CoreWeave CEO Mike Intrator on GPU economics, cloud consolidation, and why old chips still monetize

CoreWeave CEO Mike Intrator on GPU economics, cloud consolidation, and why old chips still monetize

CoreWeave is no longer pitching GPUs. Mike Intrator, co-founder and CEO, says 90% of the company's clients now use two or more of its products, and roughly 78% use three or more. The shift from raw compute vendor to full AI cloud is the frame he returns to repeatedly, and the launch of Forge at CoreWeave's Fully Connected conference is the product embodiment of that pitch. Forge consolidates software tools and infrastructure rails aimed at enterprise customers who don't have frontier-lab-level AI expertise in-house. The conference drew over 5,000 attendees against an expected 2,000.

“Compute is going to drive intelligence that is going to permeate every asset of civilization... The A100s — that's 2020 architecture — getting contracted out through 2029. Like, that kind of puts that debate to close. There are use cases that will absorb this, whether it's batch computing, or medical research... 90% of our clients use two or more of the products that we deliver.”

GPU obsolescence

Intrator has spent two years pushing back on the depreciation narrative, and he's claiming vindication. His argument is that the most capable GPUs are needed to build at the frontier, but once that building is done, a long tail of other use cases absorbs older hardware productively. His clearest data point is A100s, 2020-era architecture, now contracted out through 2029. Batch computing, medical research, and applications that haven't been invented yet because there isn't enough accessible compute are all cited as absorbers of aging silicon.

Cloud consolidation

On the question of when smaller, undercapitalized neo-clouds consolidate or fail, Intrator sees two distinct pressure points. One is an air pocket in compute demand. The other, and arguably sharper, is power scarcity: as access to power tightens, buyers will gravitate toward established providers rather than take a balance-sheet risk on a third-tier vendor. Managing dozens of smaller cloud relationships also creates drag on buyers' own balance sheets, which accelerates the flight to scale. Intrator doesn't put a number on when consolidation happens or how many players survive, "whether it's ten or seven, I don't know," but treats it as structurally inevitable in any capital-intensive market. CoreWeave's response is to build resilience through enterprise diversification, software depth, and client quality rather than wait for the cycle to play out.

Compute pricing and the time-horizon market

On the question of revenue-per-gigawatt, Intrator reframes the discussion around time horizons rather than a single pricing trajectory. His view is that two distinct markets are emerging. Beyond the roughly two-year build cycle for a new data center, there is a market for longer-duration committed capacity where buyers who plan ahead can lock in better economics, giving cloud providers the visibility to plan and price more efficiently. Inside that build cycle, it becomes a pure supply-demand squeeze, which is the dynamic Elon Musk and Jensen Huang were doing rough math around when discussing figures in the $60–70 billion per gigawatt range. Intrator sees those as separate markets that will increasingly behave differently over time.