How AI Infrastructure Gets Financed: Data Centers, Debt and Private Credit
A data center needs funding before it earns rent. Follow how equity, construction loans, customer contracts and refinancing pay for the buildout, and who carries the risk.
Guide updated October 13, 2026. The financing examples below retain their original announcement dates; forecasts and financing frameworks are not completed spending.
A data-center developer pays for land, design and construction before the facility earns rent. A signed customer contract can help obtain financing, but someone still has to fund the work and carry the risk that the building or electricity arrives late.
Artificial intelligence (AI) infrastructure gets financed through corporate spending, equity, bank loans, bonds, project debt and private credit. The combination depends on what is being financed and when it starts earning. Large technology companies can use their own cash and borrowing; developers and infrastructure owners may raise capital against projects, leases and equipment. Customer contracts support the financing but are not themselves a loan.
This guide follows how that financing works from development through repayment and refinancing. For the organizations providing the capital, use Who Finances AI Data Centers?. For the electricity projects and grid connections, see Who Finances the Power Behind AI Data Centers?.
How financing changes as a project develops
Financing can change as a site becomes an operating asset. This is a common sequence, not a requirement that every project use each step.
Development: pay for a project that does not yet earn
Owners may use equity or corporate financing for land, design and permits. Before committing construction funding, lenders examine customer contracts, power access and the developer's ability to finish.
Construction: fund work as it progresses
Construction loans may release funds as conditions are met. Cost overruns, contractor performance and completion dates affect the financing. The documents determine who must provide more money if the original budget falls short.
Operations: collect enough to service the debt
Once a facility starts operating, rent and other customer payments can support debt service. Lenders examine payment obligations, operating costs and tenant creditworthiness. A signed lease does not remove the need to deliver the contracted facility.
Refinancing: replace the original debt
An owner may refinance with longer-term loans or securitized debt. That depends on the asset's performance and market conditions; it is not assured when construction financing is signed. A delay can leave the original loan nearing maturity before the expected refinancing is available.
For the mechanics of loan payments and returns, our Academy lesson on yield, fees and credit losses provides a worked example. Its business development company, or BDC, loan example is educational, not a model of a specific data-center financing.
The quick answer: how AI infrastructure gets financed
AI infrastructure financing usually combines several layers of money.
Hyperscaler capital expenditure: very large cloud and technology companies fund chips, buildings and capacity from cash or financing raised on their own balance sheets.
Public bonds and corporate debt let investment-grade companies finance AI spending while preserving liquidity and spreading the funding burden across institutional investors.
Bank credit supports revolving credit facilities, construction loans, bridge loans, project finance, warehouse lines, and relationship lending.
Data-center leases and contracts can turn customer demand into predictable cash flows that lenders are willing to underwrite.
Real estate investment trusts, or REITs, and other property investors finance land, buildings, leasing platforms and development pipelines.
Infrastructure funds finance long-lived assets such as power generation, grid equipment, fiber, cooling systems, substations, and data-center platforms.
Asset-backed finance can support equipment, leases, receivables, contracts, power assets, and pools of cash-flowing collateral.
Private credit finances borrowers and structures that do not fit neatly into public bonds, bank loans, or ordinary real-estate finance.
Insurance capital can provide long-duration money that matches long-lived infrastructure cash flows.
Business development companies, or BDCs, may lend to private companies involved in the buildout. Exposure must be established from the BDC's own holdings, not inferred from its manager's name.
These sources can overlap. A private infrastructure fund may provide equity while a bank lends to the same project. The fund's ownership risk and the bank's loan risk are different claims on that business.
What this looks like in real deals
The following dated examples show different uses of financing. Their amounts should not be added together: one is a spending forecast, another is completed refinancing, and the third is a financing framework.
In its June 12, 2026 article, Goldman Sachs Research forecast $5.3 trillion in capital spending from 2025 through 2030 by the large technology companies leading the buildout. This is an estimate of spending, not an amount of private credit raised or already deployed.
