Google adds specialised TPUs and AMD partnership to challenge Nvidia

Google is accelerating its hardware rollout and introducing specialised Tensor Processing Units (TPUs) as it challenges Nvidia’s position in AI computing. The company is moving beyond general-purpose designs with processors tailored to security-sensitive applications and demanding workloads.

The TPU 8tx targets government agencies and regulated businesses requiring confidential computing. It pairs Google’s technology with an AMD processor to keep sensitive data and AI models encrypted at the hardware level during processing. The TPU 8i Max combines chips in a single unit to deliver double the memory and computing power for tasks including advanced mathematics, coding and drug discovery.

Market analysts estimate Google’s TPUs cost roughly 30 per cent less than Nvidia’s GPUs while offering competitive processing speeds and greater power efficiency. Apple, Meta, OpenAI and Oracle are among companies lined up to use the hardware. Anthropic reportedly has an option to purchase up to $150 billion worth of chips between 2028 and 2031.

Bloomberg Intelligence projects Google’s TPU sales will rise from roughly $12 billion this year to over $135 billion by 2028. Growth faces constraints from competing chip startups, potential slower infrastructure spending and shortages of memory components and manufacturing capacity. Every TPU Google produces is currently contracted to a buyer, prompting the company to explore supply-chain partnerships with Marvell Technology and MediaTek.