Joining the AI chip race, Google eyes CoreWeave’s Nvidia GPUs as demand outpaces supply and rivals lock in billion-dollar chip deals.
Google is reportedly in advanced talks to rent Nvidia’s latest Blackwell GPUs from cloud infrastructure firm CoreWeave, according to The Information. Sources familiar with the matter indicate that the tech giant is also exploring deals with other GPU providers to enhance its AI computing capacity.
Despite being a major buyer of Nvidia chips, Google is struggling with supply constraints. During the company’s most recent earnings call, CFO Anat Ashkenazi acknowledged the issue, stating that demand continues to exceed available capacity.
She noted Google is significantly increasing capital expenditure, which is expected to reach $75 billion in 2025, largely focused on data centres and AI infrastructure.
The potential agreement with CoreWeave is reported to be smaller than Microsoft’s $10 billion deal and OpenAI’s $12 billion contract. However, the scale could grow in the future.
In addition to GPU rentals, Google and CoreWeave are in early discussions about a separate arrangement. This deal may allow Google to lease space in CoreWeave’s data centres to host its custom tensor processing units (TPUs).
However, neither Google nor CoreWeave have commented publicly on the matter.
CoreWeave did not list Google among its customers in its recent IPO filing. Its main clients include Microsoft, Meta, IBM, Nvidia, Cohere, and Mistral AI. Microsoft accounted for 62 per cent of CoreWeave’s revenue in 2024.
Earlier reports in March suggested Microsoft might reduce some contracts, though this was denied. Still, CoreWeave’s IPO fell short of expectations, raising $1.5 billion against earlier projections of up to $4 billion.
Google’s push for more AI chip access reflects the growing competition among tech firms racing to secure resources in a constrained market.