On March 25, 2025, Switch announced completed securitized financings totaling $3.5 billion: a $2.4 billion commercial mortgage-backed securities, or CMBS, transaction and approximately $1.1 billion of asset-backed securities, or ABS. Switch said the proceeds would refinance most of its outstanding acquisition financing. This was refinancing, not a reported $3.5 billion of new construction spending.
On June 30, 2026, Bloom Energy announced that Brookfield had expanded its framework to finance AI-infrastructure power projects from $5 billion to $25 billion. A framework's size is not proof that the full amount has been funded or that the projects are operating.
Read each announcement for the transaction it describes: the borrower, the use of proceeds, the financing vehicle and what has actually closed. That is how a large headline becomes useful financial information.
Why AI infrastructure is different from software
Software can scale quickly once the code works. AI infrastructure cannot.
A model can be copied. A data center cannot. A cloud product can launch globally. A substation cannot be permitted globally overnight. A chip order can be announced in a sentence, but power capacity, grid interconnection, cooling systems, and long-term financing take years.
That is why the AI buildout behaves less like an app cycle and more like an infrastructure cycle. Infrastructure cycles are slower, more capital-intensive, and more dependent on financing terms. They also create a different kind of risk.
The market can reprice a software company in a day. A half-built data center cannot be repriced that cleanly. It has land, contracts, contractors, equipment, debt, power obligations, tenants, delays, cost overruns, and refinancing needs.
That is why AI infrastructure should be read through a credit lens, not only an equity lens.
The hyperscaler layer: the demand engine
The first layer is the hyperscaler. Microsoft, Amazon, Alphabet, Meta, Oracle, and other large platforms create much of the demand that pulls the system forward. They buy chips, lease capacity, build data centers, sign power agreements, and create the revenue expectations that make the rest of the stack financeable.
Hyperscalers can fund a large share of this through operating cash flow and balance sheets. They can also issue bonds, use leases, form joint ventures, contract with developers, or rely on outside infrastructure capital.
That matters because hyperscaler demand often turns into someone else’s financing opportunity. A cloud company signs a lease. A data-center developer raises capital. A power project gets financed. A supplier expands capacity. A private borrower needs debt. An infrastructure fund underwrites a long-term asset.
The hyperscaler starts the chain, but the capital stack carries it.
The data-center developer layer: turning demand into a project
Demand is not infrastructure until it becomes a site.
A developer needs land, permits, power access, building materials, cooling design, fiber connectivity, customer commitments, construction partners, and financing. A project may look attractive because AI demand is growing, but lenders do not underwrite a headline. They underwrite contracts, costs, collateral, completion risk, tenant quality, power availability, and exit value.
This is where AI infrastructure begins to look like project finance, real-estate finance, and infrastructure credit. A project with a strong tenant, a long lease, secured power, credible construction partners, and conservative leverage is very different from a speculative site built on loose demand assumptions and expensive debt.
Both can be called AI infrastructure. Only one may deserve cheap capital.
The lease layer: how demand becomes financeable cash flow
Leases are one of the places where the AI story becomes financeable.
A data center with no customer is an expensive box. A data center with a creditworthy tenant, a long-term lease, contracted power, and predictable revenue becomes something lenders can underwrite.
That does not remove risk. It changes the risk. The lender now asks who the tenant is, how long the lease runs, whether the facility can be re-leased, who pays for power cost increases, whether the rent supports the debt, and whether the contract matches the life of the asset.
This is why the AI financing story cannot stop at “demand is high.” The question is whether demand becomes durable cash flow.
For the deeper collateral mechanics, read Asset-Backed Finance And AI Infrastructure.
The power layer: the bottleneck inside the bottleneck
Power is becoming the hard edge of the AI infrastructure story.
Data centers need electricity that is large, reliable, available, and often tied to specific locations. That creates demand for generation, transmission, substations, transformers, batteries, fuel cells, gas plants, renewables, nuclear discussions, and long-term power purchase agreements.
