Microsoft and Intel reunite to build custom AI chips
Back in 2021, Intel made a dramatic strategic change to its business. Faced with the twin threats of companies such as Apple designing its own custom silicon, and making use of gun-for-hire foundries such as Taiwan-based TSMC to produce it, the company decided it could no longer simply design and produce its own wares.
CEO Pat Gelsinger made the dramatic decision to separate the two parts of the company, announcing that it would be opening up its fabs to anyone who needs chips manufacturing, just like its rival.
Two years on, the company has received perhaps the strongest signal yet that the pivot is paying off, as the company has inked a deal to produce custom AI chips with long-time partner Microsoft, thought to be worth around $15 billion.
The new chips are reportedly taking advantage of Intel’s 18A manufacturing process, which features a new way of building transistors, and a new method of moving electricity around inside chips, resulting in greater efficiencies.
“I don’t think it’s a big surprise that there’s [another] hyperscaler designing its own CPUs,” said Philip Kaye, co-founder of data-centre infrastructure firm Vesper Technologies. “When they have specific workloads and they’re designing their accelerator to match their workloads, then they can do something very specific with that GPU or CPU.”
From Micro
