Meta + BlackRock's $14 Billion El Paso Bet — The New AI Funding Template

On July 28, 2026, Meta Platforms and BlackRock announced a $14 billion joint venture to develop a 1-gigawatt artificial intelligence data center campus in El Paso, Texas. BlackRock-managed funds own approximately 80 percent of the venture, Meta retains roughly 20 percent and leases back the entire capacity. BlackRock is leading a debt sale of at least $12 billion to finance construction, a structure that keeps the bulk of the asset and the financing off Meta's own balance sheet. The deal is the largest single hyperscaler-data-center partnership of 2026 and the clearest signal yet that hyperscalers are turning to institutional capital and asset managers to fund the AI infrastructure buildout that their own balance sheets cannot carry.

The right read of the Meta-BlackRock deal is the practical template for the next 18-24 months of hyperscaler AI infrastructure. Hyperscalers (Meta, Microsoft, Google, Amazon) face a hard constraint: the AI capex requirement is projected at more than $730 billion across the top seven tech companies in 2026, and the hyperscalers' balance sheets cannot fund it from operating cash flow alone. The Meta-BlackRock structure solves the constraint by creating an off-balance-sheet special-purpose vehicle (SPV) that owns the data center, raises debt against it, and leases the capacity back to the hyperscaler. Meta gets the AI compute capacity it needs without diluting its equity or straining its debt capacity. BlackRock gets a 1-gigawatt data center asset with a 10-15 year lease to a credit-grade hyperscaler tenant.

The deal at a glance

Why this structure makes sense for both sides

For Meta, the off-balance-sheet treatment is the right answer to the capex problem. Meta's 2026 capex is projected at $115-145 billion for AI infrastructure, and the El Paso campus would represent roughly 10 percent of that on Meta's own balance sheet. By spinning it out into an SPV that BlackRock owns 80 percent of, Meta reduces its direct capex by $11+ billion while preserving access to the full 1-gigawatt capacity. The accounting treatment (operating lease rather than capital lease) means Meta recognizes the lease expense over the lease term rather than booking the full asset on its balance sheet, which improves Meta's return on invested capital and its debt-to-EBITDA ratio.

For BlackRock, the deal is a new asset class at scale. The $14 billion data center SPV with a 10-15 year Meta lease is a credit-quality infrastructure asset that yields 7-10 percent to BlackRock's infrastructure fund. The $12 billion bond sale is a separate transaction, with the bonds priced at investment-grade yields to bank, insurance, and pension fund buyers. BlackRock's equity stake (the 80 percent of the SPV, after the bond sale) is the residual claim on the data center after the bonds are paid. The economics for BlackRock are 15-20 percent leveraged-equity IRR, which is the standard infrastructure-fund target.

The 1-gigawatt scale matters

The El Paso campus is sized at 1 gigawatt of IT load, which is at the high end of data center campuses in 2026. For comparison, the average hyperscaler campus in 2022 was 50-100 MW. The Nvidia GB200 NVL72 rack draws approximately 120 kW per rack, and a 1-gigawatt campus can host roughly 8,300 such racks, or roughly 600,000 Nvidia Blackwell GPUs running at full utilization. That is enough capacity to train a frontier model (gpt-oss or Gemini scale) from scratch in a few weeks, or to serve billions of inference requests per day. The 1-gigawatt scale is what makes the deal structurally interesting: it is large enough to matter to Meta's AI roadmap, but not so large that Meta would build it on its own balance sheet.

Why El Paso

The site selection for El Paso is driven by two factors that have become binding constraints on data center development in 2026. First, power availability. El Paso Electric, the local utility, has secured long-term power purchase agreements (PPAs) with solar and wind generation in the Texas-New Mexico border region, which gives the El Paso campus a low-cost, low-carbon power supply that is unavailable in Northern Virginia (where Dominion Energy's interconnection queue is 4-7 years). Second, land availability. El Paso is a low-density market with large parcels of land available at low cost, which is the right input for a 1-gigawatt campus. The right read is that the El Paso site is the result of two structural shifts in data center site selection, away from the legacy carrier-hotel markets (Northern Virginia, Silicon Valley) toward markets where power and land are available.

The El Paso site is also a political statement. Texas has been the most aggressive state in promoting AI infrastructure development, with state and local tax incentives that materially reduce the total cost of ownership for a 1-gigawatt campus. Meta's choice of El Paso over a competing site in a different state is a signal to other hyperscalers and asset managers that Texas will continue to be the leading AI infrastructure market in the US through 2030. The right read for an operator is that site selection is now a state-and-local policy decision, not just a power-and-land decision.

The risk in the structure

Three concrete risks in the Meta-BlackRock structure. First, lease default risk. If Meta defaults on the lease (because of a financial crisis, a strategic pivot away from AI, or a regulatory event), BlackRock is left holding a 1-gigawatt data center asset with no anchor tenant. The secondary market for 1-gigawatt data centers is thin, and the asset is highly specialized, so the recovery value in a default scenario would be lower than the face value of the lease. Second, technology risk. The 1-gigawatt campus will be filled with Nvidia GPUs or competing accelerators, and the residual value of the campus depends on the chips being compatible with the next-generation workloads. If the AI compute architecture shifts in a direction that makes the campus less valuable (for example, if Nvidia's dominance erodes or if hyperscalers shift to custom silicon), the lease economics compress. Third, regulatory risk. The El Paso site is in Texas, which has been favorable to AI infrastructure, but the federal regulatory environment around AI and data centers is shifting. New export controls on AI chips, new environmental rules on data center power and water use, or new state-level policy changes could all affect the deal economics.

