Nvidia Pushes Back on TPU Threat, Says Its GPUs Remain ‘a Generation Ahead’

Nvidia moved to reassure investors Tuesday after its stock fell 3% on reports that Meta may shift part of its data center buildout to Google’s in-house AI chips.

In a post on X, the company said its GPUs remain “a generation ahead of the industry” and emphasized that Nvidia is the only platform that runs every major AI model across all computing environments.

The response follows rising chatter that Google’s tensor processing units (TPUs) — custom ASIC chips used to train the new Gemini 3 model — could become a stronger alternative to Nvidia’s flagship Blackwell GPUs. Analysts estimate Nvidia still controls over 90% of the AI accelerator market.

Nvidia argued that its architecture is more flexible than ASIC-based designs like TPUs, calling its chips higher-performing, more versatile, and more broadly compatible. Google does not sell its TPUs directly but makes them available through Google Cloud.

A Google spokesperson said demand is accelerating for both custom TPUs and Nvidia GPUs, adding that the company has supported a hybrid approach for years.

Nvidia CEO Jensen Huang recently downplayed TPU competition, noting that Google is still a major GPU customer and that Gemini models can run on Nvidia systems.

Huang also cited comments from DeepMind CEO Demis Hassabis reaffirming that industry “scaling laws” — the view that bigger models trained with more compute get better — remain intact, a trend Nvidia says will continue to drive demand for its hardware.

Shares of Nvidia remain volatile as hyperscalers weigh a more diversified mix of AI chips, even as the company maintains a dominant position in next-gen compute infrastructure.

Disclaimer