Why the CPU is taking center stage

Nvidia’s graphics processing units have been best-selling chips for years, but the sudden arrival of agent artificial intelligence has brought a renaissance to its more modest host chip, the central processing unit.

Now, Nvidia is set to unveil new details about its agent-optimized CPUs at its annual GTC conference, which starts Monday, with CPU-only racks likely to appear on the showroom floor.

“CPUs are becoming the bottleneck in growing this AI and agent workflow,” Dion Harris, Nvidia’s head of AI infrastructure, told CNBC this week, calling it an “exciting opportunity.”

The chip giant announced its first data center CPU, Grace, in 2021, and the next generation, Vera, is now in production. The CPUs are typically deployed in full rack-scale systems with Nvidia’s famous Hopper, Blackwell or Rubin GPUs.

The growing demand for GPUs has turned Nvidia into a household name and the world’s most valuable publicly traded company with a market cap of $4.4 trillion. Its broader chip strategy took a major turn in February when Nvidia inked a multiyear deal Meta This includes the first large-scale deployment of Grace CPUs, with plans to deploy Vera in 2027.

Thousands of standalone Nvidia CPUs are helping power supercomputers at the Texas Advanced Computing Center and Los Alamos National Lab, Nvidia told CNBC.

Bank of America It predicts that the CPU market could double from $27 billion in 2025 to $60 billion by 2030. In the most recent quarter alone, Nvidia generated more than $62 billion in data center revenue, up 75% from the previous year.

The CPU resurgence is driven by a fundamental shift in compute needs, as mass AI adoption shifts from call-and-answer chatbots to task-based agent applications.

While GPUs are ideal for training and running AI models because they have thousands of small cores narrowly focused on performing multiple operations simultaneously, CPUs have a smaller number of powerful cores that run sequential general-purpose tasks.

Agentic AI requires a lot of general compute power, as they move large amounts of data for AI workflows, orchestrating across multiple agents.

“These agent systems are spawning different agents that work as a team,” CEO Jensen Huang said on Nvidia’s earnings call last month. “The number of tokens being produced has really gone exponential, and so we need to assess at a much faster pace.”

Huang mentioned Agent AI dozens of times on the call and said that “optimal performance per watt is literally everything” as hardware requires a shift.

The company said in a press release that its standalone CPUs offer significant performance-per-watt improvements in Meta’s data centers.

“It’s a new infrastructure: a greenfield extension of racks of CPUs whose sole job is to run agent AI,” said Ben Bajarin, chip analyst at Creative Strategies. “Your software is going to sit somewhere else, your accelerators are going to just run tokens, but something has to sit in the middle and orchestrate it.”

‘Quiet Supply Crisis’

Now, the once-sleepy central processor market is facing what Futurum Group calls a “quiet supply crunch,” predicting that the CPU market’s growth rate could exceed GPU growth by 2028.

Major CPU suppliers AMD And Intel According to Reuters, consumers in China have been warned of supply shortages. CPU delivery times are up to six months and prices are up more than 10%, according to the report.

“The increase in demand over the last six to nine months has been unprecedented,” Forrest Norrod, AMD’s head of data centers, told CNBC in an interview.

Norrod said he doesn’t see “any prospect of it slowing down or stopping anytime soon,” but AMD expects a lift in demand and is “working diligently” to meet it.

An Intel spokesperson told CNBC that inventory is expected to hit its “lowest level” in the current quarter, “but we are aggressively addressing and expect supply improvement in Q2 through 2026.”

“Wafers don’t grow on trees,” Bajarin said. “It’s not like we’re just harvesting 10% more silicon wafers. There’s a crunch across the entire industry. So unfortunately, CPU wafers are constrained.”

As for whether Nvidia has seen any CPU shipment delays, Harris told CNBC, “So far, so good.”

Nvidia’s “robust supply chain” is able to handle demand, he said, because many of its CPUs are sold with GPUs in its rack-scale systems.

Optimized to ‘feed their GPUs’

Harris said Nvidia took a fundamentally different approach to design than the more general-purpose CPUs made by industry leaders Intel and AMD.

There is a big difference in the number of cores in each CPU.

AMD’s EPYC line and Intel’s Xeon high-performance server CPUs typically have 128 cores, compared to 72 cores in Nvidia’s Grace CPUs.

“If you’re a hyperscaler, you want to maximize the number of cores per CPU, and that essentially lowers the cost, dollars per core. So that’s a business model,” Harris explained.

Instead, Nvidia designed its CPU specifically to help its Star GPUs run AI workloads.

“Your single-thread performance is more important than your dollars per core because you’re trying to make sure that the most expensive resource, which is the GPU, isn’t sitting there waiting,” Harris said.

Nvidia also bases its CPUs on the arm The architecture is more commonly used for chips in low-power devices such as smartphones, but Intel and AMD base their CPUs on the traditional x86 architecture. Introduced by Intel nearly 50 years ago, x86 has been a major instruction set that has dominated PC and server processor designs since its inception.

AMD’s Norrod Nvidia said, “Their chips are well-optimized to feed their GPUs. They’re not well-suited for general-purpose applications.”

In fact, Nvidia relies on more general-purpose CPUs for some of its products. For example, Nvidia pairs its GPUs with host CPUs from Intel or AMD on its HGX Rubin NVL8 platform, which customers use as building blocks for their own AI racks.

‘Platform Agnostic’

Nvidia gains access to standalone CPUs because many of its customers are making their own ARM-based processors for their data centers.

Amazon Graviton was the first major hyperscaler to launch an internal CPU with the release in 2018. GoogleAccording to Futurum Group, the Axeon processor, released in 2024, will now handle about 30% of internal applications. Microsoft It launched its second-generation Cobalt processor in November. Arm is expected to launch its own in-house CPU this year, with the Meta as an early customer.

Mercury Research estimates that server CPU market share in the last quarter of 2025 will be dominated by Intel at 60%, AMD at 24.3% and Nvidia at 6.2%, with the remaining share split between in-house ARM-based CPUs from hyperscalers such as Amazon, Microsoft and Google.

In the face of an insatiable need for computing, Nvidia typically takes a welcoming attitude toward competition. Keeping with that tradition, Nvidia opened up its NVLink networking technology to third-party licensing in May.

The rest of 2025 saw a flurry of NVLink deals with Intel, Qualcomm, FujitsuAnd Arm eases the way for third-party CPUs to integrate with Nvidia GPUs in AI servers.

While these deals include CPUs made on the ARM or x86 architecture, Nvidia now supports the open instruction-set architecture RISC-V. Gaining traction in recent years, RISC-V allows companies like Arm to design custom processors without paying licensing fees.

In January, Nvidia struck a deal to enable US chip company SiFive to use NVLink to connect its RISC-V chip designs with Nvidia GPUs.

Regardless of how CPU demand pans out, Nvidia’s strategy remains “platform agnostic,” Harris said.

“We’re definitely building an Arm-based CPU, but we’re invested in the x86 community, we’re invested across the ecosystem, we’re going to have a strong position either way.”

Bajarin describes Nvidia’s transfer strategy as “soup-to-nuts.”

“Nvidia’s answer cannot be, ‘To compete, you buy GPUs from us or nothing else,'” Bajarin said. Whether it’s GPUs, CPUs or specialized hardware, “the way the product has to expand to meet diversity workloads,” he said.

WATCH: CNBC’s exclusive first look at Nvidia’s Vera Rubin AI system