New York, March 10, 2026, 4:23 PM EDT.
CoreWeave unveiled fresh flexible capacity options Tuesday, launching both Flex Reservations and Spot instances as it steps up efforts around inference—when AI models process real-time queries. Shares ticked up roughly 0.7%.
The timing is tricky. Back on Feb. 26, CoreWeave told Reuters it plans to ramp up capital spending in 2026 to $30 billion–$35 billion, compared with $14.9 billion slated for 2025. The company is pushing hard to grow its AI data centers and chip operations, but management flagged that the accelerated build will squeeze margins in the near term.
So, pricing and usage go beyond just what’s on offer. Inference workloads are unpredictable—demand can jump without warning, not like the steadier flow of training tasks. Customers have to choose: lock in extra capacity, or gamble on possible slowdowns. CoreWeave wants to smooth out that rollercoaster, aiming for more consistent revenues.
Flex Reservations, according to the Livingston, New Jersey-based company, gives customers a guaranteed price cap, charging standard usage rates solely for active instances. The Spot option—available now—offers cheaper, interruptible capacity, useful for jobs like batch analytics and backfills. Flex Reservations remains in preview for select customers.
“Infrastructure planning becomes as critical as deployment” after AI services go live, according to Chen Goldberg, CoreWeave’s executive vice president of product and engineering. The company is positioning the new options as a shift away from locked-in 24/7 reservations, pitching them instead as a way to better align cost and certainty with actual demand. CoreWeave
Steven Dickens, chief executive and principal analyst at HyperFrame Research, said CoreWeave is zeroing in on GPU costs as it pivots from training frontier models to rolling out more enterprise-focused deployments, where “buying characteristics change.” But Dickens noted CoreWeave still needs to show its approach can hold up against Amazon Web Services, Microsoft Azure and Google Cloud, particularly as those big providers keep expanding their inference options. DataCenterKnowledge
CoreWeave insists demand isn’t the issue here. Its revenue backlog ballooned to $66.8 billion by the end of 2025, up sharply from $15.1 billion a year before, according to Reuters. CFO Nitin Agrawal noted that the majority of this year’s capex is linked to existing customer agreements.
The line between chasing growth and delivering returns is still razor-thin. After the earnings report, AJ Bell’s Russ Mould told Reuters that while investors get the need to scale up, doubts linger about long-term profitability—and how the company plans to finance the whole expansion. CEO Michael Intrator acknowledged to Reuters the accelerated growth push would bring “some short-term pressure on the margins.” Reuters
Tuesday’s rollout isn’t just another feature drop. CoreWeave needs sharper pricing to drive higher utilization on its costly GPU fleets—and bring in extra enterprise inference jobs. That could take some heat off its aggressive spending plans. But if this doesn’t stick, those capex concerns are likely to resurface quickly.