Speaking at Nvidia’s GTC 2025 event during an Acquired podcast, Gelsinger slammed Nvidia’s AI GPUs as being ridiculously overpriced for inferencing workloads, suggesting the market will eventually wake up and realise it has been fleeced.

Gelsinger, whose tenure at Intel saw the company flailing in the AI space, grudgingly admitted that Huang had the right idea by sticking with GPUs rather than trying to compete with CPUs.

However, he was quick to add that Huang “got lucky” when AI took off. The former Intel boss believes that training AI on Nvidia’s GPUs is acceptable, but the real issue lies in deployment, where he argues that GPUs are 10,000 times more expensive than what is required.

Gelsinger has long complained about Nvidia’s dominance in AI, previously calling CUDA a “moat” and insisting that inference, rather than training, is where the real money lies.

That would be a compelling argument if Intel had done anything substantial to capitalise on AI before it became a trillion-dollar industry. Instead, Team Blue spent years fumbling its AI efforts, leading to the cancellation of its much-hyped “Falcon Shores” lineup. Now, it’s pinning its hopes on the next-gen “Jaguar Shores” architecture.

Meanwhile, Intel’s Gaudi lineup continues to underwhelm, struggling to keep up with AMD’s Instinct chips and Nvidia’s Hopper-based GPUs. In a world where AI hardware is generating profits for Nvidia and, to a lesser extent, AMD, Intel appears to be an outsider trying to climb over a back fence to gate crash a party.

With Intel’s new CEO, Lip-Bu Tan, now at the helm, there is a sliver of optimism that Chipzilla might develop a strategy that works. But given its recent track record, betting against Nvidia’s AI reign still seems like a risky move.