AI Data Centre Capital Stack: Debt, Equity and Returns

A $2.1 billion AI data centre is not funded with a single cheque. It is built in layers, each carrying a different kind of money, bearing a different level of risk, and demanding a different return. Read the stack from the cheapest debt at the bottom to the most expensive equity at the top, and the headline number resolves into something concrete: how the capital is actually assembled, and who collects what once it runs.

The familiar account of this project tells you who sits where in the chain: chipmakers and hyperscalers on top, institutional capital in the middle, local developers below. This piece asks the next question, the one a financier answers before ground is broken. In what layers does the money enter, which layer gets paid first, which absorbs the first loss, and how much does each demand to hold its position.

A $2.1 billion commitment is not one homogenous block. It is a stacked structure of debt and equity, each tranche holding a different claim on cash flow and on the assets if things go wrong. How the tranches are cut, who holds which one, and how far long-term leases support the whole structure will set the project’s cost of capital, and in the end determine who actually earns a return from this compute.

A $2.1 billion commitment is sliced into tranches, not deployed at once

Infrastructure assets of this size are almost never financed from a single source. They are cut into layers, each pricing the risk it carries. At the bottom sits debt, the cheapest money, paid first and backed by collateral. At the top sits equity, the most expensive money, paid last and the first to absorb losses if the project falls short. In between usually sits a hybrid layer, part debt and part equity, filling the gap the two ends leave open.

That ordering is not an accounting formality. It is the order in which capital gets paid when cash arrives and the order in which it is wiped out when cash runs short. A dollar of equity bears more risk than a dollar of debt, so it demands a far higher return. The more of the financing that can be pushed down into cheap debt tranches, the lower the project’s blended cost of capital, and the thicker the return left for the equity holders. The whole craft of infrastructure finance lives in that sentence.

The debt layer: cheapest money, strictest conditions

Debt sits at the base because it takes the least risk and is paid first, which also makes it the cheapest money in the structure. But lenders do not underwrite AI ambition; they underwrite whether cash flow reliably covers the loan. For a data centre, that cash flow comes from long-term leases. So how much debt the project can raise, and how cheaply, depends almost entirely on who the tenant is and how long they commit to lease.

This is where the join between finance and operations shows most clearly. A twenty-year lease to a large, highly rated cloud platform turns future cash flow into something close to certain, and lenders will extend plenty of cheap debt against it. Without such an anchor, the debt tranche shrinks, its rate rises, and the gap has to be filled with expensive equity. The tenant’s credit quality, not the size of the building, is what shapes the base of the whole structure.

The equity layer: whoever takes the first loss holds control and the tail return

At the top of the structure sits equity, paid last and lost first. In exchange for that exposed position, equity holders take two things: control, and the residual return left once every debt tranche has been paid in full. If the project performs and leases renew at higher rates, that upside flows to the equity layer, not to lenders who collect only the interest they were promised.

It is here that the question of value retained onshore is decided. If the local partner only contributes land and lends its name while real equity sits offshore, then the tail return, the fattest part of the structure, flows offshore too. A deal that keeps a meaningful equity share on the Vietnamese side keeps both decision rights and the reward for the risk. The difference is not how large the total investment is, but who holds the layer at the top.

Long-term leases are what hold the whole structure upright

What lets a data centre raise plenty of cheap debt is the certainty of its cash flow, and that certainty comes from the lease. A long lease, with a creditworthy tenant, turns a capital-intensive asset into something close to a fixed income stream, exactly the appetite of long-horizon money. Without it, the same building is a bet on future demand, and has to be financed with far more expensive capital.

This is also why the party that sets the lease usually sits high in the chain. A large cloud platform committing to a long lease is not just buying space; it is handing the project the very document that lets it raise cheap capital. That power has a price: anchor tenants typically extract favourable terms, sometimes a share of the structure itself, in exchange for the commitment that underwrites the financing. How much value the Vietnamese side keeps turns on how hard that foundation is negotiated.

Obsolescence and refinancing risk lurk at every layer

A capital structure built for a twenty-year asset has to withstand an awkward fact: the hardware inside ages far faster than the building. The chips in a compute cluster can be obsolete within a few years, while the debt that financed them runs longer than that. That mismatch forces the structure to plan ahead for rounds of hardware reinvestment and for the possibility that revenue erodes if tenants migrate to newer facilities.

Refinancing is where that risk turns into cash. The original debt usually has to be rolled over at maturity, and the terms of that roll depend on interest rates at the time, on how much lease term remains, and on whether the asset is still considered modern. A project that reaches refinancing with leases near expiry and hardware already dated will see its cost of capital jump precisely when it is weakest. A sound structure schedules debt maturities and lease renewals to line up, so that no layer has to reprice from a position of weakness.

The financing structure, not the size, decides who keeps the value

Taken together, these layers show how little the $2.1 billion figure says on its own, divorced from how it is stacked. The same total can leave most of the value onshore or pull most of it offshore, depending on the ratio of debt to equity, on who holds the equity, and on which side the lease is written to favour. The headline number is the easy part; the structure underneath is what sets the outcome.

For Vietnam, the practical consequence is specific. A domestic capital market able to supply the long-term debt layer, and local investors able to hold real equity rather than just contribute land, will retain far more from each project. That capability is not built in a single deal. But every data centre financed here is an occasion for the domestic debt and equity layers to grow, and that, rather than the announced number, is what ultimately decides where Vietnam ends up in the structure.

Lotus Investment Group advises on cross-border M&A and FDI into Vietnam and Southeast Asia. If you are underwriting Vietnam exposure, get in touch.