Google launches new AI inference chips to challenge Nvidia's chip dominance as rivals rush to buy its hardware

Google is introducing specialized chips designed for inference — the step where a trained AI model produces answers to user questions. Competitors including some of Google's biggest rivals are already stockpiling Google's current AI chips, signaling demand for alternatives to Nvidia, which controls most of the market for AI hardware.

Inference chips are cheaper and more efficient than training chips because they handle a narrower, repetitive task: once an AI model is built, inference chips just run it thousands of times. This means companies deploying AI chatbots, search tools, or recommendation systems can cut their hardware bills and power consumption compared to using general-purpose chips from Nvidia.

If Google's new chips gain adoption, AI services will become cheaper to operate — meaning companies can offer AI features to more users or at lower prices. Teams building AI products inside enterprises (banks, retailers, healthcare) also gain a second supplier option, reducing their dependence on Nvidia for equipment.