Who Pays for the AI Boom? Inside the New Infrastructure Financing Stack

AI infrastructure is becoming too large for one balance sheet. A new financing system is emerging around it.

The AI race started with models.

Then it moved to chips.

Now it is moving into capital markets.

Building AI at scale requires more than ordering processors and finding an empty warehouse. It requires data centers, power generation, substations, transmission lines, cooling equipment, fiber networks, construction crews and years of development work.

Those systems must be paid for long before the first customer sends a prompt.

That creates a simple but increasingly important question:

Who pays for all of it?

Hyperscalers still provide much of the initial capital. But as projects become larger, more expensive and more complicated, the financing is beginning to spread across corporate balance sheets, bond markets, private-credit funds, infrastructure investors, utilities, real-estate partners and equipment-financing companies.

AI infrastructure is no longer being funded through one check.

It is being assembled through a capital stack.

The Short Answer

AI infrastructure is financed through a combination of corporate capital expenditures, debt, equity, joint ventures, private credit, project finance, leases, utility investment and long-term customer contracts.

Different sources of capital fund different parts of the system.

Hyperscalers may purchase processors and other strategic equipment directly. Infrastructure funds may own the buildings and long-lived utility systems. Banks and private lenders may finance construction. Utilities may fund power infrastructure through their own capital programs. Developers may contribute land, permits and project expertise.

The final structure depends on who owns the asset, who uses it, how long it will remain valuable and who is willing to carry the risk.

The AI boom may begin with technology.

But it gets built with finance.

What Is AI Infrastructure Financing?

AI infrastructure financing is the process of raising and structuring capital for the physical systems required to train, operate and deliver artificial intelligence.

That Includes Financing For:

• AI data centers and cloud campuses

• Processors, servers and storage systems

• Power generation and energy contracts

• Substations and transmission infrastructure

• Fiber routes and network equipment

• Cooling and water systems

• Land acquisition and site development

• Construction and skilled labor

• Backup power and resilience systems

This looks very different from financing a software company.

Software can often be developed by a relatively small team and distributed at low incremental cost. AI infrastructure requires physical assets that can take years to permit, construct and connect.

The equipment inside a facility may become outdated quickly. The building around it may operate for decades.

That mismatch is one reason the financing stack is becoming more sophisticated.

You do not necessarily want to finance a three-year processor and a 30-year substation with the same money.

The Signal

The spending plans tell us how quickly the AI capital cycle is expanding.

Amazon said it expects approximately $200 billion in companywide capital expenditures during 2026, with AI, chips, robotics and satellite infrastructure among the major opportunities driving investment. The company also reported that higher property-and-equipment spending—primarily reflecting AI investment—had significantly reduced free cash flow. Amazon

Microsoft reported $41 billion in capital expenditures during its fiscal fourth quarter of 2026. Roughly two-thirds went toward shorter-lived assets, primarily CPUs and GPUs. The remainder funded longer-lived infrastructure, while another $5.6 billion of finance leases primarily supported large data-center sites. Microsoft

These figures are not directly comparable, and not every dollar is exclusively for AI.

But the direction is difficult to miss.

AI infrastructure spending is reaching a scale where even the world’s largest technology companies must think carefully about which assets they want to own, which risks they want to retain and which projects should be financed with outside capital.

A large balance sheet can launch the buildout.

It does not have to own every piece of it.

Number Worth Knowing: $45 Billion to $50 Billion

In February 2026, Oracle announced plans to raise between $45 billion and $50 billion during the calendar year to expand Oracle Cloud Infrastructure.

The company said the additional capacity was intended to meet contracted demand from major customers including AMD, Meta, Nvidia, OpenAI, TikTok and xAI.

Oracle planned to raise approximately half through equity and equity-linked securities, with the other half coming from investment-grade bonds. The equity plan included an at-the-market program of up to $20 billion. Oracle

That is not a traditional cloud expansion.

It is an industrial-scale financing program.

Oracle’s plan illustrates the new reality of the Intelligence Economy: customer demand can be enormous, contracts can be signed and the technology can be ready—but the capacity still has to be financed.

The infrastructure does not appear when the contract is signed.

Someone must raise the money, purchase the equipment and build it.

Why Are Hyperscalers Sharing the Bill?

Companies such as Amazon, Microsoft, Google, Meta and Oracle have access to capital that most businesses can only dream about.

So why bring in outside investors?

Because capital has an opportunity cost.

Every dollar committed to land, buildings and utility infrastructure is a dollar that cannot be used for processors, research, acquisitions, software development or shareholder returns.

