For years, Microsoft has been paying top dollar for Nvidia’s processors. Now, the company is quietly placing massive bets on its own silicon, hoping to finally break free from the supply chain bottleneck.
Building your own hardware is always easy on paper. In the real world, it is a massive headache. Microsoft has been trying to build its own AI chips for a while now to stop relying on expensive outside parts. Now, they are finally getting ready to show the world what they have.
According to The Information, Microsoft plans to unveil its new Maia 300 AI chip this fall. People with direct knowledge of the plans say it could happen as soon as next month.
This is a big deal because the company is currently lagging behind its biggest rivals. Microsoft introduced its first Maia chip back in November 2023. But since then, Alphabet and Amazon have moved much faster. Google is already making real money from direct sales of its custom Tensor chips. Amazon is seeing growing adoption for its Trainium processors.
Meanwhile, Microsoft is still heavily reliant on Nvidia’s costly chips to keep its AI running.
To catch up, the company is having serious talks with chipmaker TSMC. They are trying to secure manufacturing capacity for more than 300,000 units of the new Maia 300 chip, with delivery set for 2027. The ultimate goal is to lock down space for more than 1 million chips.
SEE ALSO: Meta just dropped a new AI model, and Mark Zuckerberg is giving the blueprints away
But the physical world rarely cooperates with corporate goals. Details around component supplies remain uncertain, and ongoing capacity negotiations with TSMC could easily slow down those massive production plans. Microsoft and TSMC did not immediately respond to requests for comment outside regular business hours.
The new chip has to be fast. The second-generation Maia 200, which came out in January, was built using TSMC’s 3-nanometer technology. It was packed with a heavy amount of SRAM memory to give the system a speed advantage when handling huge numbers of user requests at once.
Now the real test begins. Microsoft does not just need to build a million chips. They have to convince major cloud customers, like Anthropic, to actually switch over and use them.
If the big clients stay with Nvidia, all those factory negotiations won’t mean much.
