Google launches TPU 8tx and TPU 8i Max, completing four-system eighth-generation AI chip lineup (Nov 9)
What happened
Google announced two new TPU systems on Monday, November 9, completing its eighth-generation lineup at four systems in a single year — a first in more than a decade of TPU development. TPU 8tx is built for secure and confidential inference (plus small-scale training) for regulated enterprises and governments: it uses the same silicon as the training-focused TPU 8t, pairs it with 5th Generation AMD EPYC host CPUs and the low-latency Boardfly network, building encryption into the hardware layer so data and model weights stay protected even while in active use, with up to 2x lower latency than TPU 8t. TPU 8i Max targets the memory wall for massive context windows: it combines two TPU 8i chips in a single package, doubling the compute, on-chip SRAM and HBM bandwidth of TPU 8i and delivering the highest compute density per rack. Google also opened up TPU development to PyTorch developers, launched its TPU agent suite and XLA Serving Engine, and said shipments of the two new systems are expected next year.
Why it matters & what’s next are part of Pro, along with the audio summary.
Start your free 14-day trialWho’s affected
- Alphabet (GOOGL) — Google Cloud's in-house AI silicon story strengthens; four TPU systems in one year, first time ever, with nearly 100 customers and all available capacity booked. Nvidia (NVDA) — TPUs are the most credible in-house alternative to its GPUs; industry research cited in coverage forecasts TPU-related sales growing from about $12B this year to more than $135B by 2028, with TPUs costing roughly 30% less than Nvidia GPUs. Advanced Micro Devices (AMD) — its 5th Generation EPYC processors are the host CPUs for the new TPU 8tx system. Broadcom (AVGO) — custom-silicon peer; the TPU ramp validates the custom-AI-chip market but also sharpens competition for hyperscaler accelerator budgets.