Google announced its eighth-generation Tensor Processing Units (TPUs) during its Google Cloud Next 2026 conference in Las Vegas on 22 April 2026, unveiling two new chips designed for the growing demands of agentic AI. The lineup includes TPU 8t and TPU 8i, each built for different stages of AI development—from large-scale training to real-time inference.
The TPU 8t is focused on training advanced AI models at scale. According to Google, it delivers nearly three times the compute performance of earlier generations and can cut model development timelines from months to just weeks. It also offers about 2.8× better price-to-performance, making it more cost-efficient for organizations running compute-heavy workloads. With massive superpod configurations, the chip enables faster scaling and improved utilization for frontier AI systems.
On the inference side, the TPU 8i is optimized for running AI models and powering agent-based workflows. It’s engineered for low latency and high responsiveness, making it suitable for real-time applications, reinforcement learning, and Mixture of Experts (MoE) architectures. Enhanced memory capacity and faster interconnect speeds allow it to handle large-scale workloads more efficiently, while also improving overall cost performance.
Beyond the technology itself, the new chips signal Google’s deeper push into the AI hardware race, where it competes with companies like Nvidia and AMD. The competition is heating up as demand for high-performance AI infrastructure continues to surge.
Google is also expanding its partnerships to capitalize on that demand. Earlier this month, it announced a broader agreement with Anthropic to supply “multiple gigawatts” of next-generation TPU capacity. At the same time, the company is reportedly working to provide TPU resources to OpenAI, a key rival to Anthropic, to support its own AI services.
In addition, Meta has signed a multiyear, multibillion-dollar deal for access to Google’s TPU infrastructure, underscoring the growing reliance of major tech firms on specialized AI chips.
Both TPU 8t and TPU 8i are expected to become available to customers later this year through Google Cloud, positioning Google as a major player in delivering scalable, cost-effective AI computing power.
