Nebius reported that an auction for computing capacity on Tuesday showed continued price increases, with the company's chief revenue officer Marc Boroditsky noting higher prices per megawatt—a key metric for how much neocloud firms can charge clients for computing power.
The bullish signal from Nebius comes as investors increasingly question the sector's debt-fueled model. Neocloud companies are spending billions on Nvidia GPUs, building data centers, and leasing that capacity—a strategy now under pressure following CoreWeave's credit concerns. CoreWeave's five-year credit default swaps reached distressed levels, and its bond yields climbed.
That pressure has spread. Stock prices for Iren and Nebius both dropped sharply over the past month.
Boroditsky differentiated Nebius from CoreWeave, saying his company carries lower debt levels. He projected that Nebius would fully sell out its 2027 computing capacity, with similar demand dynamics expected in 2028.
Chase Lochmiller, CEO of Crusoe Energy, an Nvidia-backed data center builder, reinforced the pricing narrative. "We're seeing higher pricing today for A100s than we saw in 2020 or 2021," Lochmiller said, adding that H100 prices are now higher than when they were first deployed in 2023. Older-generation chips are economically critical for infrastructure companies trying to recoup multi-billion-dollar GPU investments and generate returns.
Lochmiller attributed sustained demand to the capabilities of underlying AI models, which extract more value from compute infrastructure. Higher rental rates on older Nvidia chips than their initial deployment prices could strengthen the economics of companies like Crusoe.
The core tension for neocloud operators centers on GPU depreciation. Newer, more powerful chips will eventually enter the market. If older generations can command attractive rental rates, profitability becomes viable. If not, billions in hardware spending face write-downs. Nebius and Crusoe's pricing reports suggest current demand for AI compute outweighs obsolescence concerns—but that dynamic could reverse quickly if model improvements slow or customer demand weakens.
