Pat Gelsinger, former CEO of Intel, recently appeared on the Acquired podcast at NVIDIA's 2025 GPU Technology Conference, stating that NVIDIA's pricing strategy for artificial intelligence (AI) graphics processing units (GPUs) is excessively high, hindering the widespread adoption of AI inference tasks. Gelsinger pointed out that inference is crucial for deploying AI models, and the industry should prioritize it, but NVIDIA's technology falls short in terms of cost-effectiveness.
Image Source Note: Image generated by AI, licensed from Midjourney.
He mentioned that NVIDIA's processors for AI training are priced up to 10,000 times higher than what's realistically needed. While acknowledging that NVIDIA's GPUs were instrumental in the early rapid development of generative AI, Gelsinger believes the company's strength—its CUDA software platform—may face challenges as inference becomes mainstream. He emphasized that despite these shortcomings, he admires CEO Jensen Huang's vision and tenacity, acknowledging Huang's successful early predictions regarding general-purpose GPUs and AI workloads, partly attributed to favorable timing. He even quipped, "Jensen Huang got lucky."
Under Gelsinger's leadership, Intel has faced pressure in the AI hardware competition. The company's Gaudi accelerator chips have failed to match the performance of NVIDIA's Hopper and AMD's Instinct products. Intel has shelved its Falcon Shores AI platform and is now focusing on its next-generation project, "Jaguar Shores."
Gelsinger also mentioned the potential for shifts in computer architecture, with quantum computing expected to become commercially viable by the end of the century. However, he didn't disclose Intel's specific plans for this transformation. Despite the surging demand for machine learning infrastructure, Intel's AI revenue significantly lags behind competitors, highlighting the company's overall struggles in this area.
Key takeaways:
💰 Former Intel CEO criticizes NVIDIA's high AI chip pricing, hindering large-scale inference tasks.
🔍 Gelsinger points out that NVIDIA's processors for AI training are far more expensive than necessary.
🌀 Intel is struggling in the AI hardware competition and is currently focusing on new project development.
