Neocloud Stocks Are Slumping Because the AI Buildout Has to Pay for Itself Now

Neocloud Stocks Slump: The Numbers

Doubts over whether AI can pay for its own infrastructure are weighing on the leveraged newcomers of the cloud world, according to a UOB Kay Hian report covered November 2 by Singapore Business Review. Shares of CoreWeave have fallen 46 percent and Nebius 43 percent in the second half of 2026. Australia's Firmus withdrew its planned IPO in October, Blackstone-backed AirTrunk postponed its proposed $1.5 billion Singapore REIT listing, now reportedly targeting completion by early 2027, and South Korean digital-infrastructure company Elice fell 32 percent on its first trading day on October 20. The message from the market is blunt: the AI buildout's question has changed, from how fast to who can afford to build.

Neoclouds are newer, smaller cloud providers built specifically to rent out AI computing: rows of GPUs that customers use to train and run models. Unlike hyperscalers such as Microsoft, Amazon, and Google, which also sell software, advertising, and shopping, neoclouds live or die on AI compute demand. That focus was a strength when demand exploded. It is a vulnerability now that financing is tighter, power is scarcer, and investors are asking harder questions about returns.

Why the Model Is Under Strain

The report's argument is straightforward: neoclouds spend first and earn later, borrowing heavily to buy chips and build data centers before customers sign long contracts. When capital was cheap and AI demand seemed endless, that model worked. Now power shortages, tighter financing, and environmental opposition are slowing new development, and the cost of carrying that debt is rising. A data center that takes years to build and fill is a very different bet from a software product that scales overnight.

Paradoxically, the same constraints raise the value of data centers that already exist and of vacant capacity waiting to be filled. If new supply is hard to build, existing capacity becomes more valuable, which is why the selloff is not a verdict on AI demand itself. It is a verdict on the financing model of the newest, most leveraged builders. Hyperscalers with diversified revenue can fund data centers from cash flow. Neoclouds cannot, at least not at the same scale.

What This Means for Cloud Buyers

For companies buying AI compute, the neocloud slump is a reason to check counterparty risk, not to panic. If your provider is heavily leveraged, ask about its funding runway, its contracted backlog, and its power supply. Diversify across providers where you can, and avoid locking your entire AI workload into a single neocloud without an exit plan. The NTT DATA Jakarta data center shows the other side of the market, where established operators keep expanding with contracted demand, while the Equinix record quarter shows demand for existing data center space still outrunning supply.

For investors, the AirTrunk timetable is the first checkpoint. If its REIT listing completes by early 2027, it will signal that institutional appetite for data center assets is returning. If it slips again, expect more caution. The broader context is Big Tech's own spending: Big Tech is on track to spend more than $720 billion on data centers in 2026, and that spending is increasingly funded by debt and equity raises, not just cash flow. The whole AI infrastructure trade is being repriced, and neoclouds are simply where the repricing shows up first.

The Bottom Line for the Buildout

None of this means the AI buildout is over. Demand for compute is still real, and the largest players are still building. What is changing is who builds, how they pay for it, and what return investors demand. The neocloud slump is a correction in the financing layer of AI infrastructure, not a collapse in the technology itself. The companies that survive it will be the ones with contracted customers, secured power, and balance sheets that can wait out a tighter market. The ones that do not will become cautionary tales in the next cycle's pitch decks.