Google Rolls Out Two More AI Chips, and the Pressure on Nvidia Grows

On the same day Nvidia's stock was sliding over its CEO drama, Google quietly reminded everyone that it makes AI chips too. On Monday, it announced two new members of its eighth-generation TPU family, called TPU 8tx and TPU 8i Max. Alphabet stock moved up on the news.

A quick refresher. TPUs are Google's custom chips for running and training AI models. Back in April, Google introduced TPU 8t for big training jobs and TPU 8i for fast reasoning. With these two additions, Google says it's the first year it has launched four TPU systems.

TPU 8tx is the security-focused one. It uses the same silicon as TPU 8t but runs alongside AMD's fifth-generation EPYC processors, keeping data and model weights locked away at the hardware level while they're being used. Google is targeting regulated businesses and governments that want AI but can't risk data leaks. It promises up to twice the speed of the TPU 8T on latency with almost no performance hit from the extra security.

TPU 8i Max is the muscle. It packs two TPU 8i chips into one package, doubling the compute, on-chip memory, and memory bandwidth. That's built for AI agents that need to juggle enormous amounts of context, and because it fits twice the silicon in the same rack space, data centers get more power per square foot.

Google also says it now has nearly 100 TPU customers, from AI labs to trading firms and universities, and it's making the chips easier to use with popular tools like PyTorch and coding agents such as Claude Code and Cursor. It didn't give release dates.

For years, Nvidia has had the AI chip market pretty much to itself. Google, Amazon, and others are slowly chipping away at that.

Can Google's TPUs become a real alternative to Nvidia, or will developers stick with what they know?