Read the structure. Then read the terms.
Analysis, guides and reports on how data centres, GPUs and the power behind them are financed, built from the filed documents. Choose a layer of the stack to see everything written about it.
- Pieces in the library
- 75
- Analysis articles
- 21
- Finance terms defined
- 111
Every financing is a stack: power, the building, the compute, the contracts and the capital. Point at a layer.
Where the cash stops: DSCR and cash trap levels in compute finance
Every borrower asks what DSCR the market will accept. The better question is where the cash trap bites, because that is the covenant that changes behaviour long before a default does.
Read → Deal of the monthLambda Compute I: GPU debt reaches the insurance market
US$1.008bn at 6.78% fixed, rated A (low) and Baa1. Structure, terms and rating drivers.
Read → Free report · 25-page PDF · 25 pagesThe GPU Financing Handbook
How the 31 public GPU-backed loans since 2023 are structured, priced and protected: six structure diagrams, a term benchmark and a term sheet checklist.
Get the report →The long reads.
Structure diagrams, benchmarks and checklists, drawn from credit agreements, indentures and rating reports. All reports →
The GPU Financing Handbook
How the 31 public GPU-backed loans since 2023 are structured, priced and protected: six structure diagrams, a term benchmark and a term sheet checklist.
Get the report →Free report · 12-page PDF · 12 pagesWhat Aircraft and Infrastructure Finance Teach GPU Lenders
Ten lessons from aircraft, shipping and project finance for GPU lenders, with structure diagrams and trigger charts.
Get the report → Annual report · January 2027State of AI Infrastructure Finance 2026
The year in data centre, GPU and hyperscaler debt. Subscribers receive it first.
Subscribe →What the deals show.
Short, sourced pieces on pricing, sizing, security and the contracts behind them, each with one exhibit.
Does the debt outrun the chip?
Long-dated GPU loans depend on repayment cash flows while hardware values can move sharply. Lessons from aircraft finance help frame the mismatch, but legal recovery rules remain different.
Read →Benchmarks · 5 min readWhere the cash stops: DSCR and cash trap levels in compute finance
Every borrower asks what DSCR the market will accept. The better question is where the cash trap bites, because that is the covenant that changes behaviour long before a default does.
Read →Markets · 5 min readFrom bitcoin to AI: how former miners are financing the switch
Bitcoin miners own what AI needs most: powered land. Three of them have now financed the switch to AI, in three different ways.
Read →Cross-sector lessons · 5 min readIs there a Cape Town Convention for compute? What aircraft repossession teaches GPU lenders
Aircraft lenders can find, register and repossess their collateral under an international treaty. GPU lenders have nothing comparable. The 2026 deals show how the market is filling the gap with contracts.
Read →Cross-sector lessons · 5 min readWhy data centre ABS warns earlier than GPU loans
Data centre securitisations start trapping cash at 1.30x to 1.35x. GPU project loans wait until 1.10x. That does not make GPU loans riskier. It reflects two different answers to one question: when must the debt be repaid?
Read →Pricing · 4 min readFixed or floating? Why GPU loans now come with fixed-rate tranches
Until 2025 almost every GPU loan was floating rate. In 2026 fixed-rate options appeared in most of the large deals. The reason is simple: the revenue was always fixed.
Read →Markets · 5 min readGPU financing outside the US: Iceland, the Nordics, Europe, the Gulf and Asia-Pacific
The largest GPU loans finance US campuses, but the market outside the US is growing quickly. Iceland and the Nordics lead on count. The Gulf and Australia lead on size.
Read →Deal anatomy · 4 min readLeasing GPUs like aircraft: the lessor model behind xAI's Colossus 2
Most GPU loans lend to the company that uses the GPUs. The xAI financing put the GPUs in a separate lessor and leased them back, a model aircraft finance knows well.
Read →How each structure works.
From GPU loans and data centre ABS to powered land and convertibles: the cash flows, the documents and the risks.
Hyperscaler support: separate customer credit from lender recourse
A guide to leases, guarantees, backstops and the claims they actually support.
Read →Structure guideData centre ABS: follow the pool and the waterfall
A practical reading of securitised data-centre cash flows, triggers and expected repayment.
Read →Structure guideGPU-backed loans: contracts, equipment and repayment
How GPU collateral and customer cash flows support lending, and where the two can diverge.
