Nebius has arranged its first senior secured facility against an operating GPU deployment and one customer’s cash flows. The deal adds project-level debt to its expansion toolkit, but an unnamed customer and a larger rival financing limit what it proves about the rest of Nebius’s backlog.
Nebius has shown that banks will lend against one operating GPU deployment. The harder question is whether that transaction can be repeated across its other customer commitments without concentrating more of its economics around a few hyperscalers.
Nebius entered into the approximately $775 million facility to accelerate the global buildout of its full-stack AI cloud platform. The company said the collateral consists of deployed GPU infrastructure and contracted cash flows from an investment-grade customer. It did not name that customer.
The distinction between a live deployment and the broader backlog is central. Nebius said the relevant contract is already in its servicing phase and that the facility, together with cash flows under that agreement, covers more than 100% of the capital expenditure required for the underlying GPU infrastructure. It plans to use the financing for capacity serving AI-native and enterprise customers.
That supports a narrower conclusion than the idea of a self-funding expansion loop. An operating asset and its receipts have become bankable, and the loan gives Nebius fresh capital for buildout. The disclosures do not show that every contracted deployment will qualify or that lenders will advance the full cost of the next project.
The two borrowers—Nebius Compute II, LLC and Nebius Compute II Oy—signed the facility on July 10, according to the filing-based account. Interest is one-month Term SOFR, subject to a 0% floor, plus 2.50% a year. The floating base rate means the spread is not Nebius’s total borrowing cost.
The borrowers may prepay after 10 business days’ notice without a premium or penalty, except for customary breakage costs. They must maintain a 1.15-to-1 debt-service coverage ratio and satisfy a minimum-liquidity requirement.
Subject to specified exceptions, the security covers substantially all assets of the two borrowing subsidiaries and the shares Nebius Group holds indirectly in them. That places the financed project assets and receipts at the center of lenders’ recovery rather than giving them a general claim guaranteed by the parent.
Nebius Group did not provide a general guarantee of all obligations. Its exposure is not zero, however. The same account says the parent supplied a non-recourse guarantee for specified borrower “bad acts” and performance guarantees covering certain management and data-center colocation obligations.
The structure therefore separates the facility from ordinary unsecured corporate borrowing, but it does not eliminate operational obligations. Nebius still must run the deployment, preserve the contracted cash flows and comply with the coverage and liquidity covenants.
MUFG led the deal as structuring agent, sole bookrunner and underwriter. Nebius described the transaction as significantly oversubscribed and listed nine other participating banks. That is evidence of demand for this facility on these terms; it is not evidence that lenders have committed to finance the next customer deployment.
Nebius is not first to convert a customer contract and deployed GPUs into institutional funding. IREN said it closed a $3.65 billion investment-grade facility in June to support its Microsoft AI cloud contract, secured against GPUs and associated contracted cash flows.
IREN’s package was larger and had slightly lower stated margins, but its economics are not a clean benchmark for Nebius. It combined a fixed-rate private placement with a floating delayed-draw loan, used interest-rate hedges on the latter, identified Microsoft as the customer and obtained ratings of A from Fitch and A(low) from DBRS. Nebius disclosed a single floating-rate facility and left the customer unnamed.
| Company disclosure | Debt structure and stated price | Customer support and reported coverage |
|---|---|---|
| Nebius | Approximately $775 million at one-month Term SOFR + 2.50% | Unnamed investment-grade customer; facility plus contracted cash flows cover more than 100% of underlying GPU capital expenditure |
| IREN | $2.10 billion fixed-rate private placement equivalent to SOFR + 2.13%, plus $1.55 billion delayed-draw term loan at SOFR + 2.25% | Microsoft; debt plus customer prepayments fund $5.59 billion, or about 96%, of $5.81 billion in GPU capital expenditure |
IREN also reported a 6.00% blended cost of debt. Its separate 3.31% average financing-cost figure treats a $1.94 billion customer prepayment as a 0% funding source over the contract term. Neither measure can be compared directly with Nebius’s 2.50-percentage-point margin, which excludes the prevailing SOFR base rate.
IREN attributed its access to institutional capital to both the quality of the Microsoft contract and its ownership of the data-center infrastructure housing the GPUs. That claim reinforces the competitive lesson: creditworthy offtake, control of infrastructure and an operating asset—not the label “AI cloud” by itself—shape financing access and price.
Nebius has also used corporate securities to pay for rapid expansion. It priced $2.75 billion of senior unsecured convertible notes in September 2025 and proposed a further $3.75 billion convertible offering in March 2026, the financing account says. The new facility differs because the debt, collateral and covenants sit around two project subsidiaries.
That distinction creates an option to fund some expansion outside parent-level unsecured markets, but the current disclosures do not show that Nebius can stop issuing corporate debt or equity. The company has secured up to 1.2 gigawatts of power and land for a company-owned AI factory in Pennsylvania, while data centers and GPU equipment require heavy spending before all related revenue is recognized. A $775 million facility addresses part of that capital requirement; it does not show when the Pennsylvania site will become operating capacity or how much of its hardware the loan will fund.
Nebius’s growth helps explain why it wants another funding channel. The same account says AI cloud revenue rose from $68.3 million in 2024 to $480.3 million in 2025. First-quarter 2026 consolidated revenue reached $399 million, compared with $50.9 million a year earlier, while adjusted EBITDA moved from a $53.7 million loss to positive $129.5 million.
