Nvidia built its dominance on AI-powered GPUs, but CEO Jensen Huang is now preparing investors for a renewed competitive push into CPUs — long the home turf of Intel and Advanced Micro Devices.

While GPUs remain central to AI training workloads, Nvidia is positioning its in-house CPUs as a critical component of the next phase of artificial intelligence: deployment, inference, and agentic computing.

The CPU Comeback in the AI Era

For decades, CPUs — central processing units — were considered the “brains” of computing systems, handling a broad range of general-purpose tasks. GPUs, by contrast, specialized in highly parallel mathematical computations, making them ideal for graphics rendering and, more recently, AI model training.

Huang has often described how computing shifted from roughly 90% CPU and 10% GPU workloads to the reverse during the AI boom. But that balance may be evolving again.

As AI companies pivot from training large frontier models to deploying AI agents capable of coding, research synthesis, and automation, certain workloads are increasingly CPU-intensive. Analysts suggest that inference and orchestration layers in AI systems may rely more heavily on CPUs, particularly in data-rich environments.

Nvidia’s flagship AI server, the NVL72, currently integrates 36 Nvidia CPUs alongside 72 GPUs. Some industry observers believe future AI infrastructure — particularly agentic systems — could move toward a more balanced CPU-to-GPU ratio.

Nvidia’s Strategic CPU Ambition

Nvidia’s Grace CPU platform, first introduced for data centers in 2023, is central to this strategy. Unlike traditional CPU designs, Nvidia emphasizes memory bandwidth and data throughput — aligning architecture with AI’s data-driven demands.

Huang recently suggested Nvidia could become “one of the largest CPU makers in the world,” signaling that this expansion is not a defensive hedge but a deliberate growth pillar.

A recent deal with Meta Platforms highlights that ambition. Meta will deploy Nvidia’s Grace and Vera CPUs on a standalone basis — a notable shift from Nvidia’s previous GPU-dominant server configurations. However, AMD also secured a major supply deal with Meta, reinforcing that hyperscalers are diversifying suppliers rather than abandoning incumbents.

Renewed Battle with Legacy Leaders

Intel has historically dominated the data center CPU market, while AMD has steadily gained share with its EPYC lineup. Nvidia’s entry intensifies competition in a segment already under pressure from custom silicon developed by hyperscalers.

Huang argues Nvidia’s approach differs structurally. Rather than emphasizing modular chiplet strategies used by competitors, Nvidia is designing CPUs optimized for high-throughput data processing — a core requirement for AI workloads.

Analysts suggest Nvidia’s broader aim is to challenge the assumption that Intel-style CPUs are the default backbone of computing infrastructure. Instead, CPUs become one architectural option within a more heterogeneous compute ecosystem increasingly shaped by AI requirements.

Investor Implications

Nvidia’s move does not signal weakness in its GPU franchise. Instead, it reflects vertical expansion — capturing more value inside AI data centers by supplying both accelerators and general-purpose processors.

The competitive landscape is shifting toward full-stack infrastructure control. If Nvidia succeeds in scaling CPU adoption alongside GPUs, it could increase wallet share per data center deployment and deepen customer lock-in.

With further announcements expected at Nvidia’s upcoming developer conference, investors are watching closely. The AI boom may have started with GPUs — but the next chapter may hinge on how effectively Nvidia executes its CPU offensive.