Neocloud sector faces mounting risks as AI infrastructure demand surges
Scarcity in AI computing capacity is driving strong demand for neocloud operators that rent out servers and chips despite their relatively asset-light business models. The model is also creating financial and contractual vulnerabilities that could amplify losses if demand forecasts weaken or competition expands.
Highlights
- Nscale signed a six-year, $45bn rental agreement with Anthropic and plans to go public next month after a $14.6bn valuation.
- CoreWeave boasts a $100bn-plus order backlog and its shares have more than doubled since last year, while Nebius is up threefold.
- Vendor financing and long-term rental contracts expose neoclouds to contract risks if AI infrastructure demand projections or pricing prove too optimistic.
Business model strains behind rapid expansion
As reported by Financial Times, neocloud companies such as CoreWeave and Nscale are expanding quickly by leasing out AI computing infrastructure while often depending on rented facilities, supplier credit and guarantees from chipmakers. That structure leaves them exposed to a chain of obligations that can become difficult to manage if market conditions shift.Nscale, a London-headquartered start-up, last week signs a six-year rental agreement with Anthropic worth about $45bn and is due to go public, possibly as early as next month. The company was valued at $14.6bn after its last fundraising, and its board includes former Meta Platforms executives Sheryl Sandberg and Nick Clegg.
Peer performance is supporting investor enthusiasm. CoreWeave, in the U.S., has a $100bn-plus order backlog and its shares have more than doubled since listing last year, while Netherlands-based Nebius is up threefold over the past year.
Contract and financing risks test the sector
Heavy demand for data centres is helping neoclouds generate revenue, but the sector faces the risk that AI infrastructure forecasts prove too optimistic or that elevated prices attract additional competitors. If capacity overshoots demand, the close financial ties between chip suppliers, investors and guarantors could worsen any downturn.The article points to vendor financing as a particular vulnerability, echoing risks seen during the dotcom crash of 2000 when interested suppliers helped fund excessive buildout. Another pressure point is that neoclouds sign long-term rental contracts with major AI companies while having limited visibility on how their own technology and operating costs evolve over time.
That mismatch may become more acute as hardware ages. The May agreement between Anthropic and xAI, cited from SpaceX's listing prospectus, can be terminated by either side with 90 days' notice, despite carrying an annual price tag of almost $38bn per gigawatt.
Even so, the sector has shown an ability to adapt. Earlier versions of Nscale and Ionic Digital used power and computing capacity for crypto mining, suggesting some operators may yet pivot again if AI economics change.
Investor sentiment toward AI data center and infrastructure plays has started to improve after recent political and community pushback slowed the sector’s momentum. Our earlier coverage noted that, despite the shift in tone, uncertainty around rates and fundamentals has kept investors cautious, while attention remained on how data center projects are financed and supported by tenant quality and credit structure.