CoreWeave's backlog tops $104 billion as AI cloud revenue doubles
The AI buildout's scoreboard came in overnight: CoreWeave's revenue doubled and its backlog hit a record, while China locked in its first mandatory safety rules for L3/L4 self-driving.
CoreWeave posted second-quarter revenue of $2.58 billion, up 112 percent from a year earlier and ahead of the $2.56 billion analysts expected, with the stock jumping about 10 percent in extended trading. The AI cloud provider's revenue backlog — contracted future business — climbed to $104.2 billion, management raised its full-year 2026 outlook to $12.4 billion–$13.2 billion and lifted its year-end active-power target past 1.85 gigawatts. The company also signed new capacity deals with Meta and Anthropic during the quarter, and CEO Michael Intrator framed the results as an inflection point where scale is finally translating into operating leverage, with near-term capacity effectively sold out.
The honest counterweight is the balance sheet: the net loss widened to $626 million from $290 million a year earlier, and CoreWeave carried $35 billion in debt at quarter end to pay for Nvidia GPUs and data centers. That's the neocloud bargain in one line — demand is real and compounding, but the money to build ahead of it is borrowed, and the bill comes due before the backlog converts to profit. It's the same demand signal Supermicro flashed in its own record quarter — Supermicro's record year shows AI server demand is still booming — and the clearest evidence yet that the AI buildout's appetite for compute is outrunning the supply chain's ability to deliver it.
China's market regulator approved 338 national standards including the country's first mandatory safety requirement for L3 and L4 autonomous driving, alongside new quality standards for AI medical devices. The autonomous-driving standard, GB 44721-2026, takes effect in July 2027 and applies to passenger and cargo vehicles with conditional or high automation: it sets liability boundaries between automaker and driver, requires a full safety-case archive for every model, and bars companies from using production owners as de facto test subjects to feed their systems. The AI medical device standards give developers and regulators a shared framework for quality evaluation and testing. It's Beijing writing the rules where AI safety stops being abstract — who is liable when the car drives, and whether medical AI can be trusted at the bedside.
What to watch: whether the neoclouds can convert their record backlogs into profits before the debt service bites — and how China's new rules reshape the self-driving liability debate.
The buildout keeps beating expectations — what breaks first, compute supply or the balance sheets financing it? Tell us in the comments.
