Google plans two specialized TPUs next year, targeting confidential computing and data-heavy work
On November 9, 2026, Google said it will add two specialized configurations to its eighth-generation tensor processing unit (TPU) line, with availability set for next year. Bloomberg first reported the plan, and Amin Vahdat, Google's chief technologist for AI and infrastructure, told the outlet that specialized chips can deliver more value than one design aimed at the whole market. Both systems are variants of existing eighth-generation hardware rather than new chip designs.
The TPU 8tx is aimed at sensitive workloads run by public-sector bodies and companies in heavily regulated sectors. Its key feature is chip-level encryption that keeps data protected during processing, whereas data normally has to be decrypted in memory while a machine works on it. The TPU 8i Max places two TPU 8i chips in one package, doubling compute, on-chip memory and high-bandwidth memory for long-context work such as coding tools and drug discovery research.
Bloomberg Intelligence estimates the chips cost about 30% less than Nvidia GPUs while offering competitive throughput. It projects TPU sales, including hardware and software but excluding Google Cloud usage, could exceed $135 billion by 2028, up from about $12 billion this year. CoinDesk reported that Nvidia shares fell 5% amid an investigation involving CEO Jensen Huang, and that CoreWeave, IREN and Cipher Digital are down 70% or more from their all-time highs.
Why it matters
For teams training or serving models, the practical change is a second accelerator source with a stated price advantage. The 30% gap is an analyst estimate rather than a published Google price list, so builders should measure price-performance on their own training and inference workloads once the chips reach cloud customers. Regulated buyers also gain a route to run sensitive data on Google accelerators with encryption during processing, which may widen the set of workloads that can move off GPUs.
For investors, the central question is whether Google's chip business becomes a durable revenue line that takes share from Nvidia. Bloomberg Intelligence's forecast implies more than a tenfold rise in TPU sales by 2028. Operators of GPU clouds, already well below their highs, face buyers who focus on cost per unit of compute, so cheaper accelerator supply could reset their pricing and contract terms.
Key points
- 1Google plans TPU 8tx for confidential computing and TPU 8i Max for data-heavy work, both due next year.
- 2Bloomberg Intelligence estimates the chips cost about 30% less than Nvidia GPUs.
- 3Bloomberg Intelligence projects TPU sales could exceed $135 billion by 2028, up from about $12 billion in 2026.
- 4The TPU 8i Max joins two TPU 8i chips in one package, doubling compute, on-chip memory and high-bandwidth memory.
- 5
