Google TPUs rented to Meta

The Information first published the account on Feb. 26, 2026 and the story was picked up and republished by Reuters and other outlets within hours. (investing.com) (investing.com) Sources cited in subsequent coverage said access to Google Cloud TPU capacity could begin as early as 2026, with options to acquire and install TPUs in Meta’s own data centers starting in 2027. (finance.yahoo.com) (finance.yahoo.com) (dataconomy.com) (dataconomy.com) Google’s latest TPU family — marketed as “Ironwood” or TPU v7 — is listed in Google documentation as scaling up to 9,216 chips per superpod and offering 192 GiB of HBM per chip in its published spec table. (docs.cloud.google.com) (docs.cloud.google.com) Alphabet has already struck a far larger TPU supply pact with Anthropic — announced in Oct. 2025 — giving Anthropic access to up to one million TPUs and “well over” 1 GW of compute capacity by 2026 under a deal described as worth tens of billions. (anthropic.com) (anthropic.com) (datacenterdynamics.com) (datacenterdynamics.com) Recent supply‑chain reporting flagged next‑generation HBM4 memory and TSMC wafer capacity as chokepoints this year, with analysts warning potential HBM4 shortfalls could push back Nvidia’s Rubin (aka Vera Rubin) GPU ramp while TPU orders compete for TSMC capacity. (benzinga.com) (benzinga.com) (msn.com) (msn.com) Nvidia’s public messaging this cycle reiterated confidence in its GPU roadmap — a company spokesperson said Nvidia’s GPUs are “a generation ahead” — while company posts noted they were “delighted” by Google’s progress as the market diversifies. (cnbc.com) (cnbc.com) (stocktwits.com) (stocktwits.com) Google’s move to commercialize TPUs through large cloud contracts has accelerated since late 2025, with multiple multi‑year deals cited by industry outlets as evidence that TPU capacity is no longer just an internal asset but a commercial product for hyperscalers and AI builders. (dataconomy.com) (dataconomy.com) (siliconangle.com) (siliconangle.com)