Read →Structure guideHyperscaler lease-backed project bonds
How a strong customer contract supports a project without guaranteeing its bonds.
Read →Structure guideColocation versus build-to-suit: data centre models and how each is financed
Retail colocation, wholesale, hyperscale leases, build-to-suit and powered shell compared: who the customer is, how long the contract runs and which financing tools fit each model, with public deal examples.
Read →Structure guideGPU ABS: how rated and securitised GPU financing works
What GPU ABS means, how rated GPU financings such as Lambda Compute I are structured, what rating agencies look at, and how the model compares with aircraft ABS. Public sources only.
Read →Structure guideWhat is a neocloud, and how is it financed?
A plain-English guide to neoclouds: what they are, how they differ from hyperscalers and the financing stack they use, from GPU delayed draw term loans to high-yield bonds, convertibles and vendor finance.
Read →Structure guideData centre CMBS: property credit meets operating infrastructure
How mortgage collateral, tenant income and refinance exposure fit together.
Read →Structure guideData centre construction finance
Funding a build from land and power milestones to a financeable operating asset.
Read →Structure guideGPU borrowing bases: what actually supports availability
From contract eligibility to concentration limits, valuation and a funding shortfall.
Read →Structure guideResidual value guarantees and GPU exit risk
What a value floor can cover, and the recovery risks that can remain.
Read →Structure guideConvertible finance for neoclouds
Read the equity option, cash obligations and maturity together.
Read →Structure guideFinancing powered land: prove the power milestone
A financing framework for land, connection rights and delivery risk.
Read →Structure guideECA and DFI finance for data centres
Match official finance to an eligible project and its actual obligations.
Read →Structure guideGulf and Asian capital in global compute
Separate investor geography from asset location and financing recourse.
Read →The clauses that carry the credit.
Security packages, step-in rights, cash traps and covenants, read from filed documents.
Data centre ABS triggers: cash trap, rapid amortisation and the ARD
How the DSCR trigger ladder works in data centre securitisations, with the actual cash trap and amortisation levels from the STACK, Centersquare, Compass, Vantage and Aligned master indentures.
Read →Clause explainerCash sweeps and cash traps: how lenders hold back cash
The difference between a cash sweep and a cash trap, how each is triggered and released, and where they appear in public GPU loans, data centre ABS and lease-backed project bonds.
Read →Clause explainerFinancial covenants in GPU loans: DSCR, contract realisation and loan-to-cost
The financial tests in public GPU-backed loans explained: sizing DSCR, maintenance DSCR, the contract realisation ratio and depreciating loan-to-cost tests, with figures from CoreWeave, Nscale and IREN filings.
Read →Clause explainerThe GPU security package: what lenders take in a GPU-backed loan
What lenders take as security in a GPU-backed loan: the SPV, equity pledge, GPUs, customer contracts, controlled accounts, direct agreements and parent support, with examples from public filings.
Read →Clause explainerOfftake step-in rights and direct agreements in compute financing
How step-in rights, direct agreements and recognition agreements protect lenders, customers and landlords in GPU and data centre financings, with examples from IREN, Nscale, Cipher and CoreWeave filings.
Read →Underwriting the compute stack.
Twelve visual guides to the decisions behind GPU lending, from funding to recovery.
Financing a GPU project
Start with the cash obligations that must survive a difficult launch, then choose the funding structure.
Read →Lending handbook · RecoveryKeeping a cluster operating through distress
Identify what failed, preserve lawful options and compare the value of continued service with a net sale recovery.
Read →Lending handbook · EquipmentFrom equipment value to net recovery
An equipment price, an accounting value and a lender's net recovery answer different questions.
Read →Lending handbook · DeliveryThe cash cost of a delayed launch
Link procurement, facility readiness and customer acceptance to cash rather than treating delay as one contingency line.
Read →Lending handbook · DeliveryReviewing the facility behind the loan
Review the operating envelope and the lender's practical access before treating colocation as a settled input.
Read →Lending handbook · EquipmentFrom equipment quote to collateral schedule
Reconcile price, configuration and ownership before applying an advance rate to the invoice total.
Read →Lending handbook · FinancingFinancing a modular data centre
A modular facility can shorten delivery. Its financing still depends on who controls the equipment, the site and the route to power.
