Nebius jumps; CoreWeave faces scrutiny

AI cloud is starting to look less like one big boom and more like a sorting machine. Everybody can say “we have GPUs.” But the market is getting pickier about who can turn that hardware into usable, predictable, production infrastructure. That is why Nebius and CoreWeave landed so differently this week. Nebius put up a blowout quarter on Wednesday, May 13. CoreWeave, after reporting a week earlier, is still getting pressed on whether scale alone is enough. ### What did Nebius actually report? Nebius said first-quarter 2026 revenue hit $399.0 million, up from $50.9 million a year earlier — a 684% jump. Adjusted EBITDA swung to positive $129.5 million from a loss a year ago. It also said it secured up to 1.2 gigawatts of power and land for a new owned AI factory in Pennsylvania. That is why the stock reaction was so strong — the company showed both demand now and expansion plans for later. (morningstar.com) ### Why did investors like that so much? Because this was not just “AI demand is big” hand-waving. Nebius showed revenue scale, margin improvement, and a concrete power pipeline in the U.S. Those three things matter in this business. GPUs are scarce, power is scarcer, and profitable growth is the real proof that customers are using the platform hard enough to matter. (morningstar.com) ### So why is CoreWeave under more scrutiny? CoreWeave’s quarter was huge on the surface. Revenue reached $2.078 billion, up from $982 million a year earlier. Revenue backlog hit $99.4 billion. It also highlighted a new $21 billion Meta commitment and said it had passed 1 GW of active power. But the catch is the company still posted a $740 million net loss and $536 million in net interest expense. That makes investors ask a different question — not “is demand real?” but “how expensive is this growth to finance and sustain?” (morningstar.com) ### What is CoreWeave trying to fix? Basically, CoreWeave is trying to prove it is more than a GPU landlord. In March it launched Flexible Capacity Plans — including Flex Reservations and Spot — to match how AI workloads actually behave. The key pitch is that inference demand is uneven. Teams do not want to pay for full-time peak capacity if traffic comes in waves. CoreWeave is trying to sell a more cloud-like answer: guaranteed headroom when needed, cheaper interruptible capacity when not. (investors.coreweave.com) ### Why does inference change the game? Training is lumpy but somewhat plannable. Inference is messier. Once a model is live, traffic spikes, latency targets tighten, and cost discipline matters every hour. That is why CoreWeave keeps emphasizing Dedicated Inference, which gives customers control over GPU type, runtimes, scaling, and reserved capacity, plus OpenAI-compatible endpoints. In plain English — customers want the model to stay fast, stay up, and not blow up the bill. (coreweave.com) ### Where do observability tools fit in? They matter more than they sound. CoreWeave’s Weights & Biases integration tied infrastructure signals to model runs and production monitoring. ARENA does something similar from the buyer’s side — run real workloads, see failure points, and measure cost before committing. That is a sign of where the market is heading. Buyers want evidence. They want to know where latency comes from, why jobs fail, and what scaling will cost before they sign. (docs.coreweave.com) ### Is this really Nebius versus CoreWeave? Not exactly. CoreWeave is much larger right now. But Nebius is becoming a useful contrast case. One company is showing early operating leverage and fresh enthusiasm. The other is showing massive demand, but also the capital intensity and financing burden that come with hyperscale expansion. Both are proving the AI infrastructure market is real. They are just proving different things. (prnewswire.com) ### Bottom line? The market is moving past raw capacity. The winners still need GPUs and power, obviously. But now they also need sane economics, flexible pricing, and tooling that makes production AI easier to run. Nebius got rewarded for showing that picture more cleanly this week. CoreWeave is still trying to convince investors it can. (morningstar.com)