In late July 2026, a shell company most people had never heard of placed roughly $12.5 billion of bonds due in 2048, at a coupon near 7.5%, to finish building a data center for Meta in El Paso. The common read is that Meta simply borrowed cheaply through a partner. The more interesting reality: Meta sold 80% of a gigawatt-scale AI campus to BlackRock-managed funds, kept just 20% plus total operational control, and arranged the deal so the debt sits on someone else's books while Meta's future rent quietly repays it.[1][2] The mechanism matters because it turns a capital-expenditure problem into a lease-expense arrangement, it hands infrastructure investors a bond backed by one of the world's most creditworthy tenants, and it is fast becoming the template for how hyperscalers fund the AI build-out.
Why it matters now
The dateable trigger was July 27–28, 2026, when Sopaipilla Investor, LLC placed about $12.5 billion of 7.534% amortizing notes due November 30, 2048, arranged by JPMorgan and Morgan Stanley.[3][1] Meta folded its half-built El Paso campus, roughly 1 gigawatt of capacity, into a joint venture: BlackRock's Global Infrastructure Partners and HPS Investment Partners contributed cash for 80%, Meta contributed about $2.3 billion of land and construction-in-progress for 20%, and took a $1 billion cash distribution at closing to balance the split.[1][2] The coupon landed above 7%, roughly 0.4 points higher than Meta's earlier Hyperion financing in Louisiana, a signal that even blue-chip AI infrastructure now prices in real cost-of-capital pressure.[4] The stakes: the AI arms race is colliding with balance-sheet limits, and this deal shows one way around the wall.
The longer view
Over the next two to five years, the constraint on AI is shifting from chips to the capital and power that house them. A gigawatt campus running into the billions is too large and too fast-multiplying to sit comfortably on even a cash-rich tech balance sheet, so the leverage moves to whoever can supply patient infrastructure capital: asset managers like BlackRock, whose GIP and HPS arms are built for exactly this.[1][2] The moat is no longer just model quality or engineering talent. It is access to structured financing that keeps reported debt low while still commanding compute. Watch for the regulatory and accounting response, because the whole edifice depends on how these leases are classified. If auditors or standard-setters decide a residual value guarantee near $13 billion against $12.5 billion of debt forces consolidation, the off-balance-sheet advantage evaporates.[2][5] The role of the electric utility also rises: the campus reportedly rests on an electric service agreement running to 2048 with capacity options tied to new substations and transmission.[6]
The key insight
The engine here is a swap of ownership for control. Meta gives up 80% of the equity but keeps every operational lever: sole tenant, construction manager, property manager.[1] In exchange, the debt is issued by an entity Meta doesn't fully own, and Meta's obligation reappears as rent rather than borrowing. The old world had a company build an asset, put it on its books, and finance it with corporate debt that lenders assessed against the whole firm. The new world splits the asset into an economic shell owned by outside capital and a stream of contractual rent from a single, highly creditworthy occupant. Bondholders are not really underwriting a data center. They are underwriting Meta's promise to pay.
How it works
The binding constraint is creditworthiness of cash flow. Infrastructure bonds need predictable revenue, and a half-built data center has none, so the structure manufactures certainty by anchoring everything to Meta's rent.
Start with the entities. Project Sopaipilla Holdings owns the joint venture; Sopaipilla Investor, LLC is the special-purpose issuer that sold the bonds.[7][3] A special-purpose vehicle is a legally separate company created for one financing, so its debt does not automatically count as the sponsor's debt. Proceeds fund construction; repayment relies on rent, not on Meta's corporate borrowing.[8]
The lease is a triple-net arrangement, meaning the tenant, Meta, absorbs operating costs, taxes, and property risks rather than the landlord.[9] Meta's subsidiary Roadrunner Leap reportedly signs eight leases across the campus buildings, with rent beginning December 2028 even though note coupons accrue from 2026, so roughly $1.9 billion of early interest is capitalized into the raise.[6] The bonds are secured by the issuer's accounts, revenues, and equity interests in the joint venture, not by a direct pledge of the physical buildings.[10]
What comes back to bondholders is a cascade of protections layered onto that rent: (1) a 20-year payment horizon matching the 2048 maturity, (2) construction-overrun liability that falls on Meta above 105% of budget, and (3) a residual value guarantee reportedly sized near a $13 billion aggregate threshold over 16 years, which backstops asset value if Meta declines to renew.[9][11][2] Meta can walk away without penalty only after a severe casualty delaying construction more than 18 months; otherwise abandonment triggers a hefty fee.[11][12] Fitch was reported to estimate about $1.12 of assets for every $1 of debt.[13]
Implications
For companies, this is a financing playbook, not a one-off. The near-term example is concrete: other hyperscalers facing the same AI capital wall now have a proven route to fund gigawatt campuses through infrastructure partners while reporting the spending as lease expense rather than debt.[2][14] Expect a wave of copycat structures, each tuned to the sponsor's credit and the local utility's capacity. For BlackRock, the deal validates a growth engine: channeling institutional infrastructure capital into AI data centers at yields above 7%, backed by tenant credit rather than merchant power risk.[4][15]
The directional medium-term shift is in who bears risk and who reaps optionality. In principle, BlackRock's funds absorb the equity and construction-financing risk while Meta keeps the strategic compute it actually needs.[1] In practice, Meta's overrun liability, abandonment fee, and residual value guarantee keep a large share of the real economic exposure with the tenant, which raises a harder question for finance and accounting teams. If these arrangements are eventually judged to be finance leases or consolidated variable-interest entities, the reported balance-sheet relief reverses, and the CFO's clever structuring becomes a footnote about deferred liabilities.[2][5] Work shifts accordingly: the scarce roles become structured-finance and infrastructure specialists who can engineer, price, and audit these hybrids.