What this means for the rest of the AI buildout

The Meta-BlackRock structure is the practical template that will be replicated across the AI buildout in 2026-2028. The hyperscalers that need AI infrastructure will form off-balance-sheet SPVs with asset managers (BlackRock, Blackstone, Brookfield, KKR, Apollo) that own the data center and lease capacity back to the hyperscaler. The hyperscalers get access to AI compute capacity without diluting their balance sheets. The asset managers get a credit-quality infrastructure asset class with 15-20 percent leveraged-equity returns. The total addressable market for this structure is the $300-400 billion in AI data center capex that needs to be financed over 2026-2028.

The competitive implications are significant. Hyperscalers that can form these partnerships will accelerate their AI infrastructure deployment. Hyperscalers that cannot (because of weaker credit profiles or because of regulatory constraints) will fall behind. The asset managers that can originate these deals (BlackRock, with its scale and distribution; Blackstone, with its credit and equity arms; Apollo, with its private credit focus) will capture a disproportionate share of the financing. The chip designers (Nvidia, AMD, Broadcom) benefit indirectly, because the financing structure accelerates chip deployment without constraining their pricing power. The data center REITs (Digital Realty, Equinix) face competitive pressure from the new SPV structures, which are creating institutional-grade data center supply faster than the REITs can build it.

What an operator should conclude

The Meta-BlackRock El Paso deal is the canonical example of the new AI infrastructure funding model. Hyperscalers and asset managers are forming off-balance-sheet SPVs to fund AI data centers, with the hyperscaler as anchor tenant and the asset manager as majority owner. The structure accelerates AI infrastructure deployment, preserves hyperscaler balance sheets, and creates a new asset class for institutional investors. The right model is that this structure will become the default for the next 18-24 months of AI infrastructure deals.

Three concrete takeaways. First, if you are a hyperscaler evaluating AI infrastructure financing, the off-balance-sheet SPV structure is the right answer to the capex constraint. Meta's El Paso deal is the template. The trade-off is the long-duration lease obligation (10-15 years) and the loss of direct ownership of the data center asset. Second, if you are an asset manager evaluating this asset class, the yields are 7-10 percent on the equity portion and 4-6 percent on the senior debt portion. The technology and lease default risks are new, but the underlying frameworks are the same as commercial real estate and infrastructure. Third, if you are a data center REIT (Digital Realty, Equinix, Iron Mountain), the SPV structures are competitive pressure. The REITs need to either match the SPV economics (which requires taking on more debt) or differentiate on operational expertise and interconnection density that the SPVs do not provide.

Frequently asked questions

What is the Meta-BlackRock El Paso deal

A $14 billion joint venture between Meta Platforms and BlackRock to build a 1-gigawatt AI data center campus in El Paso, Texas. BlackRock-managed funds own ~80 percent, Meta owns ~20 percent and leases back 100 percent of the capacity. BlackRock is leading a $12+ billion bond sale to finance construction, with the bulk of the asset and financing off Meta's balance sheet.

Why is the off-balance-sheet structure important

Meta's 2026 capex is projected at $115-145 billion for AI infrastructure. The El Paso campus would represent ~10 percent of that on Meta's balance sheet. By spinning it out into an SPV, Meta reduces its direct capex by $11+ billion while preserving access to the full 1-gigawatt capacity. The accounting treatment (operating lease) improves Meta's return on invested capital and debt-to-EBITDA ratio.

Why El Paso

Two factors. First, power availability. El Paso Electric has secured long-term PPAs with solar and wind generation in the Texas-New Mexico border region, which is unavailable in Northern Virginia. Second, land availability. El Paso is a low-density market with large parcels at low cost. The right read is that the El Paso site is the result of two structural shifts in data center site selection, away from legacy carrier-hotel markets toward markets where power and land are available.

What are the comparable deals

The Nvidia $500 billion compute financing platform (Aug 2026), the Broadcom $60 billion AI debt vehicle (Aug 2026), and the Marvell-Google $12.2 billion warrant (Aug 2026) are the comparable arrangements. Each one represents the same structural shift, with the asset manager or chip designer taking on the long-duration financing risk that the hyperscaler would otherwise absorb.

How much will the bonds yield

Investment-grade yields, typically 4-6 percent in 2026, with a small premium for the data center specificity and the partial Meta lease guarantee. The bonds are credit-rated based on the lease income from Meta, not on BlackRock's balance sheet, so the rating is essentially a Meta-credit rating adjusted for the lease structure.

What is the total addressable market for this structure

The $300-400 billion in AI data center capex that needs to be financed over 2026-2028, across all hyperscalers (Meta, Microsoft, Google, Amazon, Oracle) and asset managers (BlackRock, Blackstone, Brookfield, KKR, Apollo). The Meta-BlackRock deal is the template; the rest of the market will follow.

Sources

• Reuters — Meta, BlackRock partner on $14 billion El Paso data center — the deal announcement, JV structure, lease-back detail
• Quartz — Meta and BlackRock form $14 billion El Paso data center venture — 80/20 ownership structure
• ECM Source — Meta, BlackRock Team on $14B El Paso AI Data Center — $12B bond sale, accounting treatment detail