CoreWeave CEO Michael Intrator cites ‘sold out’ capacity as revenue more than doubles and backlog swells to $104 billion

CoreWeave reported Q2 revenue of $2.58B, up 112% YoY and just above analysts’ $2.56B estimate, while backlog jumped 246% YoY to $104.2B and management said near-term capacity is effectively sold out. Despite the demand, the company posted a $626M net loss, driven by $640M net interest expense, though it generated $128M of adjusted operated income excluding certain costs. Shares surged more than 14% after hours, but investors remain split over heavy AI capex and potential write-down risk from a $46.7B property-and-equipment base dominated by GPUs, alongside increasing data-center building restrictions in 18 states.

Analysis

The immediate winner is not the stock that printed the release, but the chip-and-financing stack behind it. Persistently tight GPU rental pricing is the clearest near-term read-through for NVDA: it supports both utilization of the newest parts and resale value of the installed base, while also signaling that scarcity is still being monetized rather than commoditized. The deeper second-order winner is the capital stack around AI buildout — APO, BX, BLK, and BAM can harvest fees, spreads, and AUM from financing assets that customers are unwilling to fund entirely on balance sheet.

The risk is duration mismatch. These businesses are buying or financing rapidly depreciating assets with fixed funding costs, so the equity story only works if inference demand stays elevated long enough for cash yields to outrun interest and obsolescence. In the next 1-3 months, the key catalyst is whether management teams across the AI supply chain keep raising capex and backlog guidance; over 6-18 months, the falsifier is a slowdown in booking conversion, rising financing spreads, or a next-gen chip cycle that forces earlier write-downs.

Consensus is probably over-indexing on the headline strength and underweighting how much of the upside accrues to vendors and lenders rather than leveraged operators. If demand is truly ‘sold out,’ the market should reward the scarce input providers, not chase the most balance-sheet-intensive intermediary. I’d treat CRWV as a tactical momentum name, but a poor long-duration compounder until interest expense falls as a share of revenue and property/equipment stops growing faster than cash generation.

Key Decisions for Investors

• Long NVDA / short CRWV for 1-3 months: express the view that scarcity and pricing power accrue to the supplier with the best balance sheet, while the operator remains exposed to leverage and depreciation. Falsify if CRWV demonstrates sustained FCF conversion and net interest normalizes.
• Add to BX/APO/BLK/BAM on pullbacks over the next 1-2 weeks: AI infrastructure financing should be fee-rich and capital-light relative to the underlying buildout. Best risk/reward is as a basket, sized modestly.
• Avoid chasing CRWV after the post-earnings move; if long exposure is required, use a call spread rather than stock to cap downside from any backlog-to-cash conversion disappointment over the next quarter.
• Watch NVDA and the mega-cap cloud names into the next earnings cycle: if hyperscalers keep capex elevated but neocloud returns stay leveraged, the market may rotate from ‘AI adoption’ to ‘who captures the economics,’ which is a relative-value bearish setup for CRWV.