Google Unveils New TPU 8t and TPU 8i Chips
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Google Cloud unveiled its eighth‑generation Tensor Processing Units, the TPU 8t for training and TPU 8i for inference, slated for general availability later in 2026.
- Google frames the new chips as built for the “agentic era” of AI, emphasizing specialized workloads and faster performance.
- Google claims the TPU 8t/8i chips will slash AI‑computing costs and directly challenge Nvidia’s dominance in the AI‑chip market.
- Virgo Network—a new megascale data‑center fabric—accompanies the TPUs, representing the biggest networking upgrade since TPU v4.
- Google is promoting AI agents built on the new hardware as a way to compete with OpenAI and Anthropic in enterprise AI services.
- Google Cloud pledged $750 million to accelerate partners’ development of “agentic AI” on the new TPU platform.
Why it matters: Enterprises and AI developers gain lower‑cost, workload‑specific hardware that promises faster training and inference, while Nvidia’s dominance in the AI‑chip market is challenged as Google pushes its own chips and a new networking fabric, potentially reshaping supplier dynamics for data‑center operators.
