Key Takeaways

  • Short sellers held about $6 billion in unrealized October gains across U.S. neocloud and AI data-center stocks.
  • CoreWeave and Nebius Group fell sharply despite forecasts for extraordinary growth in AI infrastructure spending.
  • High borrowing costs, power constraints and a reported $5 trillion funding gap are testing capital-intensive expansion plans.

Short sellers are benefiting from a sharp reversal in some of the market’s most prominent AI infrastructure trades. Bets against U.S. neocloud and data-center stocks generated about $6 billion in unrealized gains during October 2026, according to reporting from 24/7 Wall St., as investors reassessed the cost of financing increasingly ambitious expansion plans.

CoreWeave and Nebius Group have been at the center of the pullback. Their shares declined approximately 36% and 33%, respectively, from late-September highs. IREN and Applied Digital are among the other closely watched operators facing elevated bearish positioning. Short interest in five heavily shorted companies reached roughly 20% to 26% of free float, raising the potential for continued volatility in either direction.

Demand for AI computing has not disappeared. Gartner forecast worldwide AI spending of $2.7 trillion in 2026, an increase of 49.5% from 2025. AI infrastructure accounts for an estimated $1.48 trillion of that total. Gartner also expects global spending on data-center systems to exceed $650 billion this year, up 31.7%, while server spending is projected to rise 36.9%.

Those figures make a simple collapse in infrastructure demand look unlikely. They do not, however, resolve the financial questions confronting CoreWeave, Nebius Group, IREN and Applied Digital. Building facilities, securing high-end accelerators, arranging grid connections and installing cooling systems require large upfront commitments. Revenue generally arrives later, sometimes under contracts whose economics depend heavily on utilization rates and financing terms.

That timing gap matters. The sector is confronting a reported $5 trillion funding gap, while elevated bond yields make debt-funded construction more expensive. Gartner estimates that AI providers could spend nearly $1 trillion on infrastructure in 2026 and another $6 trillion through 2030. Yet cumulative AI-related revenue may not produce enough cash to support all planned capital spending while also delivering attractive returns.

Operators can have active enterprise customers and still face difficult economics. Determining how much revenue remains after electricity, equipment depreciation, interest expense and facility costs is becoming a more urgent priority than calculating the headline value of a multiyear capacity agreement.

Power is another constraint. Data-center electricity demand grew 17% in 2025, while consumption at AI-focused facilities increased 50%. The International Energy Agency expects total data-center electricity use to nearly double from 485 TWh in 2025 to 950 TWh by 2030. Even when financing is available, operators frequently encounter long interconnection queues, limited generation capacity and the need for costly grid upgrades.

Operational efficiency heavily influences credit quality as well as environmental performance. The Green Grid data-center efficiency framework gives operators a way to assess power and cooling overhead, while ASHRAE thermal guidelines help facility teams manage high-density computing environments. Better cooling design can reduce operating costs, but liquid cooling, electrical upgrades and heat-rejection equipment add another layer of capital spending at the outset.

Still, sentiment can turn quickly. CoreWeave and Nebius Group gained ground on October 21 after Elon Musk rejected concerns about AI overcapacity as "ridiculous." That rebound proved fragile, with Nebius Group falling again on October 22 despite fresh Citi financing and CoreWeave also losing ground. Heavy short interest can amplify rallies when bearish investors cover positions, just as leverage concerns can accelerate declines.

A wider debate is now taking shape around whether AI can generate enough revenue to justify the infrastructure being ordered. Fortune has highlighted the multitrillion-dollar difference between projected capital requirements and the revenue available to support them. The precise outcome will depend on utilization, pricing, financing structures and the pace of enterprise adoption.

For business technology leaders, the selloff does not necessarily signal that computing capacity will become abundant or inexpensive. It suggests that supplier resilience deserves closer scrutiny. Contract duration, power availability, customer concentration and access to capital dictate whether promised capacity arrives on schedule. AI demand remains extraordinary, leaving investors to evaluate which operators can actually finance these builds, run them efficiently, and earn an acceptable return.