CoreWeave Q2 2026 earnings surge

CoreWeave's headline numbers were not a marginal beat - revenue hit $2.575-$2.58 billion, up 112% year-over-year and nearly 24% sequentially, landing near the top of the company's own guidance range ^[1]. Adjusted EBITDA reached $1.5 billion at a 59% margin, more than doubling from a year earlier, while adjusted operating income jumped to $128 million from just $21 million in Q1 2026. CEO Mike Intrator called it an exceptional quarter, and CFO Nitin Agrawal described demand as exceptionally strong, with multiple customers waiting for each GPU brought online ^[2].

That demand imbalance is what let CoreWeave raise prices roughly 25% across its SKUs in July while telling investors its near-term capacity is effectively sold out ^[3]. On the customer side, Meta layered an additional $21 billion commitment (through 2032) on top of a prior $14 billion deal, and Anthropic signed a new multi-year agreement to supply compute for its Claude models. Those two deals were the primary drivers behind a revenue backlog that grew 246% year-over-year to $104 billion ^[4]. In other words, the beat wasn't just an accounting surprise - it reflects real pricing power in a market where GPU supply still can't keep up with AI compute demand.

A revenue backlog is not cash in the bank - it's the total value of contracted future work customers like Meta and Anthropic have committed to, which CoreWeave will only recognize as revenue as it delivers capacity over the life of those contracts. A bigger backlog is a genuinely bullish signal about future demand, but it says nothing about whether the company can fund the data centers and GPUs needed to fulfill it in the meantime - and that funding gap is where the story gets uncomfortable.

CoreWeave's GAAP net loss widened to $626 million in Q2, up from $290 million a year earlier, even as revenue more than doubled. The bigger flag is net interest expense, which jumped roughly 140% year-over-year to $640 million - and management guided Q3 interest expense even higher, to $860-$940 million ^[5]. That's the direct cost of financing a data center buildout largely with debt: as CoreWeave borrows more to build capacity ahead of revenue recognition, the interest bill scales with the debt load, not with how fast the backlog converts to cash. It's exactly this dynamic that led DA Davidson to argue CoreWeave's roughly 9% cost of capital still exceeds the implied returns on new capacity, calling the buildout value-destructive rather than value-creating ^[5].

Wall Street's equity analysts read the quarter as a genuine turning point. Citi kept its Buy rating and a $142 price target, calling it one of the cleanest quarters since CoreWeave's IPO and predicting shares should move meaningfully higher on improving profitability. Wells Fargo raised its target to $160 from $155 with an Overweight rating, JPMorgan said the results delivered on the margin performance investors were specifically watching for, and Morgan Stanley called it a solid quarter across demand, new commitments, and pricing ^[8].

Credit markets and short sellers tell a different story. Just two weeks before the report, CDS markets had priced roughly a 50% five-year default probability on CoreWeave's debt, with spreads topping 855 basis points ^[6], and a $2.6 billion loan facility only found buyers after CoreWeave agreed to a 100-125 basis point spread increase and tighter maintenance covenants ^[7]. That tension showed up in how retail investors read the same numbers: X/Twitter reaction to the print was almost uniformly bullish, fixated on the beat-and-raise headline, while Reddit's response was more skeptical and more technical - one widely discussed thread pointed out that most of CoreWeave's debt is financed through special-purpose vehicles collateralized by customer contracts and serviced by that contracted revenue rather than CoreWeave's own free cash flow, a structural nuance bulls use to argue the debt load is safer than the headline dollar figures suggest, and one that got little attention in mainstream coverage of the earnings beat.

CoreWeave raised its full-year 2026 revenue guidance to $12.4-$13.2 billion and set Q3 revenue guidance at $3.45-$3.6 billion, alongside a new exit ARR target of $18.5-$19.5 billion ^[2]. To hit those numbers, management is leaning further into physical buildout: the company added nearly 500 megawatts of active power in Q2 alone, taking active capacity to 1.5 gigawatts, and pushed total contracted power to about 4.2 gigawatts by mid-August ^[2]. That expansion is expensive - Q2 capex alone was $9.4 billion, and full-year capex guidance was raised to $35-$39 billion ^[3]- which means the interest-expense trajectory described above is unlikely to reverse soon, even as the backlog keeps growing (it's already up another roughly $25-30 billion in new commitments in the six weeks since quarter-end, according to one estimate) ^[10].

The bigger unknown is competitive and pricing risk. Independent analysis cited GPU rental rates down 50-70% from peak by mid-2026, and flagged new entrants including Meta's own compute buildout and SoftBank's SB Neo as potential threats to CoreWeave's pricing power and backlog economics going forward ^[5]. For now, the market is voting with its feet - CoreWeave gained 19% and peer neocloud Nebius surged 34% in a broader post-earnings 'neocloud rally' ^[9]- but that rally is a bet that demand keeps outrunning supply long enough for CoreWeave to out-earn its debt, not proof that it already has.