The CPU’s Ascendancy

Nvidia’s CPU Ambitions: Powering the Agentic AI Revolution

Nvidia’s dominance in the graphics processing unit (GPU) market has been a well-established narrative for years, fueling the artificial intelligence boom. However, the emergent landscape of agentic AI, characterized by sophisticated, task-oriented AI systems, is catalyzing a resurgence for Nvidia’s less heralded central processing units (CPUs). As the company prepares to unveil new details about its agentic-optimized CPUs at its annual GTC conference, industry watchers are taking note of a significant strategic pivot.

“CPUs are becoming the bottleneck in terms of growing out this AI and agentic workflow,” Dion Harris, Nvidia’s head of AI infrastructure, told CNBC, highlighting what he described as an “exciting opportunity.” This sentiment underscores a fundamental shift in compute demands, moving beyond the traditional strengths of GPUs towards a more nuanced requirement for general-purpose processing power.

Nvidia’s initial foray into data center CPUs began with the Grace processor in 2021, followed by its next generation, Vera, which is now entering production. These CPUs are typically integrated into comprehensive, rack-scale systems alongside Nvidia’s industry-leading GPUs such as Hopper, Blackwell, and Rubin. The company’s meteoric rise, propelled by the insatiable demand for its GPUs, has cemented its position as the world’s most valuable publicly traded company.

A pivotal moment in Nvidia’s broader chip strategy occurred in February with a multiyear agreement with Meta. This landmark deal included the first large-scale deployment of Grace CPUs as standalone units, with plans to integrate Vera in 2027. Beyond hyperscalers, thousands of Nvidia CPUs are also powering critical supercomputing infrastructure at institutions like the Texas Advanced Computing Center and Los Alamos National Lab, underscoring their growing importance in high-performance computing environments.

The market outlook for CPUs is remarkably robust. Bank of America projects the CPU market could more than double, from $27 billion in 2025 to $60 billion by 2030. Nvidia’s data center segment alone posted over $62 billion in revenue in its latest reported quarter, a staggering 75% increase year-over-year.

The driving force behind this CPU renaissance is the evolution of AI applications. While GPUs excel at parallel processing for training and running complex AI models, agentic AI systems require robust general-purpose compute capabilities to manage data orchestration and coordinate multiple AI agents. This involves substantial data movement and complex sequential task execution, areas where CPUs historically shine.

“These agentic systems are spawning off different agents working as a team,” Nvidia CEO Jensen Huang articulated during a recent earnings call. “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’s emphasis on agentic AI, mentioned a dozen times on the call, coupled with his assertion that “the best performance-per-watt is literally everything,” signals a critical hardware inflection point. Nvidia has explicitly stated that its standalone CPUs offer significant performance-per-watt improvements for workloads like those at Meta.

“This is new infrastructure: Greenfield expansion of racks of CPUs whose only job is to run agentic AI,” commented chip analyst Ben Bajarin of Creative Strategies. “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.”

A “Quiet Supply Crisis” Looms

The burgeoning demand for CPUs has led to what some analysts are terming a “quiet supply crisis.” The Futurum Group forecasts that CPU market growth could outpace GPU growth by 2028. Reports from Reuters indicate that leading CPU manufacturers like AMD and Intel have alerted customers in China to potential supply shortages, with CPU delivery lead times extending up to six months and prices experiencing increases of over 10%.

“Increases in demand are unprecedented over the last six to nine months,” stated Forrest Norrod, AMD’s head of data center, in a recent interview. He expressed no expectation of this trend slowing down soon but confirmed AMD’s proactive measures to meet the demand. An Intel spokesperson indicated that inventory levels are expected to reach their “lowest level” this quarter, with aggressive efforts underway to improve supply throughout 2026.

The inherent constraints of semiconductor manufacturing are a key factor. “Wafers don’t grow on trees,” Bajarin observed. “There’s a crunch across the entire industry. So unfortunately, CPU wafers are constrained.” Despite these industry-wide challenges, Nvidia reports that its “robust supply chain” has managed to meet demand thus far, largely due to the integrated nature of its CPU and GPU offerings.

Optimized for GPU Augmentation

Nvidia’s approach to CPU design fundamentally differs from traditional providers like Intel and AMD. Harris explained that Nvidia’s CPUs are meticulously engineered for data processing and agentic AI workflows, prioritizing single-threaded performance to ensure its high-value GPUs are not idled. This contrasts with the hyperscaler focus on maximizing core count per CPU to drive down dollars per core.

For instance, Nvidia’s Grace CPU features 72 cores, while AMD’s EPYC and Intel’s Xeon server CPUs typically offer 128 cores. This design choice reflects Nvidia’s strategy to optimize the flow of data to its GPUs. “Your single-threaded performance becomes much more important than your dollars per core because you’re trying to make sure that that very expensive resource, being the GPU, isn’t sitting there waiting,” Harris elucidated.

Furthermore, Nvidia’s CPUs are built on the Arm architecture, commonly found in mobile devices, a departure from the x86 architecture that has dominated PC and server designs for decades. AMD’s Norrod acknowledged Nvidia’s optimization, noting, “Optimized their chips very well, I think, for feeding their GPUs. They’re not well optimized for general-purpose applications.” It’s worth noting that Nvidia does integrate Intel or AMD CPUs in some of its platforms, like the HGX Rubin NVL8, for broader AI rack configurations.

Platform Agnosticism and Ecosystem Integration

Nvidia’s expansion into standalone CPUs aligns with the trend of major cloud providers developing their own Arm-based processors. Amazon launched its Graviton CPU in 2018, followed by Google’s Axion in 2024, which reportedly handles a significant portion of internal applications. Microsoft has also introduced its Cobalt processor. Arm itself is anticipated to release its own in-house CPU this year, with Meta as an early adopter.

While Intel and AMD continue to hold dominant market share in the server CPU space, the rise of in-house solutions from hyperscalers is reshaping the competitive landscape. Nvidia, however, is adopting a notably inclusive strategy. By opening up its NVLink networking technology to third-party licensing, Nvidia has fostered collaborations with companies like Intel, Qualcomm, Fujitsu, and Arm, facilitating the integration of diverse CPUs with Nvidia GPUs in AI servers.

Beyond Arm and x86, Nvidia is embracing the open-source RISC-V instruction set architecture. A recent agreement with SiFive allows the U.S. chip company to leverage NVLink for connecting its RISC-V chip designs with Nvidia GPUs. This strategic openness, which Harris describes as “platform agnostic,” ensures Nvidia maintains a strong position regardless of the underlying CPU architecture.

“We are certainly building an Arm-based CPU, but we are so invested in the x86 community, we’re so invested across the ecosystem, that we’re going to have a strong position either way,” Harris affirmed. Bajarin characterizes Nvidia’s evolving strategy as “soup-to-nuts,” reflecting a comprehensive approach to serving the diverse and rapidly expanding demands of the AI compute market. Whether it’s through GPUs, CPUs, or specialized accelerators, Nvidia appears committed to providing a complete ecosystem to meet the multifaceted needs of AI development.