CoreWeave CEO says it is locking in older NVIDIA GPUs for AI demand through 2029

CoreWeave CEO Mike Intrator used a pointed example of staying power in artificial intelligence compute to describe current market conditions, telling CNBC that the company is booking 2020-era NVIDIA GPUs through 2029. In the interview, Intrator said the bookings are happening at “full freight,” a phrase he used to convey that CoreWeave is paying for the certainty of supply rather than waiting for better terms later.

The remarks were framed as a response to how quickly technology cycles are often assumed to move in AI infrastructure. Instead of treating older GPU generations as immediately obsolete, Intrator suggested that customers and workloads still find value in those systems, and that demand for accelerated computing remains strong enough to support continued procurement.

Intrator’s comments also described the bookings as part of a broader announcement from the market about where spending is going. In his telling, continued interest in near- and mid-term GPU capacity suggests that AI infrastructure buyers want more than just next-generation hardware, they want capacity that can be delivered reliably over multiple years.

CoreWeave, whose shares trade on the Nasdaq under the ticker CRWV, is positioned as an AI infrastructure provider. While the CNBC conversation focused on GPU supply, the underlying operational model is that the company secures accelerated hardware and delivers compute capacity to customers, a structure that makes long-horizon procurement and logistics particularly important during periods of tight supply.

The “full freight” characterization matters because it implies a willingness to absorb higher shipping or related costs to ensure GPU availability. In other words, it suggests CoreWeave is not merely planning for spare capacity, but actively prioritizing timeline certainty, a choice that typically becomes more visible when AI demand is outpacing what can be quickly sourced and deployed.

For NVIDIA, the episode is a reminder that its data-center GPUs can remain commercially relevant beyond their newest releases, at least in the infrastructure layer where customers may optimize for performance-per-dollar, deployment timelines, and total compute availability. Intrator’s framing also highlights how procurement decisions by AI infrastructure operators can extend the effective lifetime of specific GPU generations in real-world deployments.

The company did not provide additional contract details in the reported interview, such as the number of GPUs involved, whether the bookings include specific NVIDIA product SKUs, or how the terms compare to earlier procurement cycles. It also did not disclose whether the 2020-era classification refers to a single GPU generation or a range of models used across customer workloads. As a result, investors and customers will have to wait for further disclosures to understand the scale and financial implications of the commitment.

Looking ahead, the key question will be whether CoreWeave’s multi-year bookings translate into sustained revenue and utilization as AI capex and cloud capacity planning cycles progress. Markets will also watch how NVIDIA’s product cadence and CoreWeave’s mix of newer versus older GPUs evolve, especially if supply constraints or customer demand patterns change over the next several quarters.