For the past two years, AI infrastructure companies have been on a pricing spree, jacking up rates as demand for high-end GPUs keeps outstripping supply. Nebius Group N.V. (NBIS) just discovered that even repeated price hikes weren't enough to cool things down.

So the company tried something different: it held an auction. And here's the kicker—customers voluntarily paid 15% to 20% more than Nebius's previous highest prices just to lock in guaranteed access to Nvidia Corp (NVDA)'s Blackwell-powered AI compute.

Nvidia's Blackwell Compute Demand

Speaking at the Goldman Sachs Communacopia + Technology Conference 2026, Nebius founder and CEO Arkady Volozh explained that traditional price increases just weren't cutting it. "When you have this imbalance between demand and supply, the best price discovery mechanism is something like an auction," he said.

The company first tried manually raising prices, but demand stayed stubbornly high. "We tried to do it manually first to raise the prices. It didn't help. The demand is still there," Volozh admitted.

That's when Nebius decided to test an auction for Blackwell compute capacity. The outcome even surprised the company. "We achieved results when we got 15%, 20% higher price than we ever sold before, after all our price raises," Volozh said.

AI Pricing Power

This experiment offers a rare peek into just how tight the AI infrastructure market really is. Instead of charging a flat rate for Nvidia Blackwell compute, Nebius let customers bid for guaranteed capacity. And according to Volozh, the winners were actually happy about it because the auction removed the uncertainty of whether they'd get the computing resources they needed.

"The guys who won this auction… were happy that they were able to get guaranteed capacity through this mechanism," he said.

The structure of these deals suggests Nebius isn't merely selling GPU time anymore—it's selling certainty. For companies building or deploying AI models, having guaranteed access to cutting-edge compute can justify a premium if delays could derail product launches or customer deployments.

AI Infrastructure

Nebius's comments also hint that the AI infrastructure market might be entering a new phase. Until now, investors have focused mostly on the surging demand for Nvidia's latest GPUs. Volozh's remarks indicate the conversation is shifting from whether customers want Blackwell capacity to how much extra they're willing to pay to secure it.

That distinction matters because it points to pricing power, not just utilization. If AI infrastructure providers can consistently command premiums through market-based pricing mechanisms while demand continues to exceed supply, profitability could improve even before additional capacity comes online.

Why It Matters

The auction wasn't just an experiment in selling AI compute—it was an experiment in discovering its true market value. If other AI cloud providers start adopting similar approaches for scarce GPU capacity, investors may begin evaluating AI infrastructure companies not only on how much compute they own, but also on how effectively they monetize periods of persistent supply shortages.