Inside European GPU-secured debt: terms, tenors and appetite
Analysis by Clara Oliver-Amorim
GPU-secured financing is emerging as the newest vertical in the European digital infrastructure financing market. The sector has been spearheaded by raises from Mistral AI, Nebius and Nscale, all of which have brought asset-level debt against hardware ready to be deployed into European datacentres, TMT Finance understands.
The European deals so far share the same template: US dollar-denominated, to match the currency of the compute contracts, fully amortising over four to five years and underwritten primarily against the creditworthiness of the hyperscaler or investment-grade customer buying the compute.
Where things differ is in how much risk sits with the neocloud itself, after which pricing, structure and lending appetites start to diverge. With contracts and hardware now being financed on an average five-year view, the main question the market is asking is about the borrowers themselves or, as one source questioned - which neoclouds will still be around in five years' time?
Does amortisation resolve the shelf-life dilemma?
The GPU-secured structure rests on a debt profile that fully amortises before the asset’s end of useful life, reflecting the relentless pace of innovation and heightened obsolescence risk inherent in GPU assets. One source told TMT Finance that while GPUs from five years' ago still hold value, it is not safe to assume that going forward the latest GPUs will also last five years. Some market participants have asked, rhetorically, how lenders can know these chips will last five years rather than four or three, given a five-year depreciation window assumes the hardware degrades in a straight line.
However, mid-life risk stands out if contracts are terminated or renegotiated in year two or three, leaving outstanding debt against hardware that has been eclipsed by one or two generations already. In the US, CoreWeave [NASDAQ: CRWV] closed a US$2.6bn delayed draw term loan, known as DDTL 5.5, on 10 August 2026, according to a press release, which unlike earlier facilities backed by customer contracts carries an around five-year maturity against underlying customer contracts averaging around three years, the release said.
CoreWeave stated the support signalled lender confidence in the long-term value of Nvidia [NASDAQ: NVDA] GPUs running on its cloud platform and a willingness to underwrite renewal risk. The Ba2/BB+ (Moody’s/Fitch) deal priced at 550 basis points (bps) over SOFR, the release said.
GPU-Secured Financings in Europe
Borrower | Country | Date | Size | Pricing | Tenor / maturity | Rating | Security / counterparty | Lenders | Source |
|---|---|---|---|---|---|---|---|---|---|
Nscale (GPU DDTL) | Norway, Portugal, Iceland, UK | Feb, 2026 | US$1.4bn | SOFR + 500 bps (possible 25 bps step-up) | Not disclosed | Unrated | Pool of GPU clusters across multiple customers, incl. contracts not yet generating revenue; Nscale group guarantee | PIMCO, Blue Owl, LuminArx Capital Management (Goldman Sachs structuring and placement agent) | Nscale press release; S-1 filing |
Mistral AI | France | Mar, 2026 | US$830m | Not disclosed | Not disclosed | Unrated | Nvidia Grace Blackwell infrastructure incl. 13,800 GB300 GPUs | Bpifrance, BNP Paribas, Crédit Agricole CIB, HSBC, La Banque Postale, MUFG, Natixis | Mistral AI statement |
NexGen Cloud | Sweden | Jun, 2026 | US$34m | Not disclosed | 3 years, non-recourse | Unrated | B200 GPU fleet and contracted cash flows | USD.AI | NexGen Cloud press release |
Nebius | Finland | Jul, 2026 | US$775m | SOFR + 250 bps | Matures Oct 2030 | Unrated | Deployed GPUs and contract with an investment-grade customer | MUFG (structuring agent, sole bookrunner); ABN AMRO, Bank of America, Deutsche Bank, HSBC | Nebius announcement (6-K) |
Verda | Finland | Jul, 2026 | E22m | Not disclosed | 4 years | Unrated | GPU purchases | Nordic Investment Bank (InvestEU-backed) | Nordic Investment Bank |
Pricing like a datacentre
Debt repayment rests on the compute contract and, so far in Europe, these are largely backed by the same investment-grade tenants that lease datacentre capacity, keeping spreads between the two verticals tight.
Just like with datacentres, everything rests on the offtake. Nscale's European GPU DDTL, a US$1.4bn facility signed in February 2026, is secured against a pool of GPU clusters spread across several European sites and customers, including some contracts not yet generating revenue, with the Nscale group guaranteeing the debt rather than a single hyperscaler contract underpinning it, according to the S-1 filing. As a result, the deal priced at 500 bps over SOFR with a potential 25 bps step-up, the filing said. The raise was led by funds managed by PIMCO, Blue Owl and LuminArx Capital Management, according to Nscale's press release.
