Meta's July 28 El Paso announcement disclosed an unusually detailed capital structure for a data-center project. Funds managed by BlackRock will hold 80% of the venture and Meta 20%. The parties have committed to fund roughly $14 billion of development costs. At financial close, Meta is to contribute land and construction in progress valued at about $2.3 billion, BlackRock is to contribute about $4.9 billion in cash, and Meta is to receive a one-time distribution of roughly $1 billion to align ownership. Part of BlackRock's investment is to be financed by $12.5 billion of debt.
Those figures do not describe a completed or energized $14 billion campus. Meta says the El Paso site is under construction and expects capacity to begin coming online in 2028. Its stated 1GW of compute is a forward-looking campus target, not disclosed delivered power, installed accelerators or billable capacity. The announcement does not disclose energization, interconnection, debt tenor, interest rate, collateral, amortization, rack count or rent per megawatt. Treating the disclosure as one gigawatt of operating supply would skip several still-unobserved delivery stages.
The useful question is therefore not how large the headline is. It is who carries the risk when buildings, substations, cooling plants and connectivity are financed as long-lived infrastructure while server generations, workloads and utilization can change much faster. El Paso provides a contractual allocation, not an operational answer. Meta's lease, management role and residual-value support keep its credit and demand attached to the project; BlackRock's equity and debt financing bring infrastructure capital into construction-period risk and return.
The Bank of England's July Financial Stability Report puts this structure into market context. It says AI companies accelerated their use of external finance in the first half of 2026, with most debt financing data-center shells and facilities rather than the servers and chips inside them. The report also identifies a potential mismatch between long-dated debt and faster, uncertain AI-hardware life cycles. That is a macroprudential observation, not a statement about El Paso's debt terms. It does, however, identify why leases, guarantees and the ability to adapt a campus matter to AI-infrastructure credit.
One campus, four layers of capital
The first disclosed layer is the asset and equity structure. BlackRock-managed funds take an 80% interest while Meta keeps 20%, with Meta contributing land and construction in progress to the venture. The second is cash and debt: BlackRock's roughly $4.9 billion cash contribution is partly supported by $12.5 billion of debt financing. The third is occupancy: Meta will lease the entire campus and be its initial sole occupant. The fourth is credit support: Meta provides residual-value guarantees that decline over time.
Each layer solves a different problem. Development cost describes the long-lived capital required for buildings and power, cooling and connectivity infrastructure. Equity determines who first absorbs changes in asset value. Debt creates fixed claims and refinancing exposure. The lease connects future use of the campus to Meta's operating demand. The residual-value guarantee can cover a specified value shortfall under specified conditions. None of those layers independently means that the campus has been built, powered or de-risked.
The figures also must not be added mechanically. Roughly $14 billion is the parties' planned total development cost. The $2.3 billion and $4.9 billion figures are asset and cash contributions at financial close. The $12.5 billion is a funding source for part of BlackRock's investment, not an additional campus cost. Meta did not disclose the debt's ultimate uses, interest rate, maturity, security, seniority or amortization. Atlas therefore treats those terms as not determined rather than filling them in with a typical project-finance template.
| Metric | Disclosure | Basis and boundary |
|---|---|---|
| Compute target | 1GW | A forward-looking Meta description of an under-construction campus expected to begin coming online in 2028. |
| Development cost | ~$14B | Planned buildings plus long-lived power, cooling and connectivity infrastructure. |
| Equity structure | BlackRock 80% / Meta 20% | Venture ownership; not proof that financial close has occurred. |
| Debt financing | $12.5B | Funds part of BlackRock's investment; debt terms are not public. |
| Residual-value threshold | ~$13B | Declines over time and applies only under contractual conditions. |
Asset-light does not mean risk-free
Meta is not handing an empty site to an independent developer. It will lease the entire campus, beginning with a four-year term and four extension options that could take the relationship to twenty years. Meta will also provide construction management, administrative and property-management services. For outside capital, those arrangements connect project cash flow to Meta's use of the site. For Meta, not owning every project asset directly does not remove the business consequence if delivery or technical fit fails.
The residual-value guarantee is especially easy to overstate. Meta says the guarantees have an aggregate threshold of roughly $13 billion that declines over time. If specified conditions are met in the first sixteen years of the lease, Meta's maximum payment equals the shortfall between then-current fair value and the applicable threshold. That is not an immediate, unconditional guarantee of all development cost. Its relevance changes with delivery, use, asset valuation and the debt and lease terms around it.
Atlas's view is that the transaction separates who owns the physical campus from who needs the compute, but it does not remove the latter from the risk chain. Meta retains economic exposure through the lease, management role and guarantee. BlackRock takes equity exposure and funds part of the investment with debt. Replicating the structure elsewhere would require more than a large financing headline: it needs an anchor tenant of comparable credit quality, a credible delivery plan and contractual support that bridges the construction period without confusing future demand with present capacity.
Asset-value exposure
The 80/20 split is the starting point for sharing changes in venture value.
Fixed-claim exposure
Debt supports part of the investment, but its maturity, pricing, security and seniority are undisclosed.
Demand enters the structure
Meta is the initial sole occupant; occupancy rights are not disclosed utilization.
