Nvidia’s CPU strategy is taking center stage at GTC 2026, with the company set to unveil new details about its agentic-optimized central processing units at its annual conference beginning Monday, March 16, 2026. A CPU-only rack is expected to appear on the showroom floor, marking a notable departure from Nvidia’s traditional GPU-centric presentations.
“CPUs are becoming the bottleneck in terms of growing out this AI and agentic workflow,” said Dion Harris, Nvidia’s head of AI infrastructure, in an interview this week. Harris described the development as an opportunity for the company to expand its data center footprint beyond graphics processors.
From Grace to Vera: Nvidia’s CPU Roadmap
Nvidia first entered the data center CPU market with its Grace processor in 2021. The next-generation Vera CPU is now in production. Both processors are typically deployed alongside Nvidia’s Hopper, Blackwell, or Rubin GPUs in full rack-scale systems.
However, the company’s CPU approach took a major turn in February 2026, when Nvidia struck a multiyear deal with Meta. The agreement included the first large-scale standalone deployment of Grace CPUs in Meta’s data centers, with plans to deploy Vera in 2027. Thousands of standalone Nvidia CPUs are also powering supercomputers at the Texas Advanced Computing Center and Los Alamos National Lab.
Agentic AI Fuels the CPU Renaissance
The CPU resurgence stems from a fundamental change in artificial intelligence compute requirements. Mass AI adoption is shifting from call-and-answer chatbots to task-oriented agentic applications. While GPUs excel at training and running AI models through thousands of narrow parallel cores, CPUs handle sequential general-purpose tasks with fewer but more powerful cores.
“These agentic systems are spawning off different agents working as a team,” said CEO Jensen Huang on Nvidia’s earnings call last month. “The number of tokens that are being generated has really, really gone exponential, and so we need to inference at a much higher speed.” Huang mentioned agentic AI a dozen times during the call.
“This is new infrastructure: Greenfield expansion of racks of CPUs whose only job is to run agentic AI. Your software is going to sit elsewhere, your accelerators are just going to run tokens, but something has to sit in the middle and orchestrate that.”
Ben Bajarin, Chip Analyst, Creative Strategies
A Quiet Supply Crisis Grips the CPU Market
The broader CPU market is facing what The Futurum Group calls a “quiet supply crisis.” The research firm predicts the CPU market growth rate could exceed GPU growth by 2028. Bank of America forecasts the CPU market could more than double, from $27 billion in 2025 to $60 billion by 2030.
Leading CPU providers AMD and Intel have warned customers in China of supply shortages, according to Reuters. CPU delivery lead times have stretched to six months, and prices have risen more than 10%. “Increases in demand are unprecedented over the last six to nine months,” said Forrest Norrod, AMD’s head of data center, in an interview with CNBC. An Intel spokesperson said the company expects inventory to reach its “lowest level” in the current quarter, but added that supply improvement is expected in Q2 through 2026.
As for Nvidia, Harris said the company’s supply chain has managed demand so far, largely because many of its CPUs are sold alongside GPUs in rack-scale systems. “So far, so good,” Harris told CNBC.
Nvidia CPU Strategy: Platform Agnostic by Design
Nvidia’s Grace and Vera CPUs are built on Arm architecture, which differs from the x86 architecture used by Intel’s Xeon and AMD’s EPYC server processors. Nvidia’s CPUs feature 72 cores, compared to the 128 cores typical in AMD and Intel high-performance server chips. Harris said Nvidia designed its CPU to maximize single-threaded performance, ensuring its GPUs are not left waiting for data.
Meanwhile, Nvidia opened its NVLink networking technology to third-party licensing in May 2025. Deals followed with Intel, Qualcomm, Fujitsu, and Arm, enabling third-party CPUs to integrate with Nvidia GPUs in AI servers. In January 2026, Nvidia also struck a deal with SiFive, allowing the U.S. chip company to connect its RISC-V chip designs with Nvidia GPUs via NVLink.
According to Mercury Research, the server CPU market share in Q4 2025 stood at Intel 60%, AMD 24.3%, and Nvidia 6.2%, with the remainder split among in-house Arm-based processors from Amazon, Microsoft, and Google. Harris said Nvidia’s approach remains “platform agnostic” regardless of how CPU demand is filled across the industry.
In the latest quarter, Nvidia generated data center revenue of over $62 billion, up 75% from a year earlier, underscoring the scale of the company’s broader AI infrastructure business as it expands into CPUs.