Google TPU 8tx

Google TPU 8tx is an eighth-generation TPU system for secure, confidential inference and small-scale training, announced on November 9, 2026.

Vendor
Google
Architecture
8th-gen TPU (TPU 8t silicon), confidential computing
Form factor
TPU system with 5th Gen AMD EPYC hosts
Status
Announced

In its November 9, 2026 announcement, Google said TPU 8tx is built on the same silicon as TPU 8t, paired with 5th Generation AMD EPYC CPU hosts, and adds hardware-level confidential computing so data and model weights stay protected while in use. It adopts the low-latency Boardfly network Google built for TPU 8i, and Google claims up to 2x lower latency than TPU 8t with a near-zero performance penalty. Google did not publish memory, bandwidth, throughput or power figures for TPU 8tx, so those fields are left blank.

Our report: Google TPU 8i Max and TPU 8tx: what Google published.

Precision note (sparse or dense)
Google has not published per-chip memory, bandwidth or FP8/FP4 throughput for TPU 8tx; it says the system uses the same silicon as TPU 8t.
Launch
Announced November 9, 2026
Vendor source
blog.google
Checked

Questions

When was Google TPU 8tx announced?
November 9, 2026.
What is TPU 8tx for?
Secure and confidential inference, and small-scale training such as private fine-tuning, according to Google.
What chip does TPU 8tx use?
Google says it is built on the same silicon as TPU 8t.
What are TPU 8tx’s memory and FLOPS?
Google has not published them.

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.

The Model Press

What are you looking for?

Search by headline, topic or keyword.