Key Takeaways
- Jensen Huang, Nvidia’s CEO, declared the company will directly challenge Intel and AMD in the central processing unit arena
- The artificial intelligence industry is transitioning from GPU-intensive model training to CPU-optimized agent deployment
- Grace and Vera, Nvidia’s CPU offerings introduced in 2023, target high-throughput data processing for data center environments
- Meta Platforms inked an agreement to purchase Nvidia CPUs independently, without bundled GPU configurations
- AMD simultaneously secured its own CPU contract with Meta, intensifying rivalry in the processor space
For years, Nvidia built its empire on graphics processing units. Now, CEO Jensen Huang is orchestrating an aggressive expansion into central processing units—territory long controlled by Intel and AMD.
During Wednesday’s fourth-quarter earnings conference call, Huang told financial analysts that Nvidia wasn’t merely prepared for the resurgence of CPU importance—the company intended to dominate it.
“We love CPUs as well as GPUs,” Huang declared during the analyst call.
Central processing units powered the majority of computational tasks for generations. Graphics processing units subsequently captured significant market share when artificial intelligence model training surged, demanding the parallel processing capabilities at which GPUs excel.
This equilibrium is experiencing another transformation. Artificial intelligence enterprises are transitioning from training their models to operationalizing them—deploying what the industry calls “agents” that perform tasks like code writing, document analysis, and report generation.
These operations perform more efficiently on central processing units, according to Creative Strategies analyst Ben Bajarin. He noted that agentic computing “is happening more and more, and sometimes primarily, on the CPU.”
Rebalancing CPU and GPU Workloads in AI
Nvidia’s premier AI server configuration, the NVL72, currently contains 36 central processing units alongside 72 graphics processing units. Bajarin projected this proportion could evolve toward parity—or that GPUs might be eliminated completely for certain applications.
Nvidia launched its Grace and Vera central processing unit products for data center deployment in 2023. Huang explained these processors feature architectural distinctions from Intel and AMD offerings, prioritizing data throughput over multipurpose adaptability.
“It is designed to be focused on very high data processing capabilities,” Huang explained.
Nvidia recently revealed a significant agreement with Meta Platforms to deliver substantial CPU volumes on a standalone basis—crucially, without accompanying GPU purchases. This represents a departure from Nvidia’s conventional sales approach.
Meta isn’t abandoning its current CPU vendors. Instead, the social media giant is diversifying its supplier base. Shortly following the Nvidia announcement, AMD disclosed its own separate CPU agreement with Meta.
Intel’s Market Leadership Now Contested
Dave Altavilla, analyst at HotTech Vision and Analysis, explained that Nvidia aims to demonstrate that the CPU “is no longer the assumed default foundation of modern compute infrastructure.”
Huang also stated during January’s CES that Nvidia CPU adoption in data centers would “explode,” adding he wouldn’t be surprised if Nvidia emerged as “one of the largest CPU makers in the world.”
Nvidia intends to reveal additional information regarding its CPU development roadmap at its upcoming annual developer conference in Silicon Valley next month.
