CoreWeave hits $5 billion in revenue, plans 8 GW of power by 2030

Growth, footprint and customers

In the letter, written after CoreWeave's first calendar year as a public company, Intrator describes the company as "The Essential Cloud for AI™". According to CoreWeave, its footprint now spans more than 850MW of active power across 43 data centers globally. The company also says nine of the leading ten model providers rely on CoreWeave Cloud.

Customer concentration has fallen sharply, by the company's account. CoreWeave Cloud customers grew by approximately 150%, and no single customer now represents more than 35% of the revenue backlog, compared with 85% at the start of the year.

Backlog and financing

The revenue backlog expanded to $66.8 billion, up from $15 billion. Weighted average contract length extended from four years to five. Intrator frames this as the result of a financing approach built on risk management: sizing opportunities carefully, building against what the company can see, and never outrunning its contracts.

• CoreWeave secured approximately $18 billion of debt and equity capital.
• It reduced its weighted average cost of debt by more than 300 basis points.
• According to the company, that equals roughly $700 million in annualized savings based on its year-end debt balance.

Intrator says the company tied capital deployment directly to long-term contracted demand and created "a new asset class in the process". He adds that CoreWeave intends to keep innovating in the capital markets and to unlock new sources of financing at lower costs.

Platform, acquisitions and hardware

The letter describes the CoreWeave Cloud as spanning the full stack: GPUs and CPUs, purpose-built networking, storage architected for AI throughput, security isolation, orchestration, tooling and application-layer services.

CoreWeave completed targeted acquisitions to widen that stack. With Weights & Biases, it brought model development, evaluation and monitoring onto its platform, and the combined teams launched a new inference service and tools for building and operating agentic applications. The acquisition of OpenPipe moved the company into reinforcement learning. It also introduced CoreWeave ARENA™, a lab where customers can benchmark real-world performance before moving workloads into production. CoreWeave Mission Control™ underpins the platform, combining security, observability and operational expertise.

On hardware, the company points to its early deployment of NVIDIA GB200 and GB300. It says it was the first cloud to achieve Exemplar Cloud status for GB200, and that SemiAnalysis named it the sole Platinum AI cloud twice.

Inference and the buildout to 2030

Looking ahead, Intrator argues that inference is where the economic returns come from: "If training builds models, inference drives the economic returns." The shift to inference and agentic workloads is expected to bring new services and developer tooling, performance improvements and the ability for customers to operate across multiple cloud environments. That will require more compute capacity in next-generation systems and across the installed base.

The company calls this a sustained buildout rather than a single-generation cycle, with existing infrastructure running alongside new deployments from Ampere to Hopper to Blackwell to Rubin. To meet demand, CoreWeave aims to add more than 5 gigawatts of capacity and reach 8 plus gigawatts of active power by 2030. It plans to look at existing regions and new markets and to build from the ground up to control its physical and financial inputs.

Intrator also stresses engagement with local leaders, regulators and utilities, and lists construction and operations jobs, workforce programs and sustained tax revenue among the expected benefits for host communities.

What it means

For customers and the market, the letter signals that CoreWeave is positioning itself as a large, diversified AI infrastructure provider with long contracts and falling financing costs. All figures come from the company's own shareholder letter, and the 2030 capacity target is an objective rather than a result.