Microsoft wants to mainly use its own AI data center chips in the future

Microsoft would similar to chiefly usage its ain chips successful its information centers successful the future, the tech giant's main exertion serviceman said connected Wednesday, successful a determination which could trim its reliance connected large players similar Nvidia and AMD.

Semiconductors and the servers that beryllium wrong information centers person underpinned the improvement of artificial quality models and applications.

Tech elephantine Nvidia has dominated the abstraction truthful acold with its graphics processing portion (GPUs), portion rival AMD has a smaller portion of the pie.

But large unreality computing players, including Microsoft, person besides designed their ain customized chips for specifically for information centers.

Kevin Scott, main exertion serviceman astatine Microsoft, laid retired the company's strategy astir chips for AI during a fireside chat astatine Italian Tech Week that was moderated by CNBC.

Microsoft chiefly uses chips from Nvidia and AMD successful its ain information centers. The absorption has been connected picking the close silicon — different shorthand word for semiconductor — that offers "the champion terms performance" per chip.

"We're not spiritual astir what the chips are. And ... that has meant the champion terms show solution has been Nvidia for years and years now," Scott said. "We we volition virtually entertain thing successful bid to guarantee that we've got capable capableness to conscionable this demand."

At the aforesaid time, Microsoft has been utilizing immoderate of its ain chips.

In 2023, Microsoft launched the Azure Maia AI Accelerator which is designed for AI workloads, arsenic good arsenic the Cobalt CPU. In addition, the steadfast is reportedly moving connected its next-generation of semiconductor products. Last week, the U.S. exertion elephantine unveiled caller cooling exertion utilizing "microfluids" to lick the contented of overheating chips.

When asked if the longer word program is to person chiefly Microsoft chips successful the firm's ain information centers, Scott said: "Absolutely," adding that the institution is utilizing "lots of Microsoft" silicon close now.

The absorption connected chips is portion of a strategy to yet plan an full strategy that goes into the information center, Scott said.

"It's astir the full strategy design. It's the networks and the cooling and you privation to beryllium capable to person the state to marque the decisions that you request to marque successful bid to truly optimize your compute to the workload," Scott said.

Microsoft and its rivals Google and Amazon are designing their ain chips to not lone trim reliance connected Nvidia and AMD, but besides to marque their products much businesslike for their circumstantial requirements.

Tech giants including Meta, Amazon, and Alphabet and Microsoft person committed to much than $300 cardinal of superior expenditures this year, with overmuch of that focused connected AI investments arsenic they look to fulfill booming request for AI.

Scott flagged that determination is inactive a shortage of computing capacity.

"[A] monolithic crunch [in compute] is astir apt an understatement," Scott said. "I deliberation we person been successful a mode wherever it's been astir intolerable to physique capableness accelerated capable since ChatGPT ... launched."

Microsoft has been gathering capableness done information centers but it's inactive not capable to conscionable demand, the CTO warned.

"Even our astir ambitious forecasts are conscionable turning retired to beryllium insufficient connected a regular basis. And truthful ... we deployed an unthinkable magnitude of capableness implicit the past twelvemonth and it volition beryllium adjacent much implicit the coming fistful of years," Scott said.