What is a neocloud, and why is it under scrutiny?
A neocloud is a cloud provider built around GPU capacity for artificial intelligence. These companies buy Nvidia-based servers, house them in data centers, and sell access under long-term contracts. They sit between chip suppliers and the AI developers that need compute quickly.
The best-known example is CoreWeave, which began in 2017 as an Ethereum miner. IREN, Core Scientific and Applied Digital also started in crypto mining, according to BBC reporting. Nebius is the former Yandex NV, renamed in 2024 after separating from its Russian parent.
The model worked while AI demand outran supply. Neoclouds could sign multi-year contracts, borrow against the hardware and the contracts, and build capacity before customers arrived. That combination turned small crypto outfits into companies with billions of dollars of revenue and debt.
It also created a fragile structure. The hardware depreciates quickly, the debt must be serviced, and revenue depends on a small number of large customers. Those features explain why the sector is now a test case for the wider AI buildout.
The debate sharpened on 1 November, when the BBC published a report on the neocloud sector. It described the firms as a once-obscure corner of the AI ecosystem that now sits in the spotlight. The report quoted analysts who question whether the model is financially sound.
The label covers AI-specialist cloud providers that sell GPU capacity rather than operating a broad cloud platform. It matters because these firms are not hyperscalers. They lack the diversified revenue that Microsoft, Amazon and Google can use to absorb losses.
Scale is part of the story. CoreWeave had 1.5 gigawatts of active power and 3.7 gigawatts of contracted power at the end of June. Those figures put it in the same conversation as much larger cloud operators, even though its revenue base is smaller.
How much debt does CoreWeave carry?
CoreWeave reported $2.575 billion of revenue in the second quarter of 2026, up 113% from a year earlier. It also reported a net loss of $626 million and an operating loss of $49 million. Net interest expense reached $640 million, up 140% year on year.
The balance sheet shows the scale of the financing. At 30 June, CoreWeave had $31.4 billion of recourse debt and $3.7 billion of non-recourse debt. That comes to about $35.1 billion, net of discounts and issuance costs, up from $21.4 billion at the end of 2025.
CoreWeave also carried $16.3 billion of operating lease liabilities, according to its quarterly filing. Adding debt and leases gives about $51 billion of obligations. That total helps explain why Reuters described a $50 billion debt load in October.
The cost of that borrowing is visible in the bond coupons. CoreWeave sold 9.75% notes due in 2031 and 9.625% notes due in 2032. It also issued euro-denominated notes at 8.5%. The yield on its October 2031 dollar bond briefly rose above 16% in mid-October, Reuters reported.
Credit markets priced more risk as well. The cost of insuring CoreWeave's debt rose above 1,000 basis points, a record. The figure comes from S&P Global Market Intelligence data cited by Reuters. That is a signal of investor concern, not proof of default.
The asset side of the balance sheet is large. CoreWeave reported a revenue backlog of about $104 billion at 30 June. It said 98% of second-quarter revenue came from committed take-or-pay contracts. Adjusted EBITDA was $1.51 billion.
Concentration is the weak point. Two customers accounted for 36% and 26% of second-quarter revenue, or 62% combined. Meta has initially committed to pay CoreWeave up to about $21 billion through 2032. A backlog is only as strong as the customers behind it.
Much of the debt is secured by assets or equity interests, while the senior notes are unsecured. That structure can protect lenders and still leave the company with limited room to maneuver in a downturn.
Why did neocloud shares fall in October 2026?
Share prices fell sharply across the group. CoreWeave closed at $54.87 on 30 October, down 37% from $87.12 on 30 September, according to Nasdaq data. Nebius fell 35% over the same period. IREN fell 41%, and Applied Digital fell 27%.
The first trigger came on 14 October, when Gartner published its forecast on AI infrastructure spending. CoreWeave fell 15.5% that day, according to Reuters. The forecast gave investors a framework for the revenue gap.
Macro factors added pressure. Firmer inflation data increased bets on interest-rate rises, which matters for companies that borrow heavily. Reuters reported that CoreWeave's shares and bonds were heading for steep weekly losses on 16 October.
Hyperscaler earnings then cooled the mood. Microsoft reported $45.2 billion of capital spending including finance leases, below the $50.67 billion consensus tracked by StreetAccount. Alphabet left its 2026 capital-spending range at $195 billion to $205 billion instead of raising it.
Alphabet's shares fell about 12% on 29 October. Its third-quarter capital spending reached $51.36 billion, and free cash flow was negative $9.07 billion. Investors read the numbers as a sign that even the largest buyers are weighing returns more carefully.
Public comments amplified the shift. Elon Musk wrote on X on 25 October that more than 200 neoclouds face "guppy-level death and consolidation". He added: "Some will make it through, but not many."
Treasury Secretary Scott Bessent pushed back on 29 October. He called neoclouds "a small component of the overall ecosystem" on CNBC. The exchange captured the split between official reassurance and market anxiety.
What does Gartner's 80% forecast actually say?
Gartner's 14 October release set out three claims that frame the debate. It said more than 80% of neocloud providers are expected to cease independent operations by 2030. It said more than $3.5 trillion of AI infrastructure spending could be exposed to impairment by the same date. And it said the revenue needed to justify the buildout is far above likely revenue.
The numbers behind the warning are stark. Gartner estimates that AI data-center owners need $8.5 trillion of cumulative revenue through 2030. The figure covers capital costs, financing and a 12% return on invested capital. Gartner forecasts no more than $3.6 trillion of attributable AI revenue.
