Gartner: 80% of Neoclouds Won't Survive

By Daniel Rao | October 14, 2026

Somebody finally did the division. On Wednesday morning, Gartner published a forecast that reads less like research and more like a foreclosure notice: more than $3.5 trillion of planned AI infrastructure spending could be exposed to impairment by the end of 2030, and over 80 percent of neocloud providers — the GPU-rental specialists that have been the purest bet on the AI boom — will cease independent operations by the end of the decade. Wall Street did not wait for the footnotes. CoreWeave cratered roughly 16 percent, DigitalOcean fell about 13 percent, Hut 8 dropped more than 10 percent, and the rest of the neocloud complex followed them down.

The scariest part of the report is not the headline number. It is the arithmetic underneath it — the kind of simple, pitiless subtraction that ends boardroom arguments.

The arithmetic nobody wanted to see

The numbers come from a Gartner report titled "Prepare for AI Market Consolidation: Capital Write-Downs Could Exceed $4 Trillion by 2032," released Wednesday from Stamford, Connecticut. "In 2026, AI providers are expected to spend nearly $1 trillion on AI infrastructure, and are forecast to invest an additional $6 trillion between 2027 and 2030," said Will Sommer, senior director analyst at Gartner. "However, our research indicates that enterprise and consumer AI revenue will be insufficient to support profitable returns on all planned AI capital expenditures during this period."

Then comes the subtraction. Assuming a 12 percent return on invested capital — the hurdle rate many companies treat as the bar for acceptable investment — Gartner estimates AI data-centre owners would need to generate a cumulative $8.5 trillion in revenue through 2030 to cover announced capital spending and financing costs. It forecasts that total enterprise and consumer AI revenue attributable to AI data centres will reach no more than $3.6 trillion over the same period. That is a shortfall of nearly $5 trillion, and Gartner says it does not expect the 12 percent return to be achieved until the end of 2032 at the earliest.

The mechanics of the pain matter as much as the total. Failing to earn a sustainable return on rapidly depreciating assets, Gartner argues, would almost certainly force operators to write down assets they can no longer recover — particularly as newer GPUs displace older hardware, stranding the chips the neoclouds are buying today. Push the trajectory out and the exposure grows: more than $4 trillion in asset value at risk of write-down through 2032. There is a further twist that should worry infrastructure owners specifically: because most AI-native providers run asset-light business models, Gartner predicts less than half of AI revenue will be earned by the providers that actually own significant AI infrastructure. The people collecting the revenue and the people holding the assets are, increasingly, not the same people.

Why neoclouds take the hit first

A neocloud is the tier of GPU-as-a-service providers sitting below the hyperscalers — companies like CoreWeave, Nebius, IREN and their peers that rent AI compute by the hour. Gartner expects this tier to absorb the blow first, predicting that "cost pressure and a lack of a differentiating business model will force over 80% of neocloud providers to cease independent operations" by 2030. The hyperscalers get a cushion, Gartner says, from massive and highly diversified revenue sources that could plausibly sustain AI investment for years. Neoclouds have no such cushion: they must soon earn returns on AI-optimized infrastructure-as-a-service and associated platforms, or they do not survive.

The segment's own growth numbers explain why. Gartner estimates spending on AI-optimized IaaS will grow from $66 billion in 2027 to $174 billion in 2030 — but Sommer said that growth will be too slow for most AI compute providers to recover their investments before the current generation of AI semiconductors is replaced. Read that sentence twice. The chips depreciate faster than the revenue arrives. That is the entire neocloud business model described as a race the runner is already losing.

The line that travelled furthest on Wednesday was not from the press release at all. It was the aside Sommer gave The Register: "Not everyone is in trouble," he said. "Some will outperform, some will underperform, and some will have to exit the market." Nobody on the neocloud side of the trade heard the first sentence.

Wednesday's tape: what fell — and what else was moving

The damage was broad. CoreWeave led the collapse, down about 16 percent on the session — TradingView's midday read had it at 15 percent, and the tape kept sliding from there. DigitalOcean fell roughly 13 percent. Hut 8 dropped more than 10 percent. IREN sank around 9 to 10 percent, and Applied Digital careened down close to 9 percent. Nebius was the day's moving target: one midday read had it down 9 percent, while Reuters later recorded a 15.1 percent slide — the difference between a morning snapshot and a market that kept selling into the afternoon. We pulled the session tape ourselves: CoreWeave opened near $72, slid to lows in the mid-$60s and closed the day down about 16 percent.

Honest accounting requires saying what else was in the air. Wednesday's selloff was not a one-cause event. Hot consumer-price data revived bets on a second Federal Reserve rate hike as soon as this month, and higher rates are poison for companies borrowing to build data centres and buy AI chips — the 10-year Treasury yield climbed 3.6 basis points and the two-year jumped 7.2. Barclays downgraded AI server makers Dell and HP Enterprise to equal-weight, sending Dell down more than 10 percent. Debt-laden hyperscaler Oracle fell 7.2 percent. And the cost of insuring the debt of some AI companies against default — including Oracle and CoreWeave — surged to record highs, according to S&P Global Market Intelligence data cited by Reuters. "Investors are questioning whether the AI buildout will generate the returns that were expected," Kim Forrest, chief investment officer at Bokeh Capital Partners, told Reuters. The options desk told its own story, too: put buying on CoreWeave surged as traders paid up for downside protection — buying insurance against their own optimism.