This layer changes the financing conversation. A data-center project is not only a real-estate asset if the power solution is uncertain. It is also a grid-access problem, an interconnection problem, a reliability problem, a permitting problem, and sometimes a political problem. Each of those risks changes the cost of capital.
When private infrastructure funds, utilities, energy platforms, and hyperscalers commit billions to AI-linked power projects, the signal is clear: the market no longer treats compute as weightless. It treats compute as load. Load needs power, and power needs capital.
The bank layer: construction risk and bridge financing
Banks still matter. The rise of private credit did not remove banks from infrastructure finance.
Banks can provide revolving credit facilities, construction loans, bridge financing, warehouse lines, letters of credit, project-level debt, and relationship lending. They are often important when a project is moving from plan to built asset, which is also when risk is high.
Construction can run late. Costs can rise. Equipment can be delayed. Power access can slip. Tenants can renegotiate. Capital markets can tighten before permanent financing arrives.
Bank financing can bridge that path, but bridge finance is only safe if there is a bridge to something. Permanent debt, a long-term lease, an asset sale, a refinancing, public bond issuance, or an infrastructure-fund takeout has to exist on the other side.
The public debt layer: when AI capex becomes a bond-market story
Large technology companies can finance AI capex with public debt. That is one reason the AI buildout can be so large.
The biggest platforms have access to deep bond markets and strong investor demand. A company with enormous cash flow, high credit quality, and strategic AI spending can borrow at a scale that smaller developers cannot.
But public debt changes the story. When capex is funded by borrowing, investors should ask what the debt is matched against. Is it funding capacity with contracted revenue? Is it funding speculative growth? Is it preserving balance-sheet flexibility? Is it increasing fixed obligations before the final AI revenue model is clear?
For the strongest companies, public debt may be a rational tool. For weaker players, debt can turn an exciting growth story into a maturity wall.
The REIT layer: public exposure to the physical bottleneck
Data-center REITs sit close to the physical asset. They own, lease, develop, and operate data-center real estate. They are not pure software companies and they are not private-credit funds. They are real-estate vehicles tied to the physical bottlenecks of digital infrastructure.
That gives them a cleaner public-market role than many other parts of the stack. Investors can compare development pipelines, leasing demand, occupancy, power access, capital costs, and balance-sheet leverage.
But REIT exposure is still not the whole AI financing story. A data-center REIT may own the building without owning the power project, financing the equipment, lending to the private supplier, or capturing the asset-backed finance layer.
The REIT is one window into the stack, not the whole building. For the distinction between real-estate exposure and private-credit exposure, read BDC vs REIT.
The infrastructure-fund layer: long-duration capital for long-duration assets
Private infrastructure funds are natural players in AI infrastructure because many of the assets are capital-intensive, long-lived, and tied to contracted cash flows.
Infrastructure capital can fund power generation, grid equipment, substations, fiber, cooling, land development, data-center platforms, and related operating companies. This capital usually wants durability: contracted revenue, essential-service characteristics, barriers to entry, inflation protection, or strategic scarcity.
AI infrastructure may offer some of that, but the phrase itself does not make an asset infrastructure-grade. A power project with a long-term contract and credible counterparty is one thing. A speculative buildout based on aggressive demand forecasts is another.
The label does not underwrite the asset. The cash flow does.
The asset-backed finance layer: when AI infrastructure becomes collateral
Asset-backed finance may be one of the least visible parts of the AI buildout, but it can be one of the most important.
Asset-backed finance starts with the asset, the collateral, and the cash flow. Equipment, leases, receivables, contracts, power assets, and pools of recurring payments can become financeable if the structure is strong enough.
That matters because AI infrastructure is full of expensive, contract-linked assets. Chips and servers may be financed. Leases may be financed. Receivables may be financed. Power contracts may be financed. Equipment pools and data-center capacity commitments may be financed.
This does not make the risk disappear. It changes the underwriting question. Instead of asking only whether a company is valuable, the lender asks what the collateral is, who pays the cash flow, how predictable the payment is, how fast the asset depreciates, whether it can be reused, and who takes the first loss.
