Who’s Really Paying for Meta’s 1 GW AI Compute? Unpacking the Capital Structure

01 – Equity vs. Cost

On July 28 Meta and BlackRock announced a joint venture to develop a 1 GW data‑center park in El Paso, Texas. BlackRock‑managed funds hold 80% of the equity while Meta retains 20%; Meta contributes roughly $2.3 B in land and construction, BlackRock provides about $4.9 B in cash, and the project is financed with roughly $12.5 B of debt, with the facility expected to go online in 2028.

Although Meta’s equity share is only 20%, this does not mean it bears only 20% of the cost. The equity determines asset ownership, not who bears the full economic burden. Meta’s goal is to secure compute usage rights rather than own the land, buildings, power, cooling, and networking infrastructure.

02 – Lease‑Based Access to Compute

Meta will lease the entire park for an initial four‑year term, with four optional four‑year renewals, giving a potential usage horizon of up to 20 years. Through the lease, Meta controls the full compute capacity despite its minority equity position.

03 – Funding Obligations and Cash‑Flow Sources

Meta’s capital contribution is limited to the 20% equity, but the project’s debt must be serviced, investors require returns, and the data centre incurs ongoing operating and maintenance expenses. As the sole initial user, Meta’s future rent payments become the primary cash‑flow source for the financing structure. Creditors provide capital today, while Meta supplies future cash flow via lease payments.

Creditors are effectively buying Meta’s credit, not the physical assets. Because the centre is custom‑built for Meta, it cannot be easily re‑let if Meta discontinues use. Therefore, creditors assess Meta’s long‑term rent‑paying ability and the asset’s value protection mechanisms.

Meta also provides a residual‑value guarantee of roughly $13 B that declines over time; if certain conditions trigger and the fair value of the assets falls below the threshold, Meta may need to cover the shortfall. Rating materials link the project’s investment‑grade credit to Meta’s lease‑payment support, its credit quality, and its responsibility for construction overruns.

04 – Beyond a Simple “BlackRock Pays, Meta Gets Compute”

The transaction is better described as external capital (funds and creditors) financing the project entity, which holds the assets and debt, while Meta supplies the lease, payment support, and residual guarantee that make the financing viable. Debt is not issued directly by Meta, but creditors ultimately rely on Meta’s ongoing profitability and rent payments.

05 – Repeating the Model

The El Paso structure is not unique. In 2025 Meta and Blue Owl used a similar arrangement for the Hyperion data centre in Louisiana: Meta kept a 20% equity stake, external capital was supplied via a special‑purpose vehicle, and Meta acted as developer, operator, and tenant. The recurring 80:20 pattern is becoming a financing template for Meta’s AI‑infrastructure expansion.

06 – AI Competition Shifts to Capital Structure

AI competition has evolved from acquiring more GPUs, to securing land, power, and data‑centre sites, and now to obtaining lower‑cost, longer‑term financing. Reuters cites U.S. bank data showing AI‑related bond issuance reaching about $270 B by early July 2026, nearly double the total issued in 2025.

07 – Risks and Long‑Term Obligations

The financing eases current capital pressure but does not eliminate future investment risk. If AI services generate sufficient revenue, the long‑term lease helps Meta acquire compute faster; if returns fall short, rent, guarantees, and financing remain regardless of technology changes. The compute can be repurposed, but the 1 GW data centre still requires sustained cash flow.

08 – Strategic Insight

The key takeaway is that large tech firms are turning corporate credit into a new form of AI production asset. The new barrier in the AI arms race is not just hardware or data, but the ability to incur and service large amounts of debt.