Google says Ironwood, its seventh-generation TPU, will be generally available on Google Cloud in the coming weeks

Google DeepMind/Tensor Processing Unit (TPU)

Google said Ironwood, its seventh-generation TPU, would be generally available in weeks, offering 10× peak performance of TPU v5p and 4× per-chip performance of Trillium for training and inference. It also announced new VMs on Axion Arm-based processors.

• Superpod scales to 9,216 chips with 9.6 Tb/s inter-chip interconnect and 1.77 PB of shared high-bandwidth memory.
• Achieves ~99.999% fleet-wide uptime via optical circuit switching and liquid cooling; GKE Inference Gateway reduces first-token latency by 96%.
• Early customers planning deployment: Anthropic (1M TPUs), Lightricks (LTX-2), and Essential AI.