Google Cloud Unveils 8th-Gen TPUs to Rival Nvidia
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Google Cloud unveiled its eighth-generation TPUs at Cloud Next 2026 — the TPU 8t for training and the TPU 8i for inference — with general availability planned for later this year, per Google's own blog and Bloomberg.
- Outlets including CNBC and TechCrunch headlined the launch as Google's "latest shot at Nvidia," while VentureBeat argued Google "doesn't pay the Nvidia tax" because of its custom-silicon strategy.
- Google simultaneously introduced the Virgo Network, a megascale data-center fabric that FundaAI called "the biggest networking upgrade since TPU v4," with major implications for optical networking suppliers.
- Google committed $750 million to accelerate partners' agentic AI development, per a PR Newswire release, layering ecosystem funding on top of the hardware reveal.
- The TPUs were co-developed with Broadcom (per Blockonomi), reinforcing Google's vertical-integration push to reduce dependence on third-party AI accelerators.
- CEO Sundar Pichai used his Cloud Next keynote to frame the announcements around "momentum and innovation at Google scale," with social and enterprise platform launches (per Mercury News and PYMNTS) bundled into the same event.
Why it matters: Google is stacking custom training chips, custom inference chips, a custom networking fabric, and a $750M partner fund into a single keynote — an explicit, vertically integrated alternative to paying Nvidia's margins on the AI compute stack. Enterprise buyers gain a second credible AI hardware vendor as agentic workloads scale, and Nvidia's pricing power faces a more coordinated challenge than any single prior TPU generation.
