Why the CPU is taking center stage

NvidiaGraphics processing units have been the best-selling chips for years, but the sudden emergence of agency AI has brought a renaissance for the more modest main chip, the central processing unit.

Now Nvidia is set to reveal new details about its brokerage-optimized CPUs at its annual event. GTC conference It launches on Monday, and a CPU-only shelf will likely appear on the showroom floor.

“CPUs are becoming the bottleneck in terms of expanding this AI and mediated 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 production of the new generation Vera started. CPUs are often deployed alongside Nvidia’s famous Hopper, Blackwell or Rubin GPUs in full rack-scale systems.

Rising demand for GPUs has made Nvidia a household name and the world’s most valuable public company, with a market cap of $4.4 trillion. The broader chip strategy got a major transformation in February when Nvidia struck a multi-year deal. Meta This included the first large-scale deployment of Grace CPUs, with plans to distribute Vera in 2027.

Thousands of discrete Nvidia CPUs also help power supercomputers at the Texas Advanced Computing Center and Los Alamos National Laboratory, Nvidia told CNBC.

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

The resurgence of the CPU stems from a fundamental change in computing needs; Mass AI adoption is shifting from call-and-response chatbots to task-focused agent applications.

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

Agency AI requires a lot of overhead computing power as it moves and organizes large amounts of data across multiple agents for AI workflows.

“These intermediary systems arise from different agents working as a team,” CEO Jensen Huang said on Nvidia’s earnings call last month. “The number of tokens being created has truly increased exponentially, and so we need to mine at a much higher rate.”

Huang mentioned agent AI a dozen times during the call, saying “best performance per watt is literally everything” as hardware needs to change.

Its discrete CPUs deliver significant improvements in performance per watt in Meta’s data centers, the company said in a press release.

“This is a new infrastructure: a greenfield expansion of CPU racks whose sole job is to run agency AI,” said chip analyst Ben Bajarin of Creative Strategies. “Your software will sit somewhere else, your accelerators will just run tokens, but something has to sit in the middle and regulate that.”

‘Silent supply crisis’

Now, the once dormant central processor market is facing what Futurum Group calls a “silent supply crisis.” CPU market growth rate may exceed GPU Growth until 2028.

Leading CPU providers AMD And Intel They warned customers in China about a supply shortage, according to Reuters. CPU delivery times are up to six months and prices have increased by more than 10%. report.

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

Norrod said he doesn’t see “any chance of this slowing down or stopping anytime soon,” but that AMD is anticipating the increase in demand and is “working diligently” to meet it.

An Intel spokesperson told CNBC that it expects inventory to fall “to its lowest level” this quarter, “However, we are aggressively addressing this issue and expect supply recovery in the second quarter through 2026.”

“Wafers don’t grow on trees,” Bajarin said. “This doesn’t mean we can just add 10% more silicon wafers. There’s a crisis in the entire industry. So unfortunately CPU wafers are in short supply.”

As for whether Nvidia sees any delays in CPU delivery, Harris told CNBC: “So far, so good.”

He said Nvidia’s “strong supply chain” has been able to manage demand, largely because most CPUs will be sold alongside GPUs in rack-scale systems.

Optimized to ‘feed their GPUs’

Harris said Nvidia is taking a fundamentally different approach to design that makes its CPUs “best suited” for data processing and mediated AI workflows, compared to 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 series and Intel’s Xeon high-performance server CPUs typically have 128 cores, while Nvidia’s Grace CPU has 72 cores.

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

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

“Your single-thread performance becomes much more important than the dollar you pay per core because you’re trying to make sure a very expensive resource like a GPU isn’t sitting there,” Harris said.

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

AMD’s Norrod said Nvidia “I think they optimize their chips very well to power their GPUs. They’re not well optimized for general-purpose applications.”

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

‘Platform independent’

Nvidia’s push into discrete CPUs comes as more customers build their own Arm-based processors for data centers.

Amazon With the launch of Graviton in 2018, it became the first major hyperscaler to launch an in-house CPU. GoogleThe Axion processor, released in 2024, now handles about 30% of internal applications, according to Futurum Group. Microsoft It launched its second-generation Cobalt processor in November. Arm expected It will launch its own in-house CPU this year, with Meta as its first customer.

Mercury Research estimates that in the last quarter of 2025, server CPU market share is 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 like Amazon, Microsoft, and Google.

In the face of insatiable need for computing, Nvidia generally takes a positive attitude towards competition. Continuing this tradition, Nvidia opened its NVLink networking technology to third-party licensing in May.

A number of NVLink deals are in place with Intel for the remainder of 2025. Qualcomm, Fujitsuand Arm is streamlining the way for third-party CPUs to integrate with Nvidia GPUs in AI servers.

While these deals include CPUs built on Arm or x86 architecture, Nvidia now also supports the open instruction set architecture RISC-V. RISC-V, which has gained attention in recent years, allows companies to design custom processors without paying licensing fees to companies such as Arm.

In January, Nvidia reached an agreement It allows US chip company SiFive to use NVLink to connect RISC-V chip designs to Nvidia GPUs.

No matter how CPU demand is met, Nvidia’s strategy will remain “platform agnostic,” Harris said.

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

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

“To compete, Nvidia’s answer is you have to buy GPUs from us or something else,” Bajarin said. Whether it’s GPUs, CPUs or specialized hardware, “this is the way the product has to expand to meet a variety of workloads,” he said.

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