Free tools Windows power users keep installed
One-click scans. No signup required.
iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
AI infrastructure is paid for by two groups: the large technology companies building data centers out of their own cash flow, and outside investors who lend to or finance that build-out. The second group is growing in importance. Bond and credit markets, project finance, leases, customer prepayments and utility investment are all part of the picture, and power networks add another layer, because utility contracts can push part of the cost onto data center customers. No public source gives a single, comparable percentage split across these channels. The useful answer is therefore a map of who is involved and how the risks are assigned.
How much money is involved
The most-cited recent figure comes from the International Energy Agency (IEA), in its 2026 report Key Questions on Energy and AI. It reports that capital expenditure by five large technology companies exceeded USD 400 billion in 2025 and is expected to rise by another 75% in 2026.
Two limits apply. The figure covers five companies, so it is not a measure of worldwide AI investment. The 2026 number is also a forecast. Applying the 75% to the 2025 total gives a 2026 figure above roughly USD 700 billion. That is simple arithmetic on the IEA’s numbers, not a separate IEA estimate.
Recommended Free Tools
Why company cash is no longer the whole answer
The IEA states the financing point directly: “Data centre investments have grown too large to be funded from company balance sheets alone, and large amounts of funding from capital markets will be critical for their buildout.”
#1 Best Overall
The Bank of England’s July 2026 Financial Stability Report shows the same shift from the lender side. It says AI companies are increasingly turning to external finance, particularly debt, to fund infrastructure. It relays a Barclays estimate that USD 240 billion of AI hyperscalers’ 2026 investment needs would be financed through investment-grade credit issuance. That is an attributed estimate of how much of the spending would be financed this way, not a total of bonds already sold. The Bank warns that debt servicing and opaque financing structures could create financial-stability risks.
Who provides the money
Five channels appear in the evidence. They are not ranked here, because no source gives comparable market-wide shares for them.
Rank #2
Corporate cash flow
Company cash is one channel among several. The IEA’s point is that it is no longer sufficient on its own. The sources do not state what share of 2026 spending comes from operating cash flow and what share from borrowing, so any percentage would be a guess.
Investment-grade bonds and other debt
Bonds are the most visible external channel. The Barclays figure above is the main quantified example, and it is a 2026 estimate for one credit channel. Debt also brings servicing obligations that continue regardless of how demand develops. The Bank of England’s concern is with servicing those obligations and with financing structures that are hard to see from outside.
Leases, joint ventures and project finance
A 2026 NBER working paper lists the channels it proposes to examine: leases, joint ventures, project debt, private credit, securitization and special-purpose vehicles (SPVs). These are tools for dividing a build-out among several parties. A lease lets a company use equipment or space without owning it. A joint venture shares ownership of a facility between partners. Project debt is borrowing secured against a single project’s expected cash flows. An SPV is a separate legal entity that holds an asset and its debt, often keeping both apart from the parent company’s balance sheet. The paper’s buildout estimates are model-based projections, and its list is an analytical framework rather than a ranking of how much each channel contributes.
GPU financing and customer prepayments
Infrastructure providers show a different mix. IREN’s FY2026 filing with the U.S. Securities and Exchange Commission says its customers include hyperscalers, frontier labs, AI developers and enterprises. It describes GPU financing and customer prepayments as sources supporting deployment. Prepayments matter for the question of who pays because they move part of the funding burden onto the customer before the service is delivered. This is one provider’s structure, and it should not be read as the template for the market.
Who pays for the power
Compute hardware is only part of the bill. The IEA’s 2025 analysis lists servers, networking, cooling, uninterruptible power supply (UPS) batteries, backup generators and grid connections as parts of data center infrastructure. It also notes that energy-system infrastructure has longer lead times than data center construction. That mismatch in timing is what brings utilities into the financing question.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
The IEA reports that electricity demand from data centers grew 17% in 2025, and demand at AI-focused data centers grew 50%. Its updated projection puts data center electricity use at 485 TWh in 2025 and 950 TWh in 2030. The 2025 figure is an estimate and the 2030 figure is a projection. Neither is a financing total.
Best Value
What utility contracts say
AEP (American Electric Power) describes, in its 2026 investor presentation, long-term agreements with data center customers. These include minimum monthly charges and customer credit and collateral protections. The company’s plans also include substantial storage and generation investment to accommodate expected loads. Those terms show how a contract can allocate risk:
- Minimum monthly charges set a floor payment regardless of how much power the customer uses.
- Credit requirements test whether the customer can meet its payment obligations.
- Collateral secures the obligation if the customer fails to pay.
The contracts do not show who bears every system-wide cost. Generation, storage and grid connection can be assigned to a data center under a contract, or their costs can be spread across other utility customers. The AEP example shows one approach. It does not show how other utilities or regulators divide these costs.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Where the risk sits
Each financing choice moves exposure to a different party. The table sets out the arrangements above by the downside each one carries and what remains open.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchPC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11| Arrangement | Downside the evidence identifies | Who is exposed, as far as the sources show | What is not established |
|---|---|---|---|
| Company cash flow | Spending that does not earn expected returns | The spending company and its shareholders | Share of spending funded from cash versus borrowing |
| Investment-grade bonds and other debt | Debt servicing burden; opaque structures; financial-stability risk (Bank of England, July 2026) | Borrowers and bondholders | Share of 2026 spending financed by bonds; the USD 240 billion figure is a Barclays estimate |
| Leases, joint ventures, project debt, private credit, securitization and SPVs | Utilization, refinancing and residual-value risk | Depends on how each deal is structured | Relative size of each channel; NBER projections are model-based |
| GPU financing and customer prepayments | Customer commitment and asset utilization | Prepaying customers and the financing provider | Prepayment size, refund rights and other terms not stated in IREN’s filing description; one issuer only |
| Utility contracts with minimum charges and collateral | Exposure if expected load does not arrive or payments fall short | The data center customer under the contract; wider ratepayers depend on each utility’s rules | How wider grid costs are shared across utility customers; AEP is one example |
Will AI customers ultimately pay?
Customers are already paying through contracts. Prepayments, minimum charges and collateral all place money or commitments on the customer side. Whether that will carry the full cost of the build-out depends on demand and returns. The IEA says so directly: “The pace of data centre growth, and the resulting increase in energy consumption, will be sensitive to market sentiment, including expectations for returns on investment in data centres and AI deployment, as well as to broader macroeconomic and financing conditions.”
That dependency cuts both ways. The evidence does not prove that current spending will earn adequate returns, and it does not prove that the spending will fail. It identifies what to watch: expected returns, the pace of demand growth, and the cost and availability of financing.
Quick Recap
A checklist for reading any AI financing announcement
- Who owns the data center, compute equipment or power asset, and who holds the debt attached to it?
- Does the customer prepay, commit to minimum charges, or post collateral?
- If utilization falls, who absorbs the shortfall and who refinances the debt?
- Who pays for generation, storage and grid connection, and is that written into the contract?
- Is the headline figure a company’s disclosed spending, a bank’s estimate, or a model projection?
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