Compare this to six months later in the US, when Nscale closed two US delayed-draw facilities totalling about US$3bn in August: up to US$1.85bn for its Ward County, Texas campus and up to US$1.2bn for Madison, North Carolina, the press release states. The floating-rate borrowings under both facilities are priced at 237.5 bps over three-month SOFR, according to Nscale's S-1 filing.
Both carried investment-grade ratings with stable outlooks. Moody's said the loan is tied to a six-year take-or-pay GPUaaS contract with a hyperscaler rated at least Aa3, fully amortises within the term of that contract, and is supported by large customer prepayments on each GPU tranche. It further added that strict funding conditions help limit lenders' exposure to delays in delivering the GPUs and the datacentre. The point stands that a hyperscaler take-or-pay contract, full amortisation inside the contract term and upfront customer cash can cut the cost of GPU debt by more than half.
A separate source said Europe's deals have so far lacked hyperscaler wraps, a key form of credit support seen in the US, such as Google's backstop of Fluidstack's lease obligations. Instead, lenders rely on direct hyperscaler offtake, upfront payments and security deposits, as TMT Finance outlined in an analysis on credit wrappers in May. But without a wrap, the same source said lenders take project-on-project risk across two SPVs rather than one, with exposure resting on the neocloud, two steps removed from the hyperscaler's cashflow.
None of the European deals carry a public rating, with one source saying strong ratings are what open the door to institutional and term loan B investors. As TMT Finance noted in May, the market could divide between contractually supported deals with tighter pricing and deeper capital access, and unsupported deals facing higher funding costs.
No ceiling on size
With a strong compute contract, there is effectively no upper limit on how large a GPU financing can be, one source told TMT Finance. Hardware makes up roughly 80% of the cost of building an AI datacentre, sources said, meaning GPU debt could start to account for the largest slice of the capital stack.
At the top end globally, so far, CoreWeave priced its US$8.5bn DDTL 4.0 facility, its first investment-grade piece, at 225 bps over SOFR, according to a March 2026 press release. The non-recourse facility was rated A3 by Moody's and A by DBRS, secured by substantially all assets of CoreWeave Compute Acquisition Co. VIII, the release said, with MUFG and Morgan Stanley as co-structuring agents and joint bookrunners.
However, one source noted lenders are increasingly cautious on building up too much exposure to Nvidia, even while the company stands as a solid counterparty. Across the board, Nvidia finds itself simultaneously playing the part of chip supplier, equity holder and, increasingly, backstop provider. A downturn in AI demand would hit the collateral and the credit support at the same time; though proofing against a downturn in AI demand seems somewhat antithetical to market sentiment at large.
Who will still be here in five years?
With four-to-five-year debt now the norm, the key question for lenders is which neoclouds will outlast their own loans, one source said. Contract quality, shareholder depth and access to cheap power are emerging as key dividing lines, TMT Finance understands, but all in all, the answer is unlikely to rest on a single metric.
If we assess staying power from a financing point of view, operators that match their debt to their contract life are able to secure cheaper debt and avoid mid-term recontracting risk. Like with datacentres, offtake agreements ring-fenced in single-purpose vehicles mean debt can be repaid from contracted cash even if the wider business – or wider sector – comes under pressure. Following on with that logic, businesses that have already converted expansion efforts into revenue-generating capacity are better off than those that still need to build their books.
Overall, the practical test for lenders may be less about which operator has the largest backlog and more about which has the contracts, funding and power to keep servicing its debt if new business slows. So far in Europe, 15 lenders have joined the GPU-secured financing lending brigade, according to press releases: Bpifrance, BNP Paribas, Crédit Agricole, HSBC, La Banque Postale, MUFG, and Natixis (Mistral AI); ABN Amro, Bank of America, and Deutsche Bank (Nebius); and PIMCO, Blue Owl, LuminArx Capital Management (Nscale DDTL), plus the Nordic Investment Bank’s support for Finnish startup Verda and USD.AI’s on-chain loan to NexGen Cloud. Doubtless, this list will likely be expanded soon.
TMT Finance subscribers can read our full coverage of this area, via our intelligence platform.
Not yet a subscriber? Get access to the very latest intelligence and analysis on M&A, financing and investment in digital infrastructure, keeping you fully informed so you can spot new opportunities and upcoming deals, before your competitors do.
