Whalesbook
AI Chip Market Dynamics Shifting
The notion that artificial intelligence is exclusively tied to Nvidia's graphics processing units (GPUs) is being actively dismantled, according to Andrew Feldman, co-founder and CEO of AI chip designer Cerebras Systems. Feldman's remarks at the World Economic Forum in Davos signal a significant potential disruption to Nvidia's current market leadership.
"I think we've already seen Nvidia's dominance being challenged," Feldman stated in an interview. "The GPU is not the only machine that can be used in AI. The mental moat for those who thought that AI equalled Nvidia has been crossed."
Evidence of Transition
Feldman cited Nvidia's reported $20 billion acquisition of Groq as evidence that GPUs have limitations. "Nvidia admitted they didn't have a solution and spent $20 billion to buy the number two player. That shows there are things GPUs don't do well," he elaborated. This move by Nvidia to acquire Groq's technology underscores a strategic effort to integrate new capabilities, even as Groq continues as an independent entity.
Google's Shift and Inference Focus
A major industry inflection point, according to Feldman, was Google's decision to train its Gemini AI model using its own Tensor Processing Units (TPUs) rather than Nvidia hardware. "What we saw was the training of a foundation model entirely without Nvidia. That was a big moment," he said.
The Cerebras CEO emphasized the market's rapid pivot towards rapid inference – the critical stage where AI models are deployed for real-world applications. "Inference is where AI meets the real economy. That's where coding, agentic work and deep research happen," Feldman explained. Cerebras Systems, founded in 2016, offers its Wafer Scale Engine (WSE), billed as the world's largest AI accelerator, aiming to compete directly with established giants like NVIDIA.
Dismissing AI Bubble Concerns
Feldman also addressed fears of an artificial intelligence market bubble, dismissing the notion. "This isn't a bubble. This is how a new technology diffuses through an entire economy... People want AI to be faster. They don't want to wait. Faster inference is what users care about most today," he concluded.