The 9%-versus-4% spread at the centre of the panic

If Gartner supplied the macro arithmetic, the micro version arrived from D.A. Davidson analyst Gil Luria, who told the Prof G Markets podcast that CoreWeave's borrowing costs run around 9 percent while its estimated return on investment is only about 1 percent — and that even hitting the company's own target of 4 to 5 percent would still fall short of its financing costs. His comparison travelled fast: it is like, he said, borrowing money at 9% to buy U.S. Treasuries yielding 5%. Luria warned that once the AI investment boom cools, those borrowing costs could prevent the company from meeting its debt obligations — with shareholders facing, in his telling, the possibility of a total loss.

CoreWeave's balance sheet is the exhibit. As of the end of June, the company's debt and lease obligations stood at approximately $52 billion, while second-quarter interest expenses nearly doubled year over year to $640 million — about 5.8 times its annualized EBITDA, a leverage ratio S&P has assessed as carrying "aggressive" financial risk. The second quarter also showed the other side of the ledger: revenue of $2.58 billion, up 112 percent from a year earlier, a contracted backlog of roughly $104 billion, and adjusted EBITDA of $1.51 billion at a 59 percent margin. The demand is real. The question Gartner is now forcing is whether demand at these prices can ever service debt at these rates.

And the trouble did not start with the stock. It started with the debt: yields on CoreWeave's $1.25 billion in unsecured bonds due in 2032 climbed to an intraday high of 15 percent before closing at approximately 14.5 percent, according to Crypto Briefing. When a company's bonds are pricing in distress while its backlog is pricing in triumph, one of those two markets is wrong.

This didn't come out of nowhere

Regular readers will recognise this story. TechPulse USA has been documenting the "who pays for the AI buildout" question for months — deal by deal, debt raise by debt raise. In October we covered AI's $113-billion debt machine hitting the wall, the financing structures straining under the weight of the buildout. We reported on Anthropic's secret bitcoin-miner deal, the scramble for power capacity that has AI labs cutting deals with the one industry that already owns energised land. And we tracked Bain-backed Bridge seeking $6.6 billion in debt for its Thailand buildout — the latest mega-raise in a stack of them. Gartner's forecast is the macro version of the same warning: the financing was always the fragile part, and the analysts have now put the industry's own numbers to it.

My take: the buildout is real; the funding model is shoddy

What follows is analysis, not reporting.

Here is the judgment call: Gartner's numbers are forecasts, not fate — research firms have buried the AI trade before, and been buried by it in return. But the honest version of Wednesday is that the neocloud business model was always a carry trade wearing a hard hat. Borrow at 9 percent, buy the scarcest chips on earth, rent them by the hour, and pray the spread holds until the GPUs are paid off. That was never a moat. It was leverage with better branding.

The botched assumption underneath the whole stack was that demand would grow fast enough to outrun both depreciation and interest — that the music would still be playing when the current generation of silicon became the previous one. Gartner is now saying, with the industry's own capex plans as evidence, that it won't. The hyperscalers survive this math because they do not need AI to pay for itself this decade; their cloud, software and advertising businesses carry the weight. The neoclouds need AI to pay for itself now, at 9 percent money, against a depreciation clock that never stops. That is why 80 percent of them, in Gartner's telling, end up absorbed or gone — and why the honest read is that most were always going to end up as divisions of bigger companies. The people who should be angriest about Wednesday are not the operators. They are the late-cycle equity buyers who paid up for "picks and shovels" and discovered they had bought a leveraged spread instead.

What to watch next

Gartner will present additional detail on AI infrastructure trends at its IT Infrastructure, Operations & Cloud Strategies conferences — London on November 16–17, Tokyo on December 1–3 and Las Vegas on December 8–10 — and the Q&A sessions there will be worth watching for how the forecast survives contact with the industry's own CIOs. Closer in, the neoclouds' next earnings prints are the real test: can any of them show the spread between borrowing costs and returns on capital actually improving, or does Luria's 9-versus-4 math hold? Watch hyperscaler capex discipline too — the moment Microsoft, Amazon, Google or Meta blink on buildout spending, the GPU order book that underwrites every neocloud backlog starts to wobble. And watch the M&A wire: if Gartner is right that four-fifths of neoclouds cease independent operations by 2030, the consolidation has to start somewhere. Wednesday's tape suggests the market has already begun pricing it.

Sources

Gartner press release, "Gartner Says Over $3.5 Trillion of AI Infrastructure Spending Could Be Exposed to Impairment by the End of 2030" (October 14, 2026, primary source); The Register; Investor's Business Daily; Reuters; TradingView / Seeking Alpha; The Hindu BusinessLine; AI Weekly; Crypto Briefing; TradingKey; ainvest. Session price data cross-checked between Finnhub market data and at least one published outlet for each ticker cited. Images: Wikimedia Commons / Library of Congress (see credits).

Daniel Rao is TechPulse USA's infrastructure correspondent, covering the data-centre economy, AI financing and the chip supply chain.

Search This Blog

TechPulse USA