AI infrastructure is increasingly being financed as an investable asset class, with a layered capital stack that has expanded rapidly in recent years.
Building AI compute at scale is no longer an equity story. It is a stack of equity, debt, asset-backed financing, and customer commitments, each funding a different layer of the build.
The amount of money going into AI infrastructure is hard to hold in your head. McKinsey estimates that roughly $5.2 trillion in capital expenditures will be required for AI-related data center capacity by 2030. No single funding source can carry a number like that, and no company should try. Funding all of that infrastructure with equity would dilute ownership, while financing it entirely with conventional corporate debt could put significant pressure on a company’s balance sheet.
So the market began doing what mature infrastructure markets have done in other sectors: building a capital stack where different layers of an asset are funded by different types of capital. Understanding that stack is now part of understanding AI infrastructure itself, whether you are building it or funding it.
Why One Funding Source Can’t Build AI Infrastructure Anymore
The core problem is that AI infrastructure mixes very different kinds of risk in one project. The company building it is a growth bet. The hardware is a depreciating asset. The facility is a long-lived piece of real estate. The customer revenue is a contract. Each of those has a natural type of capital that can price it appropriately, and forcing all of them into one instrument can mean overpaying for some and underfunding others. The capital stack exists to match each layer to the type of capital best suited to it.
Equity Funds the Company, Not the Compute
Equity sits at the bottom of the stack and does what it does best: funding the business and absorbing the first losses. It pays for the team, the software, the customer relationships, and the risk that the company itself succeeds or fails.
What the emerging capital stack makes possible is using equity for the parts of the business that require risk-bearing capital, while using debt and asset-backed financing for financeable infrastructure such as GPUs. The discipline of the new stack is using equity for what only equity can fund, and financing the rest.
Debt and Asset-Backed Financing Fund the Hardware
The hardware layer is where debt and asset-backed financing come in, because GPUs can serve as collateral in a way that a growth story cannot. This is the layer that has grown rapidly.
Over the past two years, multi-billion-dollar financings backed by GPU infrastructure and associated customer cash flows have closed, including transactions large enough to receive investment-grade credit ratings. These structures allow companies to finance large amounts of compute while aligning the cost of capital with the underlying assets and contracted cash flows.
The financing spreads the cost of the hardware over its useful economic life rather than requiring the company to fund the full purchase upfront.
Customer Commitments Fund the Capacity Before It Exists
A newer and increasingly important layer of the stack is the customer commitment, where a signed contract to buy compute can help anchor the financing for the capacity being built.
When a creditworthy customer agrees to pay for capacity over several years, that contract can strengthen the credit case for the debt used to build the capacity in the first place. This is why many of the largest GPU financings are backed by both the hardware and contracted customer cash flows: the commitment provides lenders with a clearer source of repayment alongside the underlying collateral.
In effect, the customer’s promise to pay helps support the financing of the infrastructure that serves them.
Infrastructure Financing Funds the Facility
The facility itself, the building, the power, and the cooling, is funded more like traditional infrastructure and real estate, on longer horizons that match the life of the asset.
A data center is a long-lived asset, so it can carry longer-dated financing that would not necessarily fit the shorter economic life or financing horizon of GPU hardware. Separating the facility layer from the hardware layer lets each be funded on its own timeline, which is one of the quiet reasons the stack works.
The building is financed over a longer horizon, the chips over a shorter one, and neither is forced onto the other’s schedule.
Why the Stack Is Forming Now
This structure is emerging now because the scale has grown dramatically and recent transactions have demonstrated that the model can work at institutional scale.
Recent deals have shown that GPU-backed financing can support much larger deployments and, in some cases, achieve investment-grade ratings. Each transaction adds more precedent for lenders and investors evaluating AI infrastructure as a financeable asset class.
The market is doing for AI infrastructure what it once did for real estate, telecom, and energy: turning a new kind of asset into something the full range of capital markets can fund.
What It Means for Builders and Investors
For a company building compute, the lesson is that how you fund the build is now a strategic decision, not a formality. Matching each layer to the right capital, equity for the business, debt for the hardware, customer contracts to support the financing, and longer-term capital for the facility, can influence how quickly you scale and how much equity you need to deploy.
For an investor, the same structure creates a menu of ways to participate, each with its own risk and return. And for both, coordination is the hard part, because the layers have to fit together to work.
The capital stack is only an advantage when the pieces are assembled to fit.