Outside Financing Can Help Hyperscalers:

• Develop more projects at the same time

• Preserve cash for strategic technology investments

• Match long-lived assets with long-term capital

• Transfer portions of construction and ownership risk

• Maintain flexibility as technology changes

• Reduce the amount of real estate held directly

• Expand without placing the entire project on one balance sheet

This is not necessarily a sign that a company cannot afford the project.

It is often a sign that the company is becoming more deliberate about what it wants to own.

The processor may be strategically critical.

The concrete shell around it may be financeable by someone else.

What Does the AI Infrastructure Financing Stack Look Like?

There is no single formula. But most large AI projects draw from several of the following capital sources.

1. Corporate Capital Expenditures

Corporate capex remains the foundation of the AI buildout.

Hyperscalers use their own cash to purchase processors, servers, networking equipment and other assets they consider strategically important. They may also directly fund land, construction and power infrastructure when speed or control matters more than balance-sheet efficiency.

This is the simplest structure.

The company pays for the asset and owns it.

But as projects grow from hundreds of millions to tens of billions of dollars, direct ownership becomes only one option.

2. Public Debt and Equity

Large companies can raise money through bonds, common shares, preferred securities and convertible instruments.

Debt allows a company to spread the cost of infrastructure over time. Equity provides capital without creating the same fixed repayment obligation, although it dilutes existing shareholders.

Oracle’s 2026 financing plan used both.

The objective was not merely to raise the maximum amount of money. It was to balance expansion with the company’s investment-grade credit profile.

That balance will become increasingly important across the sector.

AI companies need capital.

They also need investors and credit markets to believe that tomorrow’s revenue will justify today’s buildout.

3. Infrastructure Joint Ventures

A joint venture allows a technology company to share ownership with an outside capital partner.

This structure can be especially useful for buildings, power systems and connectivity assets that have long operating lives but are not necessarily the hyperscaler’s primary competitive advantage.

Meta’s El Paso campus provides a clear example.

In July 2026, Meta and BlackRock announced a venture to develop and own a one-gigawatt data-center campus. Funds managed by BlackRock agreed to hold an 80% interest, while Meta retained 20%.

The venture expects approximately $14 billion in total development costs for the buildings and long-lived power, cooling and connectivity infrastructure. BlackRock’s investment is partially supported by $12.5 billion in debt financing, while Meta will lease the campus and serve as its initial sole occupant. The lease structure gives Meta extension options covering a potential 20-year period. Meta

That Is the New Financing Stack In One Project:

• A hyperscaler as operator and tenant

• An infrastructure investor as majority owner

• Debt financing supporting the venture

• Long-term leases creating predictable cash flow

• Guarantees helping protect the asset’s residual value

The project may look like a data center from the outside.

Financially, it looks more like a privately financed utility-scale infrastructure asset.

4. Private Credit

Private credit refers to loans made by investment funds and other non-bank lenders.

These lenders can provide customized financing for projects that may be too large, complicated or time-sensitive for a traditional bank structure. They can finance construction, equipment purchases, acquisitions or entire portfolios of data-center assets.

Private credit is attractive because it can move quickly and accommodate unusual deal terms.

It is also expensive capital.

The lender expects to be compensated for construction risk, concentration risk, technology uncertainty and the possibility that a project takes longer than expected to become operational.

As AI projects grow, private-credit firms are moving closer to the center of the infrastructure market.

They are no longer simply lending against buildings.

They are financing the systems inside them.

5. Project Finance

Traditional corporate debt is supported by the overall company.

Project finance is different. Repayment depends primarily on the cash flow generated by a specific project.

For an AI campus, that cash flow may come from a long-term data-center lease, a capacity agreement with a cloud provider or another contracted customer commitment.

The stronger the contract, the easier the project may be to finance.

Lenders Will Examine:

• The credit quality of the tenant

• The length of the lease

• Power availability and interconnection status

• Construction costs and completion guarantees

• Equipment-delivery schedules

• Permits and environmental approvals

• Insurance and operating risks

• Expected value of the asset if the original tenant leaves

A data-center rendering does not make a project bankable.

Contracts do.

6. Leases and Sale-Leasebacks

A lease allows a company to use an asset without owning all of it directly.

In a sale-leaseback, a company sells an asset to an investor and leases it back for continued use. This can release capital that was previously tied up in land or buildings.

For hyperscalers, leasing may provide access to capacity while preserving money for processors and product development. For infrastructure investors, the lease can create a long-term income stream supported by a highly rated corporate tenant.

But a lease is not free money.

