In June, Google signed an agreement with SpaceX to lease 110,000 Nvidia GPUs for roughly $920 million per month, starting in October and running through mid‑2029. The total value of the contract approaches $30 billion. SpaceX reconfirmed the deal in its August earnings report, noting that capacity ramp‑up would begin in September and full billing would start in October.
Why this matters
Google spent years developing its own AI accelerator, the TPU (Tensor Processing Unit), and has presented the TPU as a way to reduce dependence on Nvidia. The company has also built one of the world’s largest cloud infrastructures and budgeted nearly $200 billion in capital expenditures for this year.
Despite that, when Google’s frontier models needed more compute, the company did not simply scale up TPU usage; it turned to the Nvidia cluster rented from SpaceX. The size and cost of the lease indicate that, in practice, Google still relies on Nvidia‑based capacity to meet near‑term demand.
Market context
Other industry players continue to deploy TPUs: Anthropic plans to use up to a million TPUs, and Meta has signed a multi‑billion‑dollar deal related to its hardware strategy. Nevertheless, Google’s financing and procurement choices demonstrate that public relations about independence from Nvidia do not always translate into procurement behavior when rapid increases in compute are required.
Brief takeaway
Google designed and promoted the TPU to reduce reliance on Nvidia, but the large SpaceX leasing agreement shows the company is supplementing its in‑house hardware with rented Nvidia GPUs to satisfy immediate capacity needs.