The company says it has more than $40 billion of additional contracted revenue from investment-grade customers including Microsoft and Meta. That figure also needs qualification. The account, citing Nebius’s 2025 annual report, describes Meta orders with potential value of approximately $27 billion: $12 billion for dedicated GPU clusters and up to $15 billion for access to unsold capacity from specified clusters under defined conditions. Maximum potential contract value is not the same as cash flow from an accepted deployment that lenders are ready to underwrite.
Nebius said it recently delivered the latest planned Microsoft capacity tranche and remained on schedule for the rest. Neither its loan announcement nor the filing-based account identifies Microsoft, Meta or another company as the customer behind this facility.
Nebius calls the loan a framework for financing other long-term customer deployments and expects to raise more capital on similarly attractive terms. Those are company expectations, not commitments from the current bank syndicate.
The evidence that would resolve the central question is a second facility with enough disclosure to compare it properly: the customer or its credit quality, whether the deployment is already accepted and producing cash, the amount advanced relative to project cost, the contribution from customer prepayments or future receipts, the total borrowing cost and the guarantees retained at group level.
Delivery matters as much as financing. Investors also need to see whether Nebius turns secured power and land into operating capacity on schedule, stays within the new covenants, and preserves room for a broader AI cloud business rather than building mainly around a few large contracts.
Until then, the $775 million deal proves that one mature deployment can carry asset-level debt. It does not yet prove that Nebius’s wider backlog is a repeatable financing engine.
OpenAI has put conversations and Projects back in its redesigned ChatGPT desktop app and enabled cloud Work threads to move across devices, correcting the launch's biggest usability failures without merging local Work or Codex histories.
Meta reportedly plans to put departing AWS compute executive Dave Brown to work on its data-center buildout, adding hyperscale operating experience while leaving any customer-facing cloud business conditional and undefined.
China has paired a five-year AI training offer for developing countries with cooperation centers, a weather-warning rollout and a new 29-country organization. The package gives Beijing a platform for influence, but no budget, selection rules or delivery timetable has been published.
RoboTTT extends a robot policy’s visuomotor context to 8,000 timesteps and raises its average rubric score from 42% to 79%, but the evidence comes from three author-run assembly tasks and the longest task was completed in only two of ten trials.
Zhipu reportedly reached $1 billion in annual recurring revenue in July, roughly four times a March estimate, but the unconfirmed run rate is not annual sales and still sits far ahead of recognized cloud revenue while margins remain thin.
Cursor says automatically running a repository-root git.exe on Windows does not meet its criteria for patching, while the researcher calls it an untrusted-search-path defect and separate research shows the same weakness across several competing AI coding tools.
Huawei publicly displayed a 16-cabinet Atlas 950 configuration rated at 1 EFLOPS in FP8, providing tangible evidence of its system-scale AI strategy while leaving price, power use and sustained workload performance undisclosed ahead of the full system's planned fourth-quarter release.
CIA Director John Ratcliffe said US intelligence is consistent with an estimate that Russian recruits last 20 to 30 minutes on Ukraine’s battlefield, but the public trail leads to an unsourced claim about assault troops and does not establish a representative average.
UK testing places leading open-weight models four to seven months behind selected closed-model cyber results, yet longer attack chains, U.S. benchmarks and mixed cost comparisons show why that interval is a warning signal rather than a universal capability clock.
TSMC reached the top of its second-quarter revenue guidance and raised its 2026 outlook, but a one-off investment gain boosted profit growth while the company committed more capital to 2-nanometer production, advanced packaging and an undated Arizona expansion.
Moonshot AI's Kimi K3 added to a global technology selloff with near-frontier performance, but unreleased weights, mixed cost comparisons and a recommended 64-accelerator deployment leave its effect on chip demand unresolved.
Moonshot AI has made Kimi K3 available through its apps and API, pairing a 2.8-trillion-parameter architecture with early frontier-level results, but the model's open-weight claim cannot be tested until its weights and technical report arrive.
Two binding EU decisions require Google to give rival AI services comparable access to 11 Android features and offer eligible search competitors a restricted, anonymized dataset, but phased deadlines, certification, pricing and privacy safeguards leave the competitive effect unproven.
Databricks has signed a term sheet for a Coatue-led financing at a $188 billion valuation, while unidentified sources put the round at $3 billion. The proposed capital would deepen its push into AI governance, data agents and operational databases, but the transaction remains open and the company supplied no new operating figures.
Netflix says generative AI workflows were used on roughly 300 titles in 2026, but its only quantified example tied to that disclosure covers 17 minutes and does not establish how much finished material, spending or labor changed across the slate.
Gemini 3.5 Pro missed its expected June rollout. An anonymous-source account says a late-June data change fell short of Google's coding goals, but Google has confirmed only partner testing—not the reported cause, a new date, or public results and pricing.
Google is bringing Gemini Omni editing and a reusable face-and-voice avatar to Vids, but the sharper distinction is account-level identity across Vids and Gemini rather than a new category of AI video; Vids already offered Veo generation and customizable avatars, while specialist rivals already sell digital presenters.
Intel plans to adopt Gemini Enterprise across engineering, supply-chain and corporate work while extending chip-simulation capacity into Google Cloud, but the companies supplied no rollout schedule, contract value or performance baseline.
The Navy has approved an immediate department-wide framework for turning data into operational effects, but the public rollout leaves budgets, owners, deadlines, performance measures and assurance rules to a promised implementation roadmap.
Runta has raised a $20 million seed round led by Andreessen Horowitz for an execution layer that governs agents at the operating-system and network levels, entering a market where cloud runtimes and sandbox providers already offer overlapping controls.