Jakarta, INTI - Google Cloud officially introduced its eighth-generation Tensor Processing Units (TPUs), the TPU 8t and TPU 8i, designed to accelerate artificial intelligence (AI) workloads in the Agentic era.
Amin Vahdat, Google’s SVP and Chief Technologist, AI and Infrastructure, stated that this year, two specialized chips will be launched simultaneously to welcome the era of agentic AI. He mentioned this during the Future of AI Infrastructure event, Google Cloud Next 26, in Las Vegas on Tuesday, April 22, 2026.
He also explained that the TPU 8t was designed for training and as the main engine. As for the TPU 8i, it was designed for the AI inference process. Vahdat explained that for years, TPUs have powered leading baseline models, including Gemini. This combined eighth-generation TPU will deliver scale, efficiency, and capabilities spanning training workloads, serving, and agentic needs.
Adapt to New Demands
Google Cloud believes that in the era of agentic AI, models must be able to reason problems, execute multi-stage workflows, and learn from their own actions in a continuous cycle. However, this places a new set of demands on infrastructure.
With the commercial availability of its latest chips later this year, Google Cloud is working to redefine the boundaries of AI capabilities for developers and global enterprises. The use of this technology, including in its Gemini Enterprise service, is expected to accelerate scientific discovery and transform the healthcare sector on an international scale.
Conclusion
Google Cloud launched its eighth-generation TPUs, TPU 8t and TPU 8i, to support the needs of AI in the agentic era. TPU 8t is focused on training large-scale models with high compute capacity, while TPU 8i is optimized for fast and efficient inference. Both were designed to suit their respective functions, rather than simply being derivatives of each other. This technology aims to handle the complexities of agentic AI and is expected to accelerate innovation across various sectors through services such as Gemini Enterprise.
Read more: Google is Announcing Its AI Chips Specified for Inference Processing