It replaces an upfront ownership cost with a long-term financial commitment. The value depends on lease terms, accounting treatment, renewal options and who remains responsible for operating and upgrading the asset.

The building may leave the balance sheet.

The obligation does not disappear.

7. Utility and Energy Financing

A data center without power is an expensive storage facility.

Utilities may finance substations, transmission upgrades and new generating capacity through their capital programs. Developers may contribute directly to interconnection costs. Energy companies may build generation supported by long-term power-purchase agreements.

Other projects may use behind-the-meter generation, energy-service agreements or dedicated power arrangements.

The contract is once again the key.

A long-term commitment from a creditworthy data-center customer can help an energy developer finance a power plant. Without that commitment, the same project may be considered too risky.

AI demand is therefore changing more than electricity consumption.

It is changing how new energy infrastructure becomes financeable.

8. Equipment Financing

Not every asset inside a data center has the same useful life.

Buildings, transmission lines and cooling systems may last for decades. Processors can become economically outdated much faster.

Equipment leases, vendor financing and asset-backed loans can help match financing terms with the expected life of the hardware.

This reduces the risk of repaying a ten-year loan on equipment that became obsolete after four.

It also creates a new question for lenders:

What is a used AI processor worth?

The answer matters because financing depends partly on what the lender can recover if the borrower fails.

The faster technology changes, the more carefully capital providers must think about residual value.

Why Do Contracts Matter More Than Concrete?

A proposed AI campus contains uncertainty.

A financed AI campus contains contracts.

That transformation happens when land is controlled, permits are secured, power is committed, construction costs are defined and a credible tenant agrees to use the capacity.

Each completed step reduces risk.

Lower risk can attract more lenders, improve financing terms and reduce the project’s cost of capital. Lower financing costs can then make the development more competitive.

This is why infrastructure readiness and financial readiness are closely connected.

A community may have available land.

A developer may have an ambitious plan.

An investor may have capital.

But the project becomes financeable only when the major risks have been assigned to parties capable of carrying them.

Capital Has a Geography

Money does not move toward every proposed data-center market equally.

It moves toward places where development risk can be understood.

Capital Is More Likely to Support Projects With:

• Available or contractually secured power

• Multiple high-capacity fiber routes

• Developable land with clear ownership

• Reliable water and cooling strategies

• Predictable permitting timelines

• Construction and technical workers

• Community and political support

• Customers willing to sign long-term commitments

Cheap land alone is not enough.

Neither is a line on a fiber map or a utility announcement without a delivery date.

The most financeable markets are the ones that can turn infrastructure claims into verifiable commitments.

That gives state and local leaders a direct role in the capital stack.

Clear policies, coordinated permitting and credible infrastructure planning reduce uncertainty. Reduced uncertainty can lower the cost of capital.

Company Watch: BlackRock

Everyone sees a data center.

BlackRock sees a long-duration infrastructure asset with a creditworthy tenant.

Its Meta partnership shows how the largest investment managers are moving beyond owning shares in technology companies. They are helping finance and own the physical systems those companies need.

BlackRock brings more than capital to the El Paso venture. Through Global Infrastructure Partners and HPS Investment Partners, it combines infrastructure ownership, private financing and large-scale investment management inside one platform.

That combination may become increasingly valuable.

The AI buildout crosses several traditional investment categories at once: real estate, energy, digital infrastructure, private credit and industrial development.

Firms able to assemble those pieces may become as important to AI expansion as the companies building the models.

BlackRock will not be alone.

Infrastructure managers, pension funds, sovereign-wealth funds, insurance companies and private-credit firms are all natural candidates to own long-lived assets supported by contracted technology demand.

The hyperscaler may use the campus.

The capital provider may own it.

Capital Across the Eight Pillars

Capital is one of the eight pillars of the Infrastructure of Intelligence™.

But it does not operate beside the other seven.

It moves through all of them.

Capital pays for new power generation and transmission. It funds fiber construction and network equipment. It acquires land, develops water systems and trains workers. It responds to policy, permitting and sustainability requirements.

A weakness in any pillar can increase financing costs.

A project with uncertain power may require more expensive capital. A long permitting process can increase interest expense before the facility produces revenue. Limited workforce availability can delay construction. Water risk can affect insurance, operating costs and community support.

Capital is how infrastructure plans become physical assets.

It is also how infrastructure weaknesses become financial penalties.

The Risk Has Not Disappeared

New financing structures distribute risk.

They do not eliminate it.

