The new chip is designed to run LLMs that support reasoning, which typically require more compute to generate each response.

In a podcast appearance, DeepMind CEO Demis Hassabis said that Google will eventually combine its Gemini and Veo AI models.