More than three years into the artificial intelligence (AI) race, demand for computing power continues to outstrip companies' ability to supply it. Neocloud companies like Nebius Group (NBIS +1.59%) and CoreWeave aim to capitalize on that supply/demand imbalance by building new AI-focused data centers to help meet those capacity needs. In fact, some of their largest customers are hyperscalers such as Microsoft and Meta Platforms.
Nebius has pushed to differentiate itself from smaller neocloud businesses with its own server design optimized for cooling efficiency and power usage, its own software, and complete service management. That's pushing it closer to the level of Microsoft's Azure platform than the low-level infrastructure-as-a-service offering of the average neocloud company.
Recently, Nebius started offering its data center stack of design and software management as a stand-alone product. It's asking businesses to front the cash to build a data center and pay for the infrastructure, and the company will take care of the rest. But the new model exposes a huge problem facing Nebius and other neocloud companies.
Image source: Getty Images.
Why is Nebius introducing a new strategy?
Nebius is positioning this new asset-light approach as a way for it to expand its capacity with minimal capital requirements. The company is capital-constrained: Its balance sheet shows about $8.5 billion in debt, up from about $4.1 billion at the end of 2025. It raised an additional $775 million earlier this month and expects to raise even more capital later this year.
That's all part of the neocloud business model. Nebius contracts with large customers and uses those contracts as collateral to secure financing to build the data centers to fulfill them. With a strong demand pipeline, the business model should theoretically pay off in the long run as it scales up.
NASDAQ: NBIS
Key Data Points
So, investors need to ask the question: If Nebius can secure financing using contracted revenue to build data centers itself, why would it want to let other companies build the data centers instead? The simplest explanation is that the cost of capital currently exceeds its return on invested capital.
Nebius isn't unique in that regard, and it's to be expected of a company that's heavily investing in capacity. Once it reaches a scale at which existing capacity generates meaningful recurring revenue, it may be able to deliver higher returns on invested capital. But there's a big challenge for neocloud companies.
The big neocloud problem revealed by Nebius
To produce strong returns on invested capital, you need a differentiated product. So far, the neocloud is proving relatively undifferentiated, with growth driven by demand. However, more supply is coming to market, which could weigh on potential returns for graphics processing units (GPUs), especially if it reduces utilization rates.
Space Exploration Technologies (SpaceX for short) has started offering its excess GPU capacity to other companies. Meta Platforms is considering entering this market as well. While their entries indicate huge demand for computing capacity, they're also well-funded competitors that could add significant capacity to the market. Meta could add capacity with a lower cost of capital, enabling it to compete on price, and SpaceX could, over the long run, lower the structural costs of data centers with its plans for orbital data centers, putting further pricing pressure on the market.
Although Nebius and other neocloud companies have contracts in place for the medium term, enabling them to secure financing and build new data center locations with confidence that they can recoup their investments, the long-term potential of those capital investments is in doubt. Nebius' attempt to mitigate that risk while supporting its near-term results with its new business model has highlighted that challenge for everyone else in the industry.
Only a few neocloud companies will be able to maintain the high utilization rates needed to produce returns that exceed their cost of capital. Those with more scale, like Nebius and CoreWeave, are in a better position, but there's still no guarantee.
Huge competitors like SpaceX or Meta could completely upend the market and put pressure on the smaller companies. Furthermore, it's hard to discount the possibility that AI spending could slow significantly in the near future, given that the entire business model relies on continuous growth.
