CoreWeave Posts $2.1B Revenue With $99B Backlog

The AI infrastructure company more than doubled revenue in Q1 2026, but widening losses and record capital expenditure are redefining what growth looks like in the age of hyperscale AI.

Key Takeaways

• CoreWeave’s Q1 2026 revenue of $2.078B represents 112% year-over-year growth, underpinned by a $99.4B contracted backlog that offers rare long-term revenue visibility in the technology sector.
• A net loss of $740M, driven by $536M in net interest expense, and $6.8B in quarterly capital expenditure illustrate the structural tension between explosive demand and the financial discipline required to meet it at scale.
• Strategic deals with Meta and Anthropic, combined with a $2B equity investment from NVIDIA, signal that CoreWeave has secured a position at the center of AI infrastructure, though execution risk remains the defining variable for investors.

A Record Quarter Built on Contradiction

The numbers CoreWeave released on May 7 told two stories simultaneously. Revenue of $2.078B, up 112% from $982M a year earlier, confirmed the company’s standing as one of the fastest-growing technology businesses of its generation. A contracted backlog of $99.4B, representing a 284% increase year-over-year, offered a degree of forward visibility that most public companies can only approximate. Active power capacity surpassed one gigawatt.

Then came the other side of the ledger. Net loss widened to $740M, with $536M in net interest expense accounting for the bulk of the damage. GAAP operating loss expanded to $144M from $27M. Adjusted operating income collapsed from $163M, at a 17% margin, to just $21M at a 1% margin. Adjusted EBITDA reached $1.157B, maintaining a 56% margin that reflects genuine operational efficiency, but the distance between that figure and GAAP profitability grows wider each quarter.

For investors accustomed to reading infrastructure businesses through conventional profitability metrics, CoreWeave’s results require a different frame entirely. This is a company deploying capital at a rate that precedes revenue by design, and the backlog is the intended proof that the investment is rational. The question is not whether demand exists. It manifestly does. The question is whether the financial architecture supporting that demand can sustain the weight being placed upon it.

The Infrastructure the AI Industry Actually Needs

CoreWeave was founded in 2017, went public on Nasdaq in March 2025, and has since positioned itself as what it calls “The Essential Cloud for AI.” The framing is deliberate. Unlike hyperscalers that treat AI compute as one workload among many, CoreWeave has built its entire platform around high-performance GPU clusters optimized for training, inference, and the emerging generation of agentic systems.

That specialization is increasingly valuable. As AI development matures, the center of gravity has begun shifting from large, periodic training runs toward continuous inference at scale. Production deployments, real-time applications, and autonomous agents demand infrastructure that is not merely powerful but responsive, reliable, and purpose-built. CoreWeave, operating between the foundation models and the underlying silicon, has positioned itself to capture that transition.

CEO Michael Intrator characterized Q1 as “the strongest bookings quarter in CoreWeave’s history,” and the customer roster supports the claim. The company executed multiple new agreements with Meta, including a $21B commitment signed in March. A multi-year deal with Anthropic will support the Claude family of models. Cohere, Mistral, Perplexity, Jane Street, and World Labs deepened existing relationships. These are not exploratory contracts. They represent committed capital from organizations building at the frontier of AI capability.

Product Depth and Platform Ambition

CoreWeave’s commercial momentum is reinforced by product development that extends well beyond raw compute supply. The company was among the first cloud providers designated an NVIDIA Exemplar Cloud for inference on the GB200 NVL72. It launched Flexible Capacity Plans, including Flex Reservations and Spot offerings, and introduced Dedicated Inference products and CoreWeave ARENA for evaluating production workloads under realistic conditions.

Integration with Weights and Biases extended the platform’s relevance to the development lifecycle, particularly for teams building agents and robotics applications. Earlier in 2026, CoreWeave made the NVIDIA HGX B300 available to customers and outlined plans for Rubin platform deployments in the second half of the year. The company also advanced liquid cooling and high-density rack technologies, diversified its infrastructure provider base, and continued geographic expansion to reduce latency for international customers.

These investments reflect a deliberate shift from infrastructure commodity toward full-stack AI platform provider. The distinction matters commercially. Pure infrastructure is subject to price compression over time; platform depth creates switching costs and justifies premium pricing. CoreWeave appears to understand this, and its product roadmap suggests it is building accordingly.

The Cost of Operating at This Scale

Capital expenditure in Q1 reached approximately $6.8B. To finance this level of deployment, CoreWeave secured an $8.5B investment-grade non-recourse delayed-draw term loan facility and closed a $2B Class A common stock investment from NVIDIA. The NVIDIA transaction is strategically significant beyond its financial dimensions: it represents a deep alignment between the world’s dominant AI chip supplier and one of its most ambitious cloud customers.

The liquidity these transactions provide is real. So is the leverage they introduce. Interest expense more than doubled year-over-year, and the trajectory of debt service costs will depend heavily on how quickly active capacity translates into revenue. CoreWeave’s non-recourse financing structures demonstrate financial sophistication, limiting contagion risk across the portfolio. But leverage at this scale demands flawless operational execution, particularly across supply chain, construction timelines, and power procurement.

Contracted power now exceeds 3.5 gigawatts, with more than 400 MW added during Q1 alone. The stated ambition is more than 8 GW by 2030, supported by an expanded NVIDIA partnership targeting over 5 GW of AI factories. These are extraordinary numbers by any measure, and the energy intensity they imply will attract scrutiny from regulators and policymakers as attention to AI’s physical footprint intensifies globally.

What the Market Made of It

Shares fell approximately 6.6% in aftermarket trading on May 7, settling near $132 in extended hours. The reaction reflected investor discomfort with the widened net loss and elevated capital expenditure projections rather than any fundamental deterioration in the business. CoreWeave has traded with meaningful volatility since its IPO, as the market works through the tension between extraordinary growth and the investment required to sustain it.

For long-term investors, the $99.4B backlog is the central data point. It provides a degree of revenue visibility that is genuinely unusual in technology, and it anchors the investment case even when quarterly losses expand. Near-term risks center on execution: GPU supply constraints, power availability, and competitive pressure from hyperscalers investing aggressively in proprietary infrastructure.

The Structural Bet

CoreWeave’s Q1 results are, ultimately, a document of ambition operating at industrial scale. The growth is real, the customer base is serious, and the backlog is substantial. The losses are also real, and the capital requirements are formidable. Neither fact cancels the other.

What distinguishes CoreWeave’s position in the AI infrastructure landscape is its early decision to specialize deeply rather than broadly. As the industry’s requirements become more demanding and more specific, that specialization is proving to be a durable advantage. The coming quarters will determine whether the financial structure supporting that advantage proves equally durable.

There is a broader significance to CoreWeave’s trajectory that extends beyond its own balance sheet. The company’s rise reflects a structural shift in how AI capacity is being built and who gets to build it. Hyperscalers retain enormous advantages in scale and capital, but the complexity of modern AI workloads has created space for specialized providers that can move faster, configure more precisely, and align more closely with the specific demands of frontier model development. CoreWeave has occupied that space with unusual conviction. Whether it can hold and expand it, as hyperscalers sharpen their own inference and agentic offerings, is the central competitive question of the next several years. The $99.4B backlog suggests the market has already formed a view. The operational reality of delivering against it will be the true test.