Google launches TPU 8tx and 8i Max for confidential AI and memory-hungry agents
testingcatalog · x · 2026-11-09
Google has expanded its eighth-generation TPU lineup with TPU 8tx and TPU 8i Max, bringing the generation to four systems alongside TPU 8t (large-scale training) and TPU 8i (low-latency reasoning). Pricing, regions, and general availability are unspecified.
Both are new system configurations built on existing silicon: the modular architecture lets Google change packaging, memory, networking, and security without a full chip cycle. TPU 8tx uses TPU 8t silicon with custom AMD CPU hosts and adds hardware-level confidential computing that protects data and model weights during active computation—beyond encryption at rest and in transit. It targets regulated enterprises and governments running models and autonomous agents on sensitive information, adopts the Boardfly network from TPU 8i, and Google claims up to 2x TPU 8t performance with near-zero security overhead. TPU 8i Max pairs two TPU 8i chips in one package for low-latency inference, addressing the memory needs of agents processing large codebases, lengthy videos, and million-token contexts.
Related event: Google ships four eighth-gen TPU systems in one year, adding 8tx and 8i Max(8 posts)→
More from Infra
- Google rolls out broader AI chip lineup to challenge Nvidia in a market worth hundreds of billions — dinabass · 2026-11-10
- Thinking Machines engineer returns to Modal, reflecting on a year of frontier ML infrastructure work — josh_wills · 2026-11-10
- AI Supplier Multiples Fell 15% Post-Selloff, Now Below Growth Forecasts — BenBajarin · 2026-11-10
- Samsung and SK hynix each commit $7B-$15B to ASML EUV tools to secure delivery slots — demian_ai · 2026-11-10
- Google VP explains how the company uses AI to design its own infrastructure systems — HaydnBelfield · 2026-11-10
- Musk on first principles: batteries cost $600/kWh but $80 in materials — and why he wants AI data centers in space — elonmusk · 2026-11-10