Google Unveils TPU 8t and TPU 8i with 2.8x Price-Performance Gain
Google unveiled two new AI processors, the TPU 8t for model training and TPU 8i for inference, at Google Cloud Next 2026, claiming a 2.8x price-to-performance improvement over prior units. Availability is set for later this year alongside multigigawatt capacity deals with Anthropic and OpenAI.
1. Launch Details
At Google Cloud Next 2026 in Las Vegas, Google introduced the TPU 8t and TPU 8i AI chips, marking its latest hardware push to compete with Nvidia and AMD in the data center AI market.
2. Performance and Specifications
The TPU 8t is optimized for training frontier models, reducing development cycles from months to weeks while offering 2.8x better price-to-performance versus its predecessor; the TPU 8i is tailored for inference workloads and AI agents.
3. Strategic Partnerships
Google has secured multigigawatt commitments with AI labs, including Anthropic for next-generation TPU capacity and planned provisioning for OpenAI’s offerings, reinforcing its chip ecosystem.
4. Market Implications
These new TPUs intensify pressure on Nvidia’s data center business and demonstrate hyperscalers’ shift toward proprietary silicon, potentially reshaping cloud AI service economics and supplier dynamics.