Investors Still Have to Price:

• Construction delays

• Power and interconnection uncertainty

• Equipment shortages

• Rising labor and material costs

• Interest-rate and refinancing risk

• Dependence on a small number of tenants

• Processor obsolescence

• Regulatory and environmental changes

• Community opposition

• Water availability

• The possibility that projected AI demand does not arrive on schedule

There is also a mismatch inside many AI projects.

The building may remain valuable for decades. The processors inside it may not. The power infrastructure may have alternative uses. A highly specialized cooling system may not.

The strongest financing structures will separate those risks and assign each asset to the type of capital best equipped to own it.

Long-term capital should own long-term assets.

Short-lived equipment should not be treated like a power plant.

The Hot Take

The next AI winners may not be the companies with the largest balance sheets.

They may be the companies that use those balance sheets most intelligently.

AI infrastructure has become too large to be financed as a collection of ordinary corporate technology purchases. It increasingly requires the capital engineering used for power plants, airports, pipelines and other major infrastructure systems.

That shift changes the competitive landscape.

A company able to secure processors but unable to finance the campus around them may fall behind. Another company with strong capital partners may develop more capacity without carrying every dollar of the project itself.

The AI race is becoming a financing race.

What Should Leaders Start Asking?

Companies, Communities and Investors Evaluating AI Infrastructure Should Ask:

• Who will own the land, building and equipment?

• Which assets will be funded directly by the hyperscaler?

• Where will debt, equity or private credit be used?

• Who carries construction and completion risk?

• Is the power supply committed or merely proposed?

• What customer contracts support the financing?

• How long are the leases and capacity agreements?

• Does the financing term match the useful life of the asset?

• What happens if processors become obsolete faster than expected?

• Can the facility support another tenant if the original customer leaves?

• Who pays for transmission and interconnection upgrades?

• What guarantees are being provided?

• How sensitive is the project to higher interest rates or delays?

• Which public incentives are involved?

• What economic benefits remain in the host community?

These questions reveal something a project announcement rarely does:

Whether the infrastructure is truly ready to be built.

The Golden Nugget

Capital does not fund ambition. It funds de-risked cash flow.

The campus becomes financeable when uncertainty becomes contracts. Contracts for land, power, construction, equipment and customers.

That is the difference between an announced project and a bankable one.

Frequently Asked Questions

How Are AI Data Centers Financed?

AI data centers are commonly financed through corporate capital expenditures, debt, equity, joint ventures, private credit, project finance and long-term leases. Large projects often combine several of these sources.

Why Are Private-Credit Firms Financing Data Centers?

Private-credit firms can provide large, customized loans for complex projects. They may also move faster and accept structures that do not fit traditional bank-lending models, although their capital generally carries a higher cost.

What Is a Data-Center Joint Venture?

A data-center joint venture is a partnership in which a technology company, developer or operator shares ownership with an infrastructure investor. The technology company may operate or lease the facility while the financial partner supplies a significant portion of the capital.

What Makes an AI Infrastructure Project Bankable?

A project becomes more bankable when it has controlled land, permits, committed power, credible construction plans, equipment access and long-term contracts with financially strong customers.

Why Does Financing Affect Where AI Infrastructure Gets Built?

Financing costs reflect project risk. Markets with available power, strong connectivity, predictable policy and credible development timelines can be easier and less expensive to finance.

Who Ultimately Pays For the AI Infrastructure Boom?

The initial capital comes from technology companies, investors, lenders, utilities and public markets. Over time, much of the cost is recovered through cloud fees, AI subscriptions, leases, energy payments and the prices businesses and consumers pay for AI-powered services.

The IOI Take

Wall Street often describes the AI boom as a capital-expenditure cycle.

That description is correct, but incomplete.

This is the creation of a new infrastructure-financing system.

Hyperscalers are becoming anchor tenants. Infrastructure funds are becoming owners. Private-credit firms are becoming construction lenders. Utilities are financing power expansion. Long-term customer contracts are being transformed into the cash flows supporting debt and equity investment.

Capital is connecting every pillar of the Intelligence Economy.

For companies, that means financing strategy may determine how quickly capacity comes online.

For investors, it means the opportunity extends far beyond technology stocks.

For communities, it means AI readiness must include financeability. A location that can reduce development uncertainty may attract capital before a competing market with better headlines but weaker execution.

The regions that learn how infrastructure capital makes decisions will be better positioned to capture the projects it funds.

One Last Thought

The company with the best model can create demand.

The company with the most processors can create capacity.

But the project with the strongest financing is the one most likely to get built.

The Intelligence Economy will be shaped not only by who can imagine the infrastructure.

It will be shaped by who can make it bankable.

Workforce

Policy

Capital

Sustainability