At Computex, Intel CEO Pat Gelsinger takes a direct shot at Nvidia CEO Jensen Huang's claim that traditional CPUs like Intel's are running out of steam in AI
- Gelsinger says Jensen Huang is wrong about end of Moore's Law — Intel has struggled to keep up as chip industry evolves
Context & Ripple Effects
The exchange sharpens an argument Intel had already made: [[a:847361|Gelsinger’s case that inference would matter more than training and that customers wanted alternatives to CUDA]]. It also follows Huang’s acknowledgment that Intel, Huawei, and startups were among Nvidia’s competitive challengers.
Intel’s CPU relevance is central to its broader turnaround narrative, after Gelsinger previously described deep leadership, people, and methodology problems at the company. The dispute matters because it pits Intel’s established processor base against Nvidia’s accelerator-led view of AI infrastructure.
First-order effects
- Intel publicly positions CPUs—and continued progress in their design—as part of AI systems rather than a legacy component displaced by accelerators.
- The remarks turn a technical disagreement over AI architecture into a direct competitive contrast between Gelsinger’s and Huang’s strategies.
Second-order effects
- Intel faces greater pressure to demonstrate that its processors contribute materially to AI workloads, especially alongside accelerators, rather than relying on a rhetorical defense of Moore’s Law.
- Nvidia has added incentive to reinforce the case for accelerator-centric AI systems, while customers gain a clearer basis for comparing mixed CPU-GPU deployments with more specialized stacks.
Third-order effects
- If this split persists, AI infrastructure is likely to remain heterogeneous: CPUs, accelerators, and software ecosystems will compete for distinct roles rather than converging on a single winning chip type.
- The strategic contest increasingly hinges on whether customers value integrated, flexible systems or the performance advantages of specialized AI hardware; the outcome will depend on real workload economics and software portability.
The trend: This is one data point in the AI hardware strategy split between heterogeneous systems that retain CPUs as core components and accelerator-led platforms optimized around specialized compute.