Read →Lending handbook · FinancingSizing debt against operating cash
GPU capacity becomes financeable through a sequence of deposits, deliveries, acceptance and collections. Model that sequence before choosing leverage.
Read →Lending handbook · Market structureComparing neocloud business models
Classify operators by the cash they earn, the assets they control and the entity that borrows. A company count cannot answer those questions.
Read →Lending handbook · RecoveryWhat the risk controls actually cover
Insurance, cash reserves and replacement operators act on different failures. The value lies in the response path, including what happens before cash arrives.
Read →Lending handbook · FinancingWho benefits from a vendor backstop?
Capacity purchases, landlord guarantees and residual insurance sit in different places in the financing structure. Follow the payment obligation before assigning credit.
Read →Lending handbook · Market structureRack economics: October 2026
October’s secondary-market evidence shows why a single depreciation rate can give a misleading view of a financed server fleet.
Read →Move through the structure.
Five interactive cases that keep the disclosed terms and the limits of the evidence together.
When the debt outlasts the contract
Compare the disclosed debt tenor and average customer-contract length. Then examine what renewal risk means.
Read →Interactive case studyFollow the promise. Find the limit
Trace the customer payments, limited guarantees and collateral that sit between a contract and a lender.
Read →Interactive case studyHow much is actually additional?
Separate the previous commitment, the increase and the resulting facility. Keep the amendment attached to its history.
Read →Interactive case studyOne package. Different rate exposures
Read the fixed coupon, floating spread, hedge overlay and commitment fee on their own terms.
Read →Interactive case studyCredit capacity is a step toward power
Follow a utility-credit facility from its disclosed purpose to the delivery evidence still needed.
Read →All 75 pieces.
Filter by layer of the stack or by format, or search. Newest first.
Showing all 75
- AnalysisDoes the debt outrun the chip?Long-dated GPU loans depend on repayment cash flows while hardware values can move sharply. Lessons from aircraft finance help frame the mismatch, but legal recovery rules remain different.Compute
- AnalysisWhere the cash stops: DSCR and cash trap levels in compute financeEvery borrower asks what DSCR the market will accept. The better question is where the cash trap bites, because that is the covenant that changes behaviour long before a default does.Capital
- GuideHyperscaler support: separate customer credit from lender recourseA guide to leases, guarantees, backstops and the claims they actually support.Contracts
- GuideData centre ABS: follow the pool and the waterfallA practical reading of securitised data-centre cash flows, triggers and expected repayment.Buildings
- GuideGPU-backed loans: contracts, equipment and repaymentHow GPU collateral and customer cash flows support lending, and where the two can diverge.Compute
- GuideHyperscaler lease-backed project bondsHow a strong customer contract supports a project without guaranteeing its bonds.Buildings
- PodcastCompute & CapitalThe podcast for the lawyers building AI infrastructure. Season 1 in production.Contracts
- AnalysisFrom bitcoin to AI: how former miners are financing the switchBitcoin miners own what AI needs most: powered land. Three of them have now financed the switch to AI, in three different ways.Power
- AnalysisIs there a Cape Town Convention for compute? What aircraft repossession teaches GPU lendersAircraft lenders can find, register and repossess their collateral under an international treaty. GPU lenders have nothing comparable. The 2026 deals show how the market is filling the gap with contracts.Compute
- AnalysisWhy data centre ABS warns earlier than GPU loansData centre securitisations start trapping cash at 1.30x to 1.35x. GPU project loans wait until 1.10x. That does not make GPU loans riskier. It reflects two different answers to one question: when must the debt be repaid?Buildings
- AnalysisFixed or floating? Why GPU loans now come with fixed-rate tranchesUntil 2025 almost every GPU loan was floating rate. In 2026 fixed-rate options appeared in most of the large deals. The reason is simple: the revenue was always fixed.Capital
- AnalysisGPU financing outside the US: Iceland, the Nordics, Europe, the Gulf and Asia-PacificThe largest GPU loans finance US campuses, but the market outside the US is growing quickly. Iceland and the Nordics lead on count. The Gulf and Australia lead on size.Capital
