AI Data Center Financing: Who Is Funding the $3 Trillion Buildout?

AI data center financing is moving away from hyperscaler cash flows and toward a layered stack of private funds, special purpose vehicles (SPVs) and sovereign capital. Every AI query runs in a physical building that needs land, power, cooling, chips and, above all, capital. This guide explains who is writing the cheques for the global data center investment boom, how the deals are structured, where India fits in, and what could go wrong.

Key takeaways

• The bill is about $3 trillion. Investment banks estimate global data center investment at roughly $2.9–3.0 trillion by 2029, and up to $11.3 trillion cumulatively by 2032 in faster-growth scenarios.

• Hyperscalers cannot fund it alone. They cover roughly half of projected 2029 spending, leaving a gap of about $1.4–1.5 trillion for other investors.

• A new financing stack fills the gap. Private infrastructure funds, SPVs, sovereign wealth funds and debt markets (ABS and CMBS) now supply most of the rest.

• India is already in the frame. Digital Connexion, a Brookfield, Reliance Jio and Digital Realty venture, shows how AI data center investment in India is being financed.

How big is the AI data center investment gap?

Global data center investment is heading for a figure without precedent in digital infrastructure: roughly $2.9–3.0 trillion by 2029. More aggressive scenarios, which assume faster growth in AI compute, reach $11.3 trillion cumulatively by 2032.

For most of the cloud era, the answer to "who pays" was simple. Microsoft, Amazon, Google and Meta built and owned their data centers and paid for them from operating cash flow. Hyperscalers spent about $290 billion on data center infrastructure in 2024, and nearly $200 billion of that came from those four companies.

That model is now strained. The cost of the generative AI race is rising faster than even the deepest balance sheets can comfortably absorb, which is why AI data center financing has become a question for investors, lenders and policymakers, not only for technology companies.

The AI data center financing stack at a glance

Four layers now share the cost of data center investment, with hyperscalers still deciding where capacity goes and who the anchor tenant is. Microsoft, Amazon, Google and Meta sit at the top as demand and anchor tenants: their long-term hyperscale leases set location and capacity, and they still cover roughly half of 2029 spending.

Below them, private infrastructure funds and pension investors supply the equity that owns and develops the campuses, as in Brookfield's $10 billion AI infrastructure fund. Banks then provide construction debt, and ABS and CMBS investors refinance stabilised assets, as in the $27.3 billion SPV that Meta and Blue Owl created. For the very largest projects, state-backed funds such as GAIIP and GIC share the risk as sovereign capital.

Hyperscalers: still the anchor tenants, no longer the sole funders

Hyperscalers still cover roughly half of projected 2029 data center investment, which leaves a gap of about $1.4–1.5 trillion for others to fill. The giants are shifting from a "build and own" model toward leasing, joint ventures and external debt.

They remain the anchor. They decide where capacity goes and who the tenant is, but increasingly someone else supplies the money. That makes tenant credit quality the foundation of almost every AI data center financing structure that follows.

Private infrastructure funds: the new lenders and owners of data centers

Much of the gap is being filled by private capital. After the 2008 financial crisis, tighter regulation pushed banks away from complex corporate lending, and alternative asset managers such as Blackstone, Apollo and Ares moved in. By late 2025, private infrastructure funds managed over $1.7 trillion in assets and held about $400 billion in uninvested capital.

data center investment suits these investors. Hyperscaler leases are long, tenants have strong credit and demand is growing. Stabilised assets trade at capitalisation rates of around 4.8–5.3%, while development equity targets much higher returns.

The deals are large:

• 2023: Brookfield and Ontario Teachers' Pension Plan agreed to acquire Compass Datacenters, a builder of hyperscale campuses, for $5.5 billion.

• Late 2025: Brookfield launched a $10 billion equity fund for AI infrastructure. Initial commitments of $5 billion came from Brookfield, Nvidia and the Kuwait Investment Authority, and the fund could deploy up to $100 billion once debt is added.

How SPVs finance data centers off the balance sheet

The most inventive layer of AI data center financing is the special purpose vehicle (SPV). A hyperscaler wants a campus built without loading its own balance sheet and credit rating, so a separate entity borrows against the lease instead.

Meta's "Project Beignet" in Louisiana is the best-known example. Rather than fund the roughly $30 billion campus itself, Meta partnered with Blue Owl Capital to create an SPV. The SPV raised $27.3 billion in external debt, secured by Meta's commitment to lease the facility. Meta gets the capacity, lenders get contracted cash flow from a strong tenant, and the debt sits outside Meta's consolidated accounts.

The structure is now common:

1. An SPV or joint venture sits at the centre of the project.

2. Equity comes from infrastructure funds and pension investors.

3. Construction debt comes from banks.

4. Once the campus is stabilised, the debt is often refinanced through asset-backed (ABS) or commercial mortgage-backed securities (CMBS).

Fitch Ratings reports that data center ABS issuance has roughly doubled in two years, and CMBS issuance has tripled.

Sovereign wealth funds and AI data center investment

Some projects are now too big even for the largest private pools, so risk is increasingly shared with state-backed investors.