Tensions & open questions
Off balance sheet, or just deferred. One reading holds that issuing through the SPV keeps the liability off Meta's books and converts capital spending into lease expense.[1][7] Another holds that a residual value guarantee near $13 billion against $12.5 billion of debt, plus 20% equity and full operational control, likely forces finance-lease capitalization or consolidation onto Meta.[2][5] No filed accounting determination settles it.
Four-year lease, 24-year paper. The initial lease term is only four years, yet the bonds run to 2048.[1][16] Whether Meta's rent truly secures two decades of debt depends on unexercised extension options and on the residual value guarantee acting as a synthetic extension covering outstanding principal on non-renewal, and those trigger mechanics are not document-verified.[5]
Transfer or recharacterization. Reasonable observers split on whether this genuinely moves risk to BlackRock or merely repackages Meta's own exposure. The overrun liability above 105%, abandonment fee, and an 85% minimum-demand power commitment leave much economic weight on Meta.[11][12][6]
Contested headline numbers. Reporting varies between $12.3 billion and $12.547 billion of notes, and between a ~$14 billion campus value and a $16.8 billion total structure including equity.[3][6] It is unresolved whether these reflect tranches, revisions, or rounding.
Talking points
- Meta sold 80% of a gigawatt AI data center in El Paso to BlackRock funds but kept total operational control as sole tenant, builder, and property manager.[1]
- A shell company called Sopaipilla issued about $12.5 billion of bonds at 7.5% due 2048, and they're really backed by Meta's future rent, not the buildings.[3][10]
- The clever bit: the debt stays off Meta's balance sheet and shows up as lease expense instead. Whether accountants let that stand is the open question.[2][5]
- Bondholders aren't underwriting a data center. They're underwriting Meta's promise to pay rent for 20 years.[5]
- This is becoming the template for how the whole industry funds the AI build-out.
The bottom line
- Core idea: Meta traded 80% ownership of its El Paso campus for outside infrastructure capital while keeping control and shifting $12.5 billion of debt off its books.[1][3]
- Why it matters: It's an emerging playbook for financing gigawatt AI campuses when the build-out outgrows corporate balance sheets.[14]
- What to watch: The accounting verdict on consolidation, the residual value guarantee's true mechanics, and whether rivals copy the structure.[2][5]
Technical detail
The distribution waterfall reveals how carefully the incentives are stacked. Reporting describes pro rata distributions to BlackRock's issuer and to Meta until the issuer clears an 8.25% internal rate of return and a 2.4x multiple on invested capital, after which Meta's reimbursements for cost overruns, power, and repair costs move ahead in priority.[6] Internal rate of return measures annualized profit on invested cash; multiple on invested capital measures total dollars returned per dollar in. The placement of Meta's operating reimbursements behind an equity hurdle is contested, since literal subordination of power and repair costs to an equity return reads as structurally implausible, and the precise boundary is unconfirmed.
The power arrangement is equally load-bearing. An electric service agreement reportedly runs to 2048, matching the bond maturity, permits up to 1.0 GW of committed capacity with a 1.4 GW option conditioned on three new substations and on-site generation of up to 225 MW, and carries a reported 85% minimum monthly contract-demand obligation.[6] Whether that 85% figure is a literal take-or-pay clause or an analytical description of a minimum-billing covenant turns on the unpublished agreement text, and it matters because a hard take-or-pay would deepen Meta's fixed obligations well beyond the rent.