That is the asset-backed lens. It may become one of the core financing tools of the AI infrastructure cycle.
For the full collateral map, read Asset-Backed Finance And AI Infrastructure: How Data Centers Become Collateral.
The private-credit layer: where clean categories break
Private credit enters where the financing need does not fit neatly into public bonds, bank loans, ordinary real-estate finance, or public equity.
That can include sponsor-backed companies serving data centers, equipment suppliers, software businesses, cooling companies, power-service providers, infrastructure contractors, fiber-related businesses, asset-heavy platforms, and companies with AI-adjacent demand but private capital structures.
Private credit may also finance parts of the stack through unitranche loans, asset-based structures, structured credit, delayed-draw facilities, or bespoke loans built around contracts and collateral.
This is why the category is useful: it is flexible. It is also why the category is risky. Flexibility can become discipline, but it can also become story-based lending.
When every borrower says it is exposed to AI infrastructure, the private-credit question is simple: where is the cash flow?
The insurance-capital layer: matching long assets with long liabilities
Insurance capital can be powerful in infrastructure finance because insurers often have long-duration liabilities. That means they may want long-duration assets.
AI infrastructure can potentially create assets with long-term contracts, predictable cash flows, and scale. Those features can attract insurance balance sheets, private-credit platforms with insurance partnerships, and asset managers that specialize in matching long-lived liabilities with long-lived investments.
This is one reason the AI financing story may expand beyond technology investors. It may become part of the institutional asset-allocation machine. Insurers do not need the story to be exciting. They need the cash flows to match.
Where BDCs fit: a public window into private lending
A BDC may lend to a private equipment supplier, contractor or other business involved in the buildout. To establish that exposure, read its schedule of investments and the borrower's business description. A technology classification alone does not identify a data-center loan.
Then examine the credit. Payment in kind, or PIK, means interest is added to a loan balance instead of paid currently in cash. Non-accruals are loans on which the lender has stopped recognizing interest under its accounting policy. Net asset value, or NAV, reflects the value of assets after liabilities. Those measures help assess earnings and asset quality; none proves that an AI-related investment will perform well.
A retail investor who buys an existing BDC share participates in the lender's results, including losses. That secondary-market purchase does not directly transfer the purchase price to a borrower. Our BDC guide explains the vehicle; how to read a BDC earnings report explains the disclosures.
Why liquidity terms matter in the AI financing boom
The AI financing story is not only about assets. It is also about vehicles.
If long-duration AI infrastructure is financed through semi-liquid private-market funds, investors need to understand the liquidity terms attached to those vehicles. A fund that owns private loans, infrastructure debt, equipment finance, or long-term contracts may not be able to return cash quickly if many investors ask to redeem at once.
That is why the private-credit liquidity cluster matters here. Lockups, redemption windows, gates, caps, proration, NAV marks, and queues are not side details. They decide whether investor liquidity expectations match the assets being financed.
The AI boom may pull more capital into private vehicles. The question is whether the liquidity promise will match the asset duration.
For the mechanics, read Private Credit Fund Terms Explained, Private Credit Redemptions Explained, and Private Credit Gating Explained.
What investors should watch
The most useful AI financing questions are not about labels. They are about structure.
Start with the customer. Who is paying for the capacity, how strong is that customer, and how long does the contract last?
Then look at the asset. Is the financing tied to a data center, a lease, a pool of receivables, equipment, a power project, a corporate borrower, or a fund vehicle?
Then look at the capital layer. Is the risk sitting in public debt, bank lending, project finance, securitization, private credit, infrastructure equity, insurance capital, a REIT, or a BDC?
Then look at timing. Does the debt mature before the asset stabilizes? Does the fund offer liquidity faster than the underlying asset can produce cash? Does the power arrive before the lease economics depend on it?
Finally, look at loss protection. Who takes the first loss if demand is real but slower than expected?
Those questions turn AI infrastructure from a theme into an underwriting exercise.