- AnalysisLeasing GPUs like aircraft: the lessor model behind xAI's Colossus 2Most GPU loans lend to the company that uses the GPUs. The xAI financing put the GPUs in a separate lessor and leased them back, a model aircraft finance knows well.Compute
- AnalysisThe customer sets the price: why GPU loan margins fell from 8.75% to 2.25%In three years the margin on GPU-backed loans has fallen by about three quarters. The hardware did not change much. What changed was who stands behind the contract.Capital
- AnalysisThe spares reserve: compute's version of the maintenance reserveAircraft lessors have long made airlines pay to keep the asset in condition. GPU lenders are starting to do the same, with a funded reserve for spare hardware.Contracts
- AnalysisHow lenders size a GPU loan: two tests, and the smaller one winsEvery GPU loan in the public market is sized twice: once against the cost of the hardware and once against the cash the contract will produce. Here is how the two tests work, with a worked example.Capital
- AnalysisWhy the customer contract has to sit in the borrowerIn a GPU loan the customer contract is the credit. If the contract sits with the parent, the lenders are lending against someone else's asset. The 2026 deals all move it into the borrower.Contracts
- AnalysisSplitting the GPUs from the building: inside Nscale's Ward County loanNscale's US$1.85bn Ward County financing puts the GPUs and the data centre in two separate companies, then brings both inside the lenders' security. It is one of the clearest answers yet to the question every GPU lender asks: how do I get to my collateral if things go wrong?Compute
- AnalysisWho is NVIDIA in your capital stack?NVIDIA makes the collateral in every GPU loan. In some deals it also sits in the cash flow or the equity. Lenders need to know which.Capital
- AnalysisOn-chain GPU financing in plain EnglishAlongside the bank-led loans, a small on-chain market lends against tokenised receipts for GPUs. The headline numbers are large. The commitments are not.Capital
- AnalysisFull, limited or none: how much parent support GPU lenders takeRecourse to the sponsor is one of the most negotiated points in a GPU loan. The public deals show every answer, sometimes within one sponsor's own facilities.Contracts
- AnalysisPooled or single-contract? Two ways to build a GPU facilityA GPU facility can finance a book of customer contracts or a single one. CoreWeave has done both, and the pricing shows what each design costs.Contracts
- AnalysisRated GPU debt: how a non-investment-grade sponsor gets to A (low)Three GPU financings now carry public ratings, two of them investment grade. The sponsors are not investment grade. The ratings come from the structure.Capital
- AnalysisHow big should a GPU debt service reserve be?When DBRS rated Lambda's Compute I loan, it took half a notch off for one reason: the debt service reserve was lighter than the project finance norm. That puts a price on liquidity.Capital
- AnalysisWho lends to neoclouds? From private credit to bank-led project financeIn 2023 GPU loans came from a handful of private credit funds. By 2026 the largest banks were arranging them and rating agencies were grading them. Volume followed.Contracts
- ReportThe GPU Financing HandbookHow the 31 public GPU-backed loans since 2023 are structured, priced and protected: six structure diagrams, a term benchmark and a term sheet checklist.Compute
- ReportWhat Aircraft and Infrastructure Finance Teach GPU LendersTen lessons from aircraft, shipping and project finance for GPU lenders, with structure diagrams and trigger charts.Compute
- Deal noteLambda Compute I: GPU debt reaches the insurance marketUS$1.008bn at 6.78% fixed, rated A (low) and Baa1. Structure, terms and rating drivers.Capital
- ToolDebt maturity calendarWhen AI infrastructure debt falls due: 274 maturities from public filings.Capital
- ToolFinancing clause libraryCash traps, DSCR, cures, step-in and GPU security, quoted from filed deals.Contracts
- ToolGPU security by countryHow lenders take and perfect security over GPUs in 17 countries.Compute
- AnalysisOctober 2026 research briefing: follow the obligationThree questions that connect contract credit, delivery and financing economics.Contracts
- ExplainerData centre ABS triggers: cash trap, rapid amortisation and the ARDHow the DSCR trigger ladder works in data centre securitisations, with the actual cash trap and amortisation levels from the STACK, Centersquare, Compass, Vantage and Aligned master indentures.Buildings
- ExplainerCash sweeps and cash traps: how lenders hold back cashThe difference between a cash sweep and a cash trap, how each is triggered and released, and where they appear in public GPU loans, data centre ABS and lease-backed project bonds.Capital
