CoreWeave Hits $5.1B Revenue as AI Cloud Backlog Surges to $66.8B

Long-term AI compute contracts and rapid infrastructure scaling are turning a specialist GPU provider into a strategic pillar of enterprise cloud.

Key Takeaways

• CoreWeave hit $5.1B in annual revenue in 2025, growing 168% year-over-year, driven by surging demand from AI labs, hyperscalers, and enterprises deploying GPU-intensive workloads at scale.
• Remaining performance obligations more than quadrupled during 2025, with average contract durations of five years, providing investors rare forward visibility in a volatile sector.
• $14.9B in capital expenditure and $1.2B in net interest expense drove GAAP losses, but management targets 25–30% long-term EBITDA margins as the infrastructure base matures and deployment density improves.

A Record Set in Plain Sight

No cloud infrastructure company has reached $5 billion in annual revenue faster than CoreWeave. That distinction, confirmed when the Livingston, New Jersey firm published its fiscal 2025 results, is more than a marketing milestone. It reflects the structural shift now underway in enterprise computing, where specialized GPU capacity has become as strategically vital as power or bandwidth, and where the companies that control it command pricing, tenure, and influence that general-purpose cloud providers are scrambling to match.

CoreWeave’s full-year revenue of $5.1 billion represented a 168% increase over the $1.9 billion it generated in 2024. Fourth-quarter revenue reached $1.6 billion, up 110% from $747 million a year earlier. These are not the numbers of a company riding a temporary wave. They reflect a firm that has, in a compressed window, built itself into indispensable infrastructure for the institutions training and deploying large-scale artificial intelligence.

The Mechanics of Hypergrowth

To understand what CoreWeave sells, it helps to understand what its customers cannot easily find elsewhere. The company operates GPU clusters optimized for the specific demands of AI model training and inference, workloads that require not just raw compute but low-latency interconnects, precision memory bandwidth, and software stacks tuned for parallelism. Its proprietary Mission Control system, which earned CoreWeave designation as NVIDIA’s first Exemplar Cloud for training on the GB200 NVL72 platform, sits at the operational center of this offering.

That platform served a broadening roster of customers through 2025, encompassing AI laboratories, hyperscalers, and enterprises across industries. Named customers include Cognition, CrowdStrike, Cursor, Mercado Libre, Midjourney, and Runway. Reserved-instance additions doubled in the fourth quarter, a metric that speaks to the depth of commitment rather than opportunistic spot purchases. When customers lock in multi-year contracts at this rate, they are not hedging against a trend. They are building against a conviction.

The Backlog as Evidence

The figure that arguably defines CoreWeave’s investment case is not its revenue, impressive as that is, but its revenue backlog. By December 31, 2025, remaining performance obligations and estimated future revenues from committed contracts reached $66.8 billion, more than quadrupling from the start of the year and up $11.2 billion sequentially in the final quarter alone.

Backlogs of this magnitude, with average contract durations stretching to five years, are unusual in cloud infrastructure, where workloads have historically been elastic and commitments short-dated. Their presence here signals a market in transition, one where AI compute is increasingly treated as a fixed input rather than a variable cost, and where customers are willing to sacrifice flexibility for capacity assurance. For CoreWeave, it translates into revenue visibility that most technology companies, even mature ones, cannot claim.

CEO Michael Intrator described 2025 as “a defining year,” citing intensifying demand from a broader customer base running an expanding set of workloads. CFO Nitin Agrawal framed the backlog’s growth as evidence of disciplined execution, a pointed phrase in an environment where capital deployment and contract conversion are subjects of close scrutiny.

Infrastructure at Scale, and at Cost

Growth at this pace does not arrive cheaply. CoreWeave invested $14.9 billion in capital expenditures during fiscal 2025, including $8.2 billion in the fourth quarter alone, well beyond analyst expectations of roughly $6 billion. The company ended the year with over 850 megawatts of active power capacity, having added 260 megawatts in the final quarter. Contracted power reached 3.1 gigawatts across a diversified provider portfolio, with a stated ambition, in partnership with NVIDIA, of exceeding 5 gigawatts by 2030.

These investments generate real leverage: adjusted EBITDA reached $3.1 billion for the year, a 60% margin, with the fourth quarter delivering $898 million at 57%. But the GAAP picture is considerably more complex. Net losses totaled $1.2 billion for the full year, widening to $452 million in the fourth quarter from $51 million in the year-earlier period, driven primarily by $1.2 billion in net interest expenses on the company’s substantial debt load. To fund continued expansion, CoreWeave raised $2.6 billion through convertible senior notes and extended its revolving credit facility to $2.5 billion. These are the financing instruments of a company building long-cycle physical assets, and the strategy is coherent, but it concentrates risk in the cost of capital.

Platform Depth and Competitive Moat

CoreWeave has spent the past two years broadening its platform well beyond raw compute. The acquisition of Weights & Biases, integrated throughout 2025, brought machine learning experiment tracking and model management into the CoreWeave ecosystem, with Mission Control and W&B Models now jointly powering training diagnostics. Subsequent acquisitions deepened this further: Monolith extended the company’s reach into industrial AI, bridging digital simulations with physical manufacturing environments, while Marimo brought an open-source AI-native Python notebook into the stack.

The launch of Serverless RL, the first fully managed reinforcement learning environment offered as a cloud service, represents a meaningful product innovation for developers training AI agents without the operational burden of managing underlying infrastructure. AI Object Storage arrived as a purpose-built solution for high-throughput AI workloads, paired with zero-egress migration tools to ease multi-cloud transitions. These additions collectively form a coherent argument: CoreWeave is not simply a GPU rental operation, but an AI cloud platform with deliberate stickiness designed into its architecture.

Market Reaction and the Margin Question

Investors, at least in the immediate hours after the results, were unconvinced. Shares closed at $97.63 on February 26, before declining a further 5 to 6 percent in after-hours trading. The adjusted EBITDA figure of $898 million fell short of the $936 million consensus, and the fourth-quarter capital expenditure number prompted concerns about near-term margin trajectory. Management’s guidance for 2026 revenue of $12 to $13 billion was well-received on its face, but the caveat that EBITDA margins would bottom in the first quarter before recovering to low double-digits by year-end introduced a period of uncertainty that equity markets are not well-suited to absorb patiently.

That tension is inherent to the CoreWeave model. The infrastructure being deployed today generates commitments that run five years into the future, but the cost of deploying it lands in the current quarter. Investors accustomed to software-style margin expansion on a quarterly cadence will find the model disorienting until the revenue base matures sufficiently to absorb the fixed cost structure.

The Larger Moment

CoreWeave’s 2025 results are most usefully read not as a company-specific story but as a proxy for where enterprise AI stands. The volume of compute being committed, the duration of those commitments, and the pace at which organizations across industries are entering long-term infrastructure contracts all suggest that AI has moved decisively past its experimental phase. Production deployments, not pilots, are now driving purchasing decisions.

The risks are real and worth naming plainly. GPU supply chains remain tight and susceptible to geopolitical disruption. Power procurement at gigawatt scale is operationally complex and politically contingent. Competition from hyperscalers with deeper balance sheets and established enterprise relationships is not standing still. And the debt burden, while manageable in the current growth environment, leaves limited room for a demand slowdown.

What CoreWeave has built, however, is a business with genuine structural advantages in the fastest-growing segment of enterprise technology. The backlog says so. The customer list says so. The infrastructure scale, built at a pace that has no historical precedent in cloud computing, says so. Whether the financial model matures as gracefully as the operational one is the question that will define the next chapter. The foundation, at least, is not in doubt.