What could go wrong
The first risk is overbuilding. If capital assumes AI demand will be infinite, projects can get financed at prices that leave little room for disappointment. Too much supply, wrong locations, weak tenants, changing chip economics, or lower-than-expected utilization can turn today’s scarcity into tomorrow’s refinancing problem.
The second risk is power. A data center without reliable, affordable power is not a finished asset. Delayed interconnections, transmission constraints, rising power costs, political backlash, emissions pressure, and fuel-price volatility can all change project economics.
The third risk is contract quality. Not every lease or customer commitment is equally valuable. A long-term contract with a strong counterparty is different from a speculative commitment tied to a weak customer or uncertain demand.
The fourth risk is technology change. If chips become more efficient, model architectures change, inference patterns shift, or workloads move differently than expected, some assets may be less valuable than their financing assumed.
The fifth risk is refinancing. Infrastructure booms often look safe when capital is abundant. The test comes when maturities arrive, rates move, credit spreads widen, or investors stop accepting optimistic assumptions.
The sixth risk is private-credit drift. When lenders compete to finance a hot theme, covenants can loosen, leverage can rise, spreads can compress, and underwriting can become more story-driven. That is when a real boom becomes a credit problem.
What to look for in a financing announcement
Identify what the money will pay for and what will repay it. Then check whether the transaction is proposed, committed or completed, and whether it provides debt or equity. A development fund, a loan secured by equipment and a refinancing of existing buildings do different jobs even when all three carry an AI label.
For a BDC shareholder, the additional step is to find the holding in that company's filings. A manager's infrastructure announcement does not establish exposure for every fund or BDC it operates.
Investor Quick Answers
How does AI infrastructure get financed?
AI infrastructure is financed through hyperscaler capex, public bonds, bank credit, data-center leases, power contracts, REIT capital, infrastructure funds, asset-backed finance, private credit, insurance capital, and BDCs.
Why is AI infrastructure a financing story?
AI infrastructure requires physical assets before the final economics are fully known. Data centers need land, power, cooling, fiber, equipment, contracts, and long-term capital. That makes financing terms central to the AI buildout.
What is the AI data-center capital stack?
The AI data-center capital stack includes the layers of money used to finance the buildout: hyperscaler balance sheets, bonds, banks, leases, REITs, private infrastructure, private real estate, asset-backed finance, private credit, insurers, and BDCs.
What are examples of AI infrastructure financing?
Examples include hyperscaler capital spending, data-center securitizations, bank credit facilities, data-center leases, private infrastructure funds, AI-linked power partnerships, asset-backed financing, and private-credit loans to companies tied to the data-center buildout.
How does private credit finance AI infrastructure?
Private credit may finance AI-adjacent borrowers, equipment suppliers, infrastructure-service companies, software firms, power-related assets, asset-backed structures, leases, receivables, and other private companies tied to the data-center buildout.
What is asset-backed finance in AI infrastructure?
Asset-backed finance uses specific collateral and cash flows, such as equipment, leases, receivables, contracts, data-center assets, or power assets, to support lending. It may become more important as AI infrastructure becomes more physical and contract-based.
Why does power matter so much for AI financing?
Power is a bottleneck because data centers need large, reliable electricity supply. Grid access, generation, transmission, substations, power contracts, and interconnection timelines can determine whether a data-center project is financeable.
Are BDCs AI infrastructure investments?
BDCs are not usually direct AI infrastructure investments. They are public private-credit vehicles that may lend to private companies touched by AI infrastructure demand. Their value is as a credit signal, not as a pure AI stock substitute.
What is the biggest risk in AI infrastructure financing?
The biggest risk is that capital chases AI demand faster than cash flows, power access, contracts, collateral, and underwriting discipline can support. If assumptions miss, the pressure can show up in refinancing, NAV marks, non-accruals, asset values, or redemption terms.
Follow the AI infrastructure financing system
- Start here: this guide explains how the full capital stack fits together.
- Identify the capital providers: Who Finances AI Data Centers? separates hyperscalers, banks, REITs, infrastructure funds, private credit and BDCs.
