CoreWeave Doubles Revenue, Takes Backlog to $104 Billion and Signals Capex of Up to $39 Billion by 2026

CoreWeave reported revenue of $2.58 billion in the second quarter of 2026, an increase of 112% year-over-year, exceeding the $2.56 billion that analysts expected. The adjusted loss per share was $1.03, narrower than the projected $1.22. The stock rose 14% in after-hours trading on August 11, and sustained gains on August 12.

What drew the most attention was not the quarterly revenue, but the backlog. The company raised its future contract portfolio to $104 billion and indicated it has already signed another $25 billion in new commitments in the first weeks of the third quarter. The annual forecast was revised upward for revenue, adjusted operating income, and notably, for capex, which is now projected to be in the range of $35 billion to $39 billion.

Michael Intrator, CEO, described this moment as "an important turning point where our scale begins to translate into expanding operational leverage." This statement is significant. Until this quarter, the opposing thesis towards CoreWeave argued that the company was renting expensive GPUs with tight margins. Management's interpretation is that the density of contracts allows rapid dilution of fixed costs.

More Aggressive Guidance than Market Expected

CoreWeave raised its revenue projection for 2026 to a range of $12.4 billion to $13.2 billion, with adjusted operating income between $960 million and $1.15 billion. Capex was the figure that captivated the buy-side the most. By raising it to $35-39 billion, it signals that there is demand to fill virtually all the metal that can fit in 2026, and that NVIDIA, CoreWeave's main GPU supplier, has customers willing to pay for the next Blackwell tape-out.

The Other Side of the Balance Sheet

The backlog is the most discussed variable among analysts. Techtimes estimates that the portfolio jumped $29.6 billion in six weeks and is nearing $129 billion when counting the commitments signed in early August. This number exceeds any traditional valuation multiple for the company. Additionally, it is concentrated: Microsoft continues to anchor the portfolio, while Meta, OpenAI, and a growing group of banks have joined the base over recent quarters, though the top contract still hinges on a single giant.

Analysts cited by Techtimes assign a probability of over 50% of default in a severe stress scenario, such as the cancellation of a multi-year contract by a dominant buyer. CoreWeave's management counters that diversification is advancing, with American investment banks that have spent years focusing on private cloud now contracting external GPU cloud to scale internal LLM workloads.

Market Interpretation

In the United States, CoreWeave's results, combined with those of SuperMicro during the same window, support the thesis that the demand for AI infrastructure has not peaked. The Nasdaq rose 0.54% at the close on August 12, with CoreWeave and SMCI leading the infrastructure group. This adds pressure on CIOs from the Global 2000 who still consider GPU as an optional expense. If CoreWeave's marginal buyer is an American investment bank, the annual investment measure in proprietary stack versus external leasing is expected to change.

In Europe, CoreWeave has opened operations in Amsterdam and London and is negotiating expansion into Germany, alongside pressure from the European Commission for data sovereignty. For banks like Deutsche Bank and UBS, which already contracted regional GPU cloud in 2025, the dilemma is choosing between the cost efficiency of CoreWeave, which has a global scale, and compliance with local sovereign cloud regulations. The announced backlog suggests that the company has the capacity to cater to both sides, provided it secures datacenter space within the EU.

In Brazil, the effect is indirect yet significant. Banks and consultancies consuming GPU from OpenAI and Anthropic essentially end up, at the bottom of the stack, purchasing compute from CoreWeave, as Anthropic booked significant capacity from the neocloud in 2025. The increase in capex and the speed of the backlog reduce the risk of delays for new generative AI workloads contracted by Itaú, Bradesco, or Nubank, but also solidify the country’s reliance on a supply chain with a single bottleneck: NVIDIA.

What Intrator did not disclose during the call was the timeline for converting the backlog into cash flow. A ten-year contract that requires capex now, revenue later, and amortization over a long period is a structure that makes sense in project mathematics, but the economics of a company that went public in March still needs to prove that this curve can withstand a credit tightening cycle.