• GAIIP: The Global AI Infrastructure Investment Partnership brings together Microsoft, BlackRock, Global Infrastructure Partners and MGX, the UAE-backed investment firm. It aims to raise an initial $30 billion in equity and mobilise up to $100 billion in total capital for data centers and the energy infrastructure that powers them.

• GIC: Singapore's sovereign fund has put capital into a roughly $15 billion US joint venture with CPP Investments and Equinix. In 2025 it also formed a BRL 2 billion (about $329 million) data center venture in Brazil with local partner Alianza.

For sovereign funds and pensions, the appeal is long-dated, inflation-linked income and exposure to the AI buildout.

AI data center investment in India: the Digital Connexion model

This financing story is no longer only American. AI data center investment in India is already being structured the same way. Brookfield, Reliance Jio and Digital Realty each hold a one-third stake in Digital Connexion, a joint venture developing campuses of about 40 MW in Mumbai and 20 MW in Chennai.

The venture combines a global infrastructure fund, a listed data center REIT and a domestic partner with market access and customer relationships. That combination is likely to be repeated across India, the Gulf and Southeast Asia, where demand for local capacity is rising and where regulation, power availability and data localisation shape where capital can go.

For anyone planning data center investment in India, the lesson is that capital follows secured power, credible tenants and a partner who understands local rules.

Key risks in AI data center financing

The scale of capital should not hide four risks.

Completion risk

Power connections, permits, labour shortages and supply-chain delays can stall projects. Lenders now insist on fixed-price construction contracts and secured grid interconnection before committing.

Tenant concentration

Many structures depend on a handful of hyperscalers. If a tenant's demand falls short or its lease terms change, the debt supported by that lease comes under pressure.

Off-balance-sheet opacity

SPVs keep debt out of a corporate's headline leverage, but the obligations are real. Investors and rating agencies are paying closer attention to lease commitments.

Policy and energy regulation

The EU is moving toward a binding energy-efficiency rating scheme for large data centers. Some US localities have debated moratoria over concerns about power and water. Projects now need community and regulatory support, not just capital.

There is also a more basic question. Capital is flowing in on the expectation that AI demand will justify the buildout. If returns on AI spending disappoint, the first signs would appear in lease renewals, rental growth and the pricing of securitised debt.

What this means for investors, operators and policymakers

Investors: The opportunity in data center investment is moving down the stack. The headline returns are no longer in owning the chips or the software but in the power, land, financing and contracted cash flows that make the buildings possible. Diligence should focus on tenant quality, lease length, power security and exit routes.

Operators and developers: Capital is available but conditional. Funders back projects with credit-worthy tenants, secured power and fixed-price contracts. A site without grid access is hard to finance, however good its location.

Policymakers: Financing is now an infrastructure policy question. Clear permitting, reliable power and predictable data rules do as much to attract capital as tax incentives.

The $3 trillion question has an answer, but not a single source. Hyperscalers supply the demand, private funds and sovereign investors supply equity, and debt markets supply most of the rest. Whether that system holds up depends less on how much capital is available than on whether the power, the tenants and the returns arrive as expected.

FAQs: AI data center financing

Who is funding AI data centers?

Hyperscalers still cover roughly half of projected spending. The rest comes from private infrastructure funds, SPVs and joint ventures, sovereign wealth funds and debt markets.

How much will AI data center investment reach by 2029?

Investment banks estimate roughly $2.9–3.0 trillion by 2029, with more aggressive scenarios reaching $11.3 trillion cumulatively by 2032.

What is an SPV in data center financing?

A special purpose vehicle owns and borrows against a campus, secured by a hyperscaler's lease, so the debt sits outside the tenant's consolidated accounts. Meta's Project Beignet is the best-known example.

Is AI data center investment in India growing?

Yes. Digital Connexion, a joint venture of Brookfield, Reliance Jio and Digital Realty, is developing campuses of about 40 MW in Mumbai and 20 MW in Chennai.

What are the main risks in AI data center financing?

Completion delays, tenant concentration, opaque off-balance-sheet obligations and tightening energy regulation. If AI returns disappoint, lease renewals and securitised debt pricing would show it first.

About the author

Shivam is Vice President, Strategy at Xvello (Intelligence · Strategy · Growth), where he advises on infrastructure and growth strategy. [Add before publishing: years of experience, relevant deals or advisory work, LinkedIn link and a short bio page URL. These are the strongest experience and authority signals for this page.]

Sources and references

Link each source below to its original page before publishing, and add "Last reviewed: October 2026" near the byline.

• Investment bank research on global data center spending (2029 and 2032 estimates)

• Fitch Ratings: data center ABS and CMBS issuance trends

• Brookfield and Ontario Teachers' Pension Plan: Compass Datacenters acquisition (2023)

• Brookfield: $10 billion AI infrastructure fund announcement (late 2025)

• Meta and Blue Owl Capital: Project Beignet financing, Louisiana

• GAIIP: Microsoft, BlackRock, Global Infrastructure Partners and MGX announcement

• GIC: CPP Investments and Equinix joint venture; Alianza Brazil venture (2025)

• Digital Connexion: Brookfield, Reliance Jio and Digital Realty