenter, which could be "100 times more costly" than some of the biggest operating centers today. Stargate would be the largest in a string of datacenter projects the two companies hope to build in the next six years, and executives hope to have it running by 2028.
OpenAI and Microsoft are building these supercomputers in phases, the report says, and that Stargate would be a phase 5 system. A phase 4 system would cost less and may launch as soon as 2026, The Information's sources say, and may be looking to start in Mt. Pleasant, Wisconsin. The system could require several Stargate could need so much power ("at least several gigawatts") that Microsoft and OpenAI are considering alternative sources of power, such as nuclear.
Sources suggested a datacenter of this scale would be challenging, partially because the existing designs require "putting many more GPUs into a single rack than Microsoft is used to, to increase the chips' efficiency and performance." That means also devising novel ways to keep everything cool.
It sounds like the companies are also potentially using this phase of design to move away from reliance on Nvidia. The report claims that OpenAI wants to avoid using Nvidia's InfiniBand cables in Stargate, even though Microsoft uses them in current projects. OpenAI claims it would rather use Ethernet cables.
Much is still to be determined, so it seems like the price and plans could all change and it's unclear when details will be finalized. The Information also states that it has yet to be determined where this computer will be located, and whether it will be built in a single datacenter or in "multiple datacenters in close proximity."
Earlier this year, reports stated that OpenAI CEO Sam Altman had ambitions to build AI chips and was looking to raise as much as $7 trillion to build foundries to produce them. Last year, Microsoft revealed its 128-core Arm datacenter CPU and Maia 100 GPUs specifically for AI projects. There have also been reports of Microsoft developing its own networking gear for AI datacenters. As AI has taken off, Nvidia's GPUs have been in high demand, so it makes sense that companies like Microsoft and OpenAI might want to have some other options.
"We are always planning for the next generation of infrastructure innovations needed to continue pushing the frontier of AI capability," Microsoft chief communications officer Frank Shaw told The Information, though he apparently did not comment directly on the supercomputing plans.
Microsoft has poured billions of dollars into its partnership with OpenAI, largely in the form of computing power to run its models. If Stargate or something like it comes to pass, the partnership will only grow deeper as the investments get larger — and more complicated.
