Google adds a two-chip TPU package and one for confidential inference

single source· 1 articles · confidence: low · first seen 2026-11-09 17:09 UTC

What this means for you

Nothing to act on today. Availability is reported for 2027, with no price, no throughput figures and no published benchmark. The confidential-inference part is the one worth tracking: it is a hardware claim that no source here has tested.

Google has added two parts to its eighth-generation TPU line, the chips it designs to run AI models. TPU 8i Max puts two TPU 8i chips in a single package and is aimed at agents working over long context — models handling long inputs across several steps. TPU 8tx adds confidential inference in hardware: a request is meant to stay unreadable while it is processed. Bloomberg reports both are due in 2027. No prices, throughput figures or benchmark results were given.

Key facts

  • ·Google added TPU 8i Max and TPU 8tx to its eighth-generation TPU lineup. source
  • ·TPU 8i Max packs two TPU 8i chips per package, aimed at long-context agents. source
  • ·TPU 8tx adds hardware confidential inference. source
  • ·Bloomberg reports 2027 availability for both chips. source

What the sources say

  • MarkTechPost — Short write-up listing the two additions, with availability taken from Bloomberg and no specifications or pricing.

Sources

The original reporting. Follow these — they did the work.

← the wire