- Find the bottleneck: Who Finances the Power Behind AI Data Centers? follows electricity, interconnection and project-finance risk.
- Test the collateral: Asset-Backed Finance and AI Infrastructure examines leases, equipment, receivables and securitization.
Read Next
For the public-company distinction, BDC vs REIT compares a lender with a property owner. For private-fund liquidity, Private Credit Fund Terms Explained explains how withdrawal terms can differ from the life of the investments.
To learn how borrowing costs affect a lender's earnings, take the Academy lesson on BDC leverage and funding. Or start with the Academy for the broader learning sequence.
Follow the financing news
BDC Weekly follows company filings, borrowing, dividends and developments in private credit. Our AI power-delay issue applies the timing questions in this guide to a reported project.
Subscribe to The Drift for the next Weekly in your inbox.
Terms used here
AI: artificial intelligence. Capex: capital expenditure, or spending on long-lived assets. Hyperscaler: a very large cloud or technology company. BDC: business development company. REIT: real estate investment trust. ABS: asset-backed securities. CMBS: commercial mortgage-backed securities. PIK: payment in kind, interest added to the loan balance rather than currently paid in cash. NAV: net asset value. Unitranche: a loan structure that combines senior and subordinated financing into a single facility. Warehouse line: interim financing used to build a pool of assets before longer-term financing. Debt service: required interest and principal payments. Securitization: financing that issues securities supported by specified assets or cash flows.
Source Notes
The three amounts in the examples are reported source figures, not calculations by The Drift. They are not summed. Goldman Sachs' $5.3 trillion figure is a June 2026 forecast; Switch's $3.5 billion figure is its rounded March 2025 refinancing announcement; Bloom's $25 billion figure describes a June 2026 financing framework, not completed deployment.
The project-life-cycle explanation draws on Norton Rose Fulbright's June 11, 2025 discussion of construction, operating and refinancing structures. It is a simplified sequence; terms and lender protections differ by transaction.
This article draws on Goldman Sachs Research on private markets’ expected role in data-center financing; International Energy Agency analysis of data-center electricity demand; S&P Global Ratings and S&P Global Market Intelligence research on data-center financing, bank credit, project finance, lease terms, and private equity investment; Moody’s data-center credit-risk commentary; Switch’s data-center securitized debt financing announcement; Reuters reporting on AI-linked power infrastructure financing; Bloom Energy and Brookfield’s AI infrastructure power-financing announcement; and The Drift’s private-credit mechanics coverage on fund terms, redemptions, gating, PIK income, NAV, non-accruals, refinancing risk, BDCs, and data-center financing.
Source links:
- Norton Rose Fulbright: Data center financing structures
- Goldman Sachs Research: Private markets are expected to have a growing role in data center financing
- IEA: Energy demand from AI
- S&P Global Ratings: Key data center financing takeaways from PPIF 2026
- S&P Global Market Intelligence: Banks meeting data center demand with billions in credit facilities and bonds
- S&P Global Market Intelligence: Private equity investment surge sends US data center deals to five-year high
- Moody’s: Data centers — managing risk amid a market boom
- Switch: Switch announces $3.5 billion in securitized debt financings
- Reuters: Bloom Energy and Brookfield expand AI infrastructure power partnership
- Bloom Energy: Brookfield and Bloom Energy expand AI infrastructure partnership to $25 billion
Disclosure
The Drift is published by Drift Research LLC for informational and educational purposes only. Nothing published by The Drift constitutes personalized investment advice, financial advice, tax advice, accounting advice, legal advice, or a recommendation to buy, sell, or hold any security. The Drift is not a registered investment adviser, broker-dealer, financial planner, or fiduciary. Data and calculations are derived from sources believed reliable and from methods described in the applicable source and calculation notes, but they may contain errors, estimates, rounding differences, or information that has become outdated. Readers should review the original sources and make their own assessment. All investments involve risk, including possible loss of principal. Past performance and hypothetical results do not guarantee future results. Consult qualified professionals before acting.
