Meta & Google TPU spending
Meta is reportedly committing $60B+ to AMD and also buying Google TPUs in multi‑billion dollar deals, signaling hyperscaler GPU+TPU co‑investment argued. That reinforces a multi‑vendor procurement trend among the largest AI buyers argued.
The AMD pact commits Meta to deploy up to 6 gigawatts of Instinct GPU capacity valued at roughly $60 billion over a multi‑year term ([bloomberg.com)], and the agreement includes a performance‑based 160 million‑share warrant that could let Meta acquire up to about 10% of AMD stock if milestones are met ([jonpeddie.com)]. AMD has said the program will begin with an initial 1‑gigawatt deployment in the second half of 2026 and that each additional gigawatt carries a “double‑digit billions” price tag, according to CEO Lisa Su’s comments and follow‑up coverage ([cnbc.com)]. Separately, Meta agreed to a multibillion‑dollar, multi‑year rental of Google Cloud TPUs to train and run next‑generation models, a deal first reported by The Information and picked up by Reuters; reporting also says Meta may move from rented capacity to outright TPU purchases as early as 2027. ([siliconangle.com)] Earlier in the month Meta announced a multiyear agreement with NVIDIA to buy “millions” of Blackwell and Rubin GPUs and to deploy standalone Grace CPUs across its hyperscale clusters, with Nvidia describing joint codesign work and no public price disclosure on the pact ([investor.nvidia.com)]. Meta’s own filings show the company guided 2026 capital expenditures of $115–$135 billion, up from $72.22 billion in full‑year 2025—funding that underpins simultaneous, multi‑vendor hardware commitments across GPUs, CPUs and cloud‑leased accelerators ([investor.atmeta.com)]. Industry coverage frames the sequence—NVIDIA expansion, the AMD 6GW/warrant package, and the Google TPU rental—as a hyperscaler playbook to lock capacity while diversifying suppliers, a pattern noted across Axios, The Information and trade outlets reporting on the three deals. ([axios.com)]