Gartner projects that more than $4 trillion in asset value could face write-downs through 2032 if current trends hold. Gartner does not expect a sustainable 12% return before late 2032.
The forecast also sizes the market that neoclouds are chasing. Gartner expects spending on AI-optimized infrastructure-as-a-service to grow from $66 billion in 2027 to $174 billion in 2030. That is rapid growth, but it may still be too small for the number of firms borrowing to compete.
Impairment is an accounting term, not a bankruptcy filing. It means an asset's carrying value falls below the cash it is expected to generate. For a GPU landlord, that can happen if rental rates fall, utilization drops, or newer chips make older ones less valuable.
The $3.5 trillion figure refers to spending that may not earn its cost of capital, not money already lost. It is a warning about future write-downs if utilization and pricing disappoint. The figure is a projection, and projections can be wrong in either direction.
"Cease independent operations" is also broader than collapse. It includes acquisitions and consolidation. Forrester analysts expect at least three neoclouds to be acquired soon as finances get shaky, according to the BBC.
Gartner's forecast is a scenario, not an observed outcome. It assumes current spending and revenue trends continue. A faster rise in AI revenue, slower capacity growth, or stronger pricing would change the arithmetic.
Who is exposed if neoclouds stumble?
The first group is creditors. CoreWeave's bonds and loans are held by institutions that accepted high coupons for high risk. Credit default swaps let others hedge or bet on default. A rise above 1,000 basis points shows that protection has become expensive.
Shareholders are the second group. Nvidia is CoreWeave's second-largest investor, according to Reuters. That link creates a circularity: Nvidia supplies the chips, invests in the buyer, and benefits when the buyer expands.
Customers are exposed too, but unevenly. AI developers need capacity, and neoclouds provide it faster than some hyperscalers can. A forced merger or restructuring could disrupt delivery schedules. Large customers, however, usually have more bargaining power than small suppliers.
Landlords, power providers and construction firms also carry risk. CoreWeave's $16.3 billion of operating lease liabilities show how much of the buildout sits in lease obligations. If a tenant retrenches, the property owner keeps a specialized asset that may be hard to re-let.
Local governments and communities have a stake as well. Authorities often offer tax abatements and power commitments to attract data centers. A delayed or canceled project can leave those plans stranded, even if the shell of the building remains.
Employees face the usual consolidation risk. Back-office, sales and engineering roles overlap when two capacity providers merge. Construction jobs are temporary by nature, and the permanent operational workforce at a data center is small.
Equity holders sit at the bottom of the capital structure, while secured creditors recover more. That hierarchy explains why a stock can fall 37% while the bonds still trade, albeit at distressed yields.
What is the case against the bearish view?
The strongest counterargument is that demand still exceeds supply. CoreWeave's $104 billion backlog and 3.7 gigawatts of contracted power show that customers have signed up for years of capacity. The company had 1.5 gigawatts of active power at the end of June.
Bessent's point is also relevant. Neoclouds are a small part of the AI financing system. Hyperscalers are still spending at extraordinary levels. Alphabet's capital-spending range remains $195 billion to $205 billion for 2026.
Some analysts argue that bond yields are extrapolating a business failure that has not happened. Evercore maintained an outperform rating on CoreWeave in October, according to Reuters. Nvidia's Jensen Huang said the company's financial strength was "improving every day".
Gartner's 80% figure includes acquisitions, and consolidation is not the same as value destruction. A neocloud bought by a larger cloud provider can keep its data centers and customers. Bondholders may be repaid even if equity holders lose independence.
The bear case still has force. A backlog depends on delivery and on customers paying. GPU depreciation is fast, and refinancing at 9% to 16% is expensive. A company can be solvent on paper and still struggle to fund its next quarter of construction.
The sector also has genuine customers with deep pockets. Meta's commitment is measured in billions, and large AI labs continue to buy compute. Those contracts do not disappear because share prices fall, though they can be renegotiated if a supplier is weakened.
The honest conclusion is that the sector faces a financing test, not a verdict. CoreWeave's contracts may prove durable. Its cost of capital may fall if rates ease. Or the gap between infrastructure spending and AI revenue may force the consolidation Gartner predicts.
What to watch next?
CoreWeave reports third-quarter results on 10 November. The key numbers are revenue, backlog conversion, capital spending and interest expense. Analysts will also watch whether customer concentration has improved.
The company's 2027 spending plans matter more than the quarter itself. If CoreWeave slows construction, that would signal weaker demand or tighter financing. If it raises guidance again, the growth story remains intact for now.
Bond and credit markets will give an earlier signal. Watch the yield on the 2031 and 2032 notes, the cost of credit default swaps, and the appetite for new neocloud debt. A failed or repriced bond sale would be more informative than another day of share-price moves.
The IPO pipeline is another test. Anthropic and OpenAI are expected to sell shares in the coming months, according to the BBC. Their listings would give investors a new way to own AI exposure and could pull capital away from second-tier infrastructure firms.
Watch the casualties and the consolidators. Microsoft withdrew from a deal to anchor Nscale's Loughton project. Firmus shelved its Australian IPO in early October. Both show that power delays and market conditions can derail plans quickly.
Finally, watch policy and power. Planning delays, water rules and electricity queues can hurt neoclouds as much as higher interest rates. The companies that locked in power years ago have an advantage over those still waiting.
The near-term calendar is crowded. CoreWeave's results, the AI lab listings and the next round of hyperscaler guidance all land within weeks. Each will update the same question: who pays for the buildout, and at what cost of capital?