NEW YORK, March 5, 2026, 14:33 (ET)
- CoreWeave signed a multi-year partnership to run Perplexity’s AI “inference” workloads
- Deal leans on dedicated Nvidia GB200 NVL72 clusters; financial terms were not disclosed
- Shares fell in Thursday trading as investors kept focus on funding and spending
CoreWeave said it has entered a multi-year strategic partnership with AI search firm Perplexity, pitching dedicated Nvidia GB200 NVL72 clusters to handle “inference” — the work of running trained AI models to answer users in real time. The companies did not disclose financial terms. “AI applications running in production require more than just access to raw infrastructure,” CoreWeave senior vice president of revenue Max Hjelm said, while Perplexity chief business officer Dmitry Shevelenko called CoreWeave “an essential partner” in its infrastructure push. 1
The tie-up lands as CoreWeave tries to prove it can keep its newest data centers busy without overbuilding. Shares were down about 6.6% in Thursday trading.
Investors have been rattled by the pace of CoreWeave’s spending, even as the company argues most of it is matched to signed customer demand. Chief executive Michael Intrator told Reuters in late February the company chose “to build faster” to deliver more infrastructure, while D.A. Davidson analyst Alexander Platt said the heavy buildout was “a positive signal” if capacity keeps coming online. 2
Perplexity said it is pursuing a multi-cloud approach, and the partnership with CoreWeave is designed to support new services as usage rises. CoreWeave said it will also deploy Perplexity Enterprise Max internally for staff. 3
Axios reported Perplexity will use specialized Nvidia Grace Blackwell-powered clusters at CoreWeave for inference work, a bet that low-latency performance will matter more as consumer-facing AI tools push into heavier, always-on use. 4
CoreWeave sits in an awkward middle ground: it is not a hyperscaler like Amazon or Google, but it is spending like one, buying high-end chips and power capacity to compete for the same AI workloads. Smaller “neocloud” rivals, including Nebius, are chasing similar customers, often by promising faster access to scarce GPUs.
But the Perplexity deal does not settle the big question hanging over the stock: whether demand stays strong enough — and deployments stay on schedule — to cover a buildout that is capital-hungry and sensitive to financing costs. If customers slow spending, or if data center timelines slip, revenue tied to delivery can move later while bills keep arriving.
CoreWeave has leaned on long-term contracts and new financing to scale, and it has told investors the spending is tied to signed customer commitments. The next few quarters will test whether deal announcements translate into steady revenue and steadier margins, not just bigger hardware.
