Perplexity signs a multi-year deal to run AI workloads on CoreWeave's infrastructure

Perplexity, the AI search company, signed a multi-year deal to run its AI workloads on CoreWeave’s cloud platform, and investors took notice, pushing CRWV shares up roughly 5.7% in pre-market trading.

The deal puts Perplexity on NVIDIA GB200 NVL72-powered clusters through CoreWeave’s infrastructure. Those clusters will carry the load for Perplexity’s fast-growing AI products, along with its Sonar and Search API services.

CoreWeave is also bringing Perplexity Enterprise Max into its own offices. Staff will use it to search the web, pull from internal knowledge bases, run multi-step research, look through data, and tap into advanced AI models, all from a single place.

Perplexity has already started running workloads through CoreWeave’s Kubernetes service as part of its first deployment phase. It is also using W&B Models to train, fine-tune, and manage its models from early testing through to live production. The move fits Perplexity’s wider strategy of spreading its infrastructure across more than one cloud provider, while adding to CoreWeave’s growing list of AI clients running at production scale.

Max Hjelm, CoreWeave’s SVP of Revenue, said production AI demands more than raw computing power. “AI applications running in production require more than just access to raw infrastructure; they require best-in-class performance and reliability as well as a cloud platform designed end-to-end for AI that simplifies compute operations,” he said. Perplexity’s Chief Business Officer Dmitry Shevelenko called CoreWeave an “essential partner” for where the company is headed.

Looking ahead, CoreWeave is planning to spend between $30 billion and $35 billion on capital expenditures in 2026, a sharp jump from $10.31 billion in 2025. It wants to hit more than 1.7 gigawatts of active power by year-end, ahead of the analyst consensus sitting at 1.59 gigawatts, and grow beyond five gigawatts past its contracted footprint by 2030.

The partnership gives CoreWeave a high-profile new customer outside its Microsoft/OpenAI concentration problem, fresh ammunition for diversification, and a premarket stock bump, all before the investor conference.

Co-Founder and Chief Development Officer Brannin McBee will speak at the Morgan Stanley TMT Conference in San Francisco on Wednesday, March 4, 2026, starting at 4:05 p.m. Eastern. Vice President Nick Robbins will follow at the Cantor Global Technology Conference in New York on Tuesday, March 10, 2026, at 2:30 p.m. Eastern.

As previously reported by Cryptopolitan, Nvidia has put $2 billion into CoreWeave, picking up Class A shares at $87.20 each. CEO Mike Intrator said the money will help the company “accelerate our build” and spread its customer base. “This will lead to continued diversification,” he said.

CoreWeave makes its money by renting out GPU-heavy computing capacity from its data centers, the kind of muscle companies need to train AI models and keep them running. That puts it in a growing class of cloud providers built for one thing: powering AI.

Want your project in front of crypto’s top minds? Feature it in our next industry report, where data meets impact.