Google expands specialised TPU lineup to challenge Nvidia

Google is broadening its AI chip portfolio with the TPU 8tx and 8i Max, specialised variations designed for secure computing and applications that need to process more information per request. Available in 2027, the processors reflect a shift towards hardware tailored to specific workloads rather than a single design serving the entire market.

The 8tx pairs with an AMD CPU and enables chip-level encryption for customers such as government agencies and regulated businesses. The 8i Max combines TPU chips in a single package to provide more computing power and memory for long-context applications, including drug discovery, mathematics and coding. Both are reconfigured versions of existing designs rather than completely new silicon.

Google has attracted customers including Apple, Meta Platforms, Oracle and Anthropic. Bloomberg Intelligence estimates TPU sales, including related software and hardware, will exceed US$135 billion (S$173 billion) by 2028, excluding usage through Google Cloud. Analyst Mandeep Singh said TPUs cost about 30% less than Nvidia’s GPUs while offering competitive throughput. Nvidia is also expanding its specialised offerings for AI agents and inference.

Growth faces constraints from component shortages, limited manufacturing capacity and potential weakness in infrastructure spending. Google says every available TPU is already contracted to a customer. Competition is also widening as start-ups and customers including OpenAI and Anthropic pursue their own chip designs.