- ExplainerFinancial covenants in GPU loans: DSCR, contract realisation and loan-to-costThe financial tests in public GPU-backed loans explained: sizing DSCR, maintenance DSCR, the contract realisation ratio and depreciating loan-to-cost tests, with figures from CoreWeave, Nscale and IREN filings.Capital
- ExplainerThe GPU security package: what lenders take in a GPU-backed loanWhat lenders take as security in a GPU-backed loan: the SPV, equity pledge, GPUs, customer contracts, controlled accounts, direct agreements and parent support, with examples from public filings.Compute
- ExplainerOfftake step-in rights and direct agreements in compute financingHow step-in rights, direct agreements and recognition agreements protect lenders, customers and landlords in GPU and data centre financings, with examples from IREN, Nscale, Cipher and CoreWeave filings.Contracts
- GuideColocation versus build-to-suit: data centre models and how each is financedRetail colocation, wholesale, hyperscale leases, build-to-suit and powered shell compared: who the customer is, how long the contract runs and which financing tools fit each model, with public deal examples.Buildings
- GuideGPU ABS: how rated and securitised GPU financing worksWhat GPU ABS means, how rated GPU financings such as Lambda Compute I are structured, what rating agencies look at, and how the model compares with aircraft ABS. Public sources only.Capital
- GuideWhat is a neocloud, and how is it financed?A plain-English guide to neoclouds: what they are, how they differ from hyperscalers and the financing stack they use, from GPU delayed draw term loans to high-yield bonds, convertibles and vendor finance.Compute
- BenchmarkAI infrastructure credit index: Q3 2026US$129.8bn across 44 deals. Volume index 157.Capital
- GuideData centre CMBS: property credit meets operating infrastructureHow mortgage collateral, tenant income and refinance exposure fit together.Buildings
- GuideData centre construction financeFunding a build from land and power milestones to a financeable operating asset.Buildings
- GuideGPU borrowing bases: what actually supports availabilityFrom contract eligibility to concentration limits, valuation and a funding shortfall.Compute
- GuideResidual value guarantees and GPU exit riskWhat a value floor can cover, and the recovery risks that can remain.Compute
- GuideConvertible finance for neocloudsRead the equity option, cash obligations and maturity together.Capital
- GuideFinancing powered land: prove the power milestoneA financing framework for land, connection rights and delivery risk.Power
- GuideECA and DFI finance for data centresMatch official finance to an eligible project and its actual obligations.Capital
- GuideGulf and Asian capital in global computeSeparate investor geography from asset location and financing recourse.Capital
- HandbookFinancing a GPU projectStart with the cash obligations that must survive a difficult launch, then choose the funding structure.Capital
- HandbookKeeping a cluster operating through distressIdentify what failed, preserve lawful options and compare the value of continued service with a net sale recovery.Compute
- HandbookFrom equipment value to net recoveryAn equipment price, an accounting value and a lender's net recovery answer different questions.Compute
- HandbookThe cash cost of a delayed launchLink procurement, facility readiness and customer acceptance to cash rather than treating delay as one contingency line.Buildings
- HandbookReviewing the facility behind the loanReview the operating envelope and the lender's practical access before treating colocation as a settled input.Buildings
- HandbookFrom equipment quote to collateral scheduleReconcile price, configuration and ownership before applying an advance rate to the invoice total.Compute
- HandbookFinancing a modular data centreA modular facility can shorten delivery. Its financing still depends on who controls the equipment, the site and the route to power.Capital
- HandbookSizing debt against operating cashGPU capacity becomes financeable through a sequence of deposits, deliveries, acceptance and collections. Model that sequence before choosing leverage.Capital
- HandbookComparing neocloud business modelsClassify operators by the cash they earn, the assets they control and the entity that borrows. A company count cannot answer those questions.Capital
- HandbookWhat the risk controls actually coverInsurance, cash reserves and replacement operators act on different failures. The value lies in the response path, including what happens before cash arrives.Compute
- HandbookWho benefits from a vendor backstop?Capacity purchases, landlord guarantees and residual insurance sit in different places in the financing structure. Follow the payment obligation before assigning credit.Capital
- HandbookRack economics: October 2026October’s secondary-market evidence shows why a single depreciation rate can give a misleading view of a financed server fleet.Capital
