GoKawiilGoogle adds TPU 8tx and TPU 8i Max to eighth-gen TPU lineup
Google has expanded its eighth-generation TPU family with two new variants: TPU 8tx, built on the same silicon as its TPU 8t training chip but using a 5th Gen AMD EPYC host CPU and Boardfly networking for confidential, hardware-isolated inference, and TPU 8i Max, which packages two TPU 8i chips together to double compute, SRAM, and HBM bandwidth for long-context agentic workloads. Both are listed as 'coming soon,' joining the previously announced TPU 8t and 8i chips revealed in April.
GoKawiil's interpretation of the reporting above, not reported fact.
The modular design lets Google tailor TPU systems for specific needs—like regulated-industry security or agentic reasoning—by recombining existing silicon rather than engineering new chips from scratch, which Google says can cut development time to months instead of years. This approach could help Google compete for AI infrastructure customers at a moment when it claims nearly 100 TPU customers, including a notable recent shift by Meta toward TPUs, though actual performance and adoption of these new variants remain unproven until they ship.
- TPU 8tx reuses TPU 8t silicon but swaps to an AMD EPYC host CPU for confidential inference aimed at regulated enterprises and governments.
- TPU 8i Max combines two TPU 8i chips on one package, roughly doubling compute, SRAM, and HBM bandwidth for long-context agent tasks.
- Google says nearly 100 customers now use its TPUs, with Meta recently adopting the chips, as the company touts faster, modular chip development cycles.
Source: tomshardware.com — Shane Downing, 2026-11-12
Published there as: “Google reworks TPU silicon into specialized systems for agents and security”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.