NVIDIA Wants to Help Companies Build Custom AI Chips

According to a Reuters exclusive, NVIDIA has already held talks with mega-players like Amazon, Meta, Microsoft, Google and OpenAI about producing bespoke silicon solutions.

NVIDIA Wants to Help Companies Build Custom AI Chips

NVIDIA is launching a new business unit focused on designing customized processors tailored to major companies’ AI and other workload needs. According to a Reuters exclusive, NVIDIA has already held talks with mega-players like Amazon, Meta, Microsoft, Google and OpenAI about producing bespoke silicon solutions.

The new NVIDIA unit will be led by semiconductor veteran Dina McKinney. Previously Vice President responsible for AMD’s CPU design and Marvell’s infrastructure processors, McKinney will oversee teams building custom chips for cloud computing, 5G telecom, gaming, automotive and more.

NVIDIA's dominance in the AI chip market has been a key driver behind its soaring market value, which has seen an impressive 40% increase this year alone, reaching $1.73 trillion. They have attracted heavyweights like OpenAI, Microsoft, Alphabet, and Meta Platforms, who all leverage NVIDIA's H100 and A100 chips for their generative AI endeavors.

However, while NVIDIA GPUs dominate AI training and high-performance computing in the cloud today, hyperscale providers are increasingly rolling out their own custom designs too. For certain workloads, homegrown chipsets allow more optimization around cost, capabilities and power needs at scale.

In fact, Amazon, Google and Microsoft already deploy custom silicon alongside NVIDIA processors in their datacenters. Letting these big customers instead tap NVIDIA's IP directly for bespoke solutions could defend against further in-house encroachment on NVIDIA's datacenter stronghold.

“If you’re really trying to optimize on things like power, or optimize on cost for your application, you can’t afford to go drop an H100 or A100 in there,” said Greg Reichow of Eclipse Ventures. “You want to have the exact right mixture of compute.”

The custom chip market, particularly for data centers, is poised for rapid growth. According to 650 Group’s Alan Weckel, the data center custom chip market could reach $10 billion this year and double by 2025. The broader custom chip market was worth approximately $30 billion in 2023, representing a significant slice of the global chip sales pie​​.

NVIDIA's push builds on previous moves to open up its networking technology for integration into customer’s custom designs last year. While scant details have emerged since, the conversations with hyperscalers suggests it is serious about becoming a one-stop shop for tailored silicon across data center, AI, 5G, automotive and other explosive growth markets.

As more cloud giants look to take chip design in-house, the company will have its work cut out maintaining its grip on the accelerating AI chip market.

Let’s stay in touch. Get the latest AI news from Maginative in your inbox.

Subscribe
Mastodon