Google TPU 8i

Google TPU 8i is the inference and reinforcement-learning chip of Google’s eighth TPU generation.

Vendor: Google
Architecture: 8th-gen TPU (inference)
Memory: 288 GB HBM
Memory bandwidth (TB/s): 8.601 TB/s
FP4 (PFLOPS): 10.1 PFLOPS
Form factor: Cloud TPU
Status: Announced

Google announced TPU 8i on April 22, 2026 at Google Cloud Next. It pairs 288 GB of HBM with 384 MB of on-chip SRAM (three times the previous generation) and doubles ICI bandwidth to 19.2 Tb/s; Google claims 80% better performance per dollar for inference than the prior generation. Google said general availability would come later in 2026.

Google lists peak FP4 per chip as 10.1 PFLOPS, with no sparse or dense label, and memory bandwidth as 8,601 GB/s.

Memory (GB): 288 GB
Precision note (sparse or dense): Google lists peak FP4 per chip as 10.1 PFLOPS with no sparse/dense label. Bandwidth given as 8,601 GB/s.
Launch: Announced April 22, 2026 (Google Cloud Next); general availability planned later in 2026
Vendor source: cloud.google.com
Checked: October 5, 2026

Questions

When was Google TPU 8i announced?

Google announced TPU 8i on April 22, 2026 and said general availability would come later in 2026.

How much memory does TPU 8i have?

TPU 8i has 288 GB of HBM at 8,601 GB/s plus 384 MB of on-chip SRAM, three times the previous generation.

What is TPU 8i used for?

TPU 8i is the inference and reinforcement-learning chip of Google’s eighth TPU generation; Google claims 80% better performance per dollar for inference than the prior generation.

What is TPU 8i’s FP4 performance?

Google lists peak FP4 per chip as 10.1 PFLOPS, with no sparse or dense label.

Source

Every figure is taken from the vendor’s own datasheet, product page, technical documentation or announcement, linked on each chip with the date it was last checked. How we pick and check sources: editorial standards.