CoreWeave's Cloud Conundrum: Growth at Full Throttle, Losses in Tow

CoreWeave, one of the most prominent new names in AI infrastructure, posted second-quarter revenue of $1.21 billion, up 207% from a year earlier and comfortably ahead of Wall Street's $1.08bn forecast. The surge reflects an AI boom that continues to strain computing supply chains, as enterprises rush to deploy larger and more complex models. Yet despite the top-line beat, the shares fell 10% after hours. This may not reflect disappointment with the results: the stock had already climbed 33% over the past week, suggesting much of the good news was priced in well before the release.

The Livingston, New Jersey-based company operates 33 data centres in the United States and Europe. It provides large-scale, GPU-optimized infrastructure—originally built for crypto mining but now tailored for AI, machine learning, and visual effects—serving major clients like Microsoft, Nvidia, and OpenAI. AI chips remain scarce amid an arms race for AI capacity. Demand visibility appears strong: CoreWeave's order backlog swelled to $30.1bn at the end of June, up from $25.9bn just three months earlier and 86% higher than a year ago. A $4bn contract expansion with OpenAI and fresh hyperscaler deals helped drive the increase, alongside a 600 MW boost in contracted power that takes the total to 2.2 GW.

Not all growth is unconstrained. "Ultimately the most significant challenge right now is accessing power shells capable of delivering the scale of infrastructure that our clients are requiring," said CEO Michael Intrator. To ease the bottleneck, CoreWeave is pressing ahead with a $9bn all-stock acquisition of crypto miner Core Scientific, securing 1.3 GW of existing and future power capacity. The deal faces opposition from Core Scientific's largest shareholder.

The spending needed to build such capacity is steep. Operating expenses jumped to $1.19bn in the quarter from $318m a year earlier, driven by higher infrastructure, technology, and administrative costs. Operating income fell to just $19.2m from $77.7m a year earlier, as interest expenses and capital requirements mounted. Even so, adjusted EBITDA rose to $753m, maintaining a 62% margin.

Analysts at Jefferies described the results as "solid" and highlighted the hyperscaler expansions as proof of "unrelenting demand for high performance compute" and CoreWeave's "best-in-class capabilities." They cautioned that sequential backlog growth was smaller than some investors hoped, as the OpenAI expansion had been signed in May and was already priced into expectations. Still, they raised revenue forecasts for 2026 and 2027, arguing that backlog growth remains closely tied to available power—and CoreWeave is adding capacity at a clip.

The AI workloads driving this demand are themselves evolving. Chain-of-thought reasoning, increasingly used in large models, boosts accuracy but requires far more computation. That dynamic has turned companies like CoreWeave into both the enablers and captives of the AI surge: the more sophisticated the models, the higher the demand—and the higher the infrastructure bill.

CoreWeave now expects full-year revenue of $5.15bn to $5.35bn, up from its earlier $4.9bn to $5.1bn forecast, and reaffirmed its capital expenditure guidance. The share price—up nearly threefold since its March IPO—suggests investors still see plenty of runway. But with customer concentration high, costs surging, and physical capacity a binding constraint, the firm must prove it can turn scale into sustainable margins.