CoreWeave's debt hits $35bn: What it means for the neocloud refinancing wall

CoreWeave, the largest publicly traded neocloud, reported total debt of $35bn as of 30 June 2026, up from $22.7bn in long-term debt at the end of the first quarter. Net interest expense reached $640m for the quarter, more than double the $267m recorded a year earlier, as the company financed a $9.4bn quarterly capital expenditure programme. Revenue backlog grew to $104bn, up from $99.4bn at the end of Q1, and the company forecasts $35bn to $39bn in annual capital expenditure for 2026.

The figures update a debt picture that has been building since 2023, when CoreWeave and its peers began financing GPU fleets through asset-backed loans to meet AI infrastructure demand hyperscalers could not fulfil fast enough. Analysts have described loan-to-value ratios in the region of 60-70% on these facilities, though methodology varies by source. That debt is now maturing inside a tightening window between 2026 and 2028, a pattern some industry commentators have termed the neocloud refinancing wall.

CoreWeave is not the only borrower carrying it. Lambda closed a $1bn syndicated senior secured credit facility in May 2026, upsized from $275m the previous August, and is now targeting a public listing in the second half of the year after its initial H1 timeline slipped.

FluidStack has reportedly secured borrowing capacity of up to $10bn from Macquarie and other lenders, according to sector analysis, though the company has not disclosed this publicly. Neither Lambda nor FluidStack discloses consolidated debt on the same schedule as a listed company, so sector-wide exposure is harder to verify than CoreWeave’s own figures, though estimates circulating earlier in the year of around $20bn now look conservative against CoreWeave’s number alone.

The backstop question

Nvidia has moved to address the exposure directly. On 1 July 2026, the company introduced a backstop financing model under which it agrees to rent back unused GPU capacity from neocloud partners at a fixed rate, underwriting a utilisation floor that makes lenders more willing to extend credit. Sharon AI’s six-year, 40,000-GPU deployment agreement was the first deal structured this way.

That mechanism sits underneath the $500bn financing platform Nvidia confirmed with six institutional partners this week, in which Nvidia may itself backstop up to $125bn of the total facility. Where that platform addresses capital access at the hyperscaler and AI lab level, the backstop financing model is aimed squarely at the smaller merchant GPU cloud operators carrying the refinancing wall, the same companies whose Q2 numbers are now landing.

Whether the backstop resolves the refinancing risk or simply relocates it onto Nvidia’s own balance sheet is unresolved. A neocloud financed under the backstop model carries Nvidia’s revenue-share obligations into its pricing, meaning vendor-financed and independent GPU capacity are no longer directly comparable products, a distinction increasingly relevant to enterprise buyers price-shopping between providers.

What connectivity providers should watch

The refinancing wall matters beyond the balance sheets of the neoclouds themselves. As covered in Capacity’s look at Middle East neocloud expansion, regional operators seeking comparable financing face a narrower set of lenders and less established backstop infrastructure than their US counterparts, a gap the $500bn platform’s US-institution concentration does not close.

Collateral quality remains the underlying variable. Nvidia’s own exposure now extends well beyond chips, into power through its $3bn Lancium stake and now into credit risk through the backstop and the Wall Street platform. If GPU residual values hold across hardware generations, the debt amortises as planned. If they do not, the same company sits at the centre of both the collateral and the guarantee.

The question is expected to feature at Metro Connect Fall x Datacloud USA in Austin (2-3 September), where the Neocloud Leaders Forum & Breakfast, a closed-door, invitation-only session held under Chatham House Rules on 3 September, will bring together neocloud, hyperscaler, investor and fibre operator executives to discuss capital, strategy and long-term AI infrastructure expansion.

Why Nvidia is said to be becoming a power company: The $3bn Lancium stake

Can telcos compete in the Middle East’s neocloud and AI infrastructure market?

Nvidia’s $500bn Wall Street pact turns AI chips into assets