- Case studyWhen the debt outlasts the contractCompare the disclosed debt tenor and average customer-contract length. Then examine what renewal risk means.Contracts
- Case studyFollow the promise. Find the limitTrace the customer payments, limited guarantees and collateral that sit between a contract and a lender.Contracts
- Case studyHow much is actually additional?Separate the previous commitment, the increase and the resulting facility. Keep the amendment attached to its history.Capital
- Case studyOne package. Different rate exposuresRead the fixed coupon, floating spread, hedge overlay and commitment fee on their own terms.Capital
- Case studyCredit capacity is a step toward powerFollow a utility-credit facility from its disclosed purpose to the delivery evidence still needed.Power
- BenchmarkQ4 2026 credit monitor: interimThe register at a glance, its composition and the questions still open.Capital
- ToolInside a data centreA scroll cutaway of the building, the compute and the land and power behind it.Power
- ToolGPU borrowing baseHow debt capacity falls as hardware ages and contracts run off.Compute
- ToolFinancing cost calculatorAll-in cost of a GPU or data centre facility from disclosed margins.Capital
- BenchmarkGPU loan marginsDisclosed margins over SOFR, set against customer credit.Capital
- BenchmarkTrigger ladderWhere cash traps, amortisation triggers and defaults sit.Capital
- BenchmarkThe tenor gapDebt tenor against customer contract length.Contracts
- BenchmarkLTV and reservesLoan to value, loan to cost and reserve sizing across deals.Capital
- BenchmarkHyperscaler bondsThe investment-grade bonds funding AI build-out.Capital
Nothing matches.
Cash trap
A contractual mechanism that retains cash otherwise available for distribution when a specified test is failed. In filed GPU loans the trap usually sits at 1.10x to 1.15x DSCR; in data centre ABS at 1.30x to 1.35x.
All 111 terms →Notes from the register.
Dated notes on facility histories, project debt and utility credit support, kept with their sources.
An upsize needs two numbers.
Vantage announced two different transactions in June 2025: construction debt for New Albany and an increase to a corporate facility. The corporate amendment added $2.75bn to a $3bn commitment, producing a $5.75bn facility. The increment and the resulting capacity are two descriptions of one amendment.
DataBank’s October 2025 construction facility similarly increased from $725m to $1.6bn. The difference is $875m. Neither the resulting commitment nor the arithmetic increment establishes how much was drawn.
Sources: Vantage, 3 June 2025 ↗; DataBank, 17 October 2025 ↗.
Inspect the facility history →A coupon leaves questions open.
Yondr’s public investor-relations page confirms that Yondr JK 1, LLC issued $715m of 6.875% senior secured notes on 30 June 2026. The earlier pricing announcement describes funding part of the construction of a 48MW critical-IT project in Loudoun County, together with reserves and fees.
The nominal coupon is a checked term. It is not a complete borrower financing cost: the public material does not supply the full dated amortisation, issue-price, fee and reserve schedule. Confidential investor material was not accessed. The project link establishes stated purpose; it does not verify completion or energisation.
Sources: Yondr pricing release, 24 June 2026 ↗; public issuer confirmation, checked 22 October 2026 ↗.
Read the source observations →Credit support is one part of securing power.
Yondr’s June 2026 disclosure describes a global letter-of-credit facility supporting utility security requirements. That purpose is relevant to development bankability, but no commitment amount, reserved megawatts or named grid connection is established by the announcement.
We classify the facility as contingent credit and retain the amount as unknown. We do not add it to funded borrowing or infer a project-level power allocation.
Source: Yondr corporate facilities announcement, 2 June 2026 ↗.
Inspect the contingent facility →Read the scope beside the number.
The register includes historical facilities, amendments, package components and capital observations. Amounts cannot be added into a complete market total. Disclosed margins and coupons do not establish comparable all-in financing costs.
Coverage remains incomplete. Source links and record qualifications identify what the public disclosures support and where terms remain unknown.
Read the methodology →The week in compute finance.
Financing announcements and the terms worth reading, in one short email. Work email only.
Subscribe → Podcast · coming soonCompute & Capital
Conversations with the general counsel building data centres and GPU clouds.
Find out more →