Partial residual-value exposure
The guarantee is conditional and time-varying, not a blanket project backstop.
The same buyer is separating asset, occupancy and power choices
El Paso is an asset-venture route: external funds hold most of the long-lived infrastructure while Meta retains use and management. Meta's other public infrastructure disclosures show that a hyperscaler does not use only one model. Its April infrastructure explainer lists El Paso, Richland Parish, Lebanon and Tulsa as AI-optimized data centers under construction. That common status matters. A build plan can show capital direction; it does not prove delivered power, system readiness or current supply.
In Jamnagar, India, Meta announced an agreement to lease capacity at a Reliance AI-enabled data center while separately backing nearly one gigawatt of renewable energy with two energy providers. Asset ownership, use of capacity and energy procurement are therefore split among different agreements. The announcement does not disclose leased megawatts, actual generation from those energy projects or a cost basis comparable with El Paso, so the different structure cannot support a claim that one route is cheaper or faster.
Richland Parish, Louisiana illustrates a heavier build-and-power configuration. Meta said in July that it was expanding the site's compute target to 5GW with more than $50 billion of investment, and that its Entergy agreement would fund new gas generation, grid-scale batteries, nuclear uprates and other purchased power. The capacity, customer-savings and cost-allocation statements are Meta's own and do not substitute for utility or regulatory records. Their analytical value is narrower: a campus financing model cannot be assessed only through the data halls when power, transmission, storage and water determine delivery timing.
These are not a ranking of Meta's preferred models. Together they pose a testable question: can asset ownership, rights to use capacity and power responsibility close on the same timetable? El Paso connects financeable assets with an anchor tenant's credit. It does not automatically supply energized power, installed systems or productive workload.
- 01Asset ownership
El Paso puts long-lived facilities into a venture rather than keeping every asset solely on Meta's balance sheet.
- 02Capacity rights
Meta leases the whole El Paso campus and separately leases capacity in Jamnagar; neither disclosure establishes utilization.
- 03Power responsibility
Richland Parish shows company-described generation, storage and grid arrangements becoming part of campus economics.
- 04Delivery closure
Financial close, grid delivery, equipment acceptance and real load must occur in sequence before capacity is operational.
Five gates separate financing from usable compute
The first gate is financial close. Meta said the transaction was expected to close in the coming days; that is not proof that the conditions have been met. The second is civil works and long-lead equipment. Buildings plus power, cooling and connectivity infrastructure must arrive to budget and schedule, but no mechanical-completion percentage is public. The third is power and interconnection. A one-gigawatt compute target requires campus power and transmission arrangements, yet Meta did not disclose delivered megawatts, an interconnection agreement, a generation source or a commercial-operation date.
The fourth gate is system installation and acceptance. Accelerators, networks, cooling loops and controls have to be installed and pass reliability tests. Meta did not disclose suppliers, rack count, liquid-cooling architecture or acceptance cadence. The fifth is occupancy and utilization. The lease establishes who has the right to use the campus. It does not disclose a minimum utilization level, rent per megawatt or an actual production workload.
Only after the five gates have been crossed can the one-gigawatt figure be discussed as usable computing capacity. Announced, under construction, energized, system-ready and occupied are separate states. Collapsing them turns a forward-looking development objective into inflated supply, revenue and credit quality.
A 12–36 month view built from conditions, not forecasts
In an on-plan conversion, four public signals appear in sequence: financial close; power and long-lead-equipment milestones; system installation and acceptance; and the first capacity coming online in 2028. Only when the latter stages show actual workload, lease accounting or another use metric does the lease-and-guarantee structure begin to move from construction-period credit support toward observable asset and usage cash flow.
A second path is capital in place but delivery delayed. The venture can close while interconnection, equipment or construction moves later. In that case equity and debt carry construction-period risk longer, and the one-gigawatt target remains outside operating supply. This path does not require demand to disappear. It requires physical delivery to move more slowly than the financing documents.
A third path is demand or technology reconfiguration. If the lease, guarantee thresholds, intended use or asset-upgrade plan changes, redeployability and residual value become more important than the initial development cost. Conversely, if Meta activates the full campus on the initial timetable without changing the long-term arrangement, the original structure has closed the loop between asset, workload and technical fit.
This is not a probability call on any path. It converts foresight into observable leading indicators. Financial close verifies the capital structure. Energization verifies physical infrastructure. Acceptance verifies system availability. Workload and accounting disclosure verify economic use. None can stand in for the other three.
Does the capital structure close?
Watch financial close, debt borrower, tenor, security and guarantee terms—not only the financing headline.
Do grid and construction close together?
Watch interconnection, energization and long-lead-equipment milestones before treating construction as supply.
Does online become used?
Watch actual megawatts, racks, availability, lease accounting or revenue recognition.
What changes the view?
A formal reduction in timing, scope, use structure or guarantee support would change risk ownership.
Scale cannot substitute for verifiability
The first risk layer is transaction terms. Without the debt's pricing, tenor, seniority, amortization and covenants, actual leverage, interest coverage and refinancing pressure cannot be calculated. The second is construction and conversion. A 2028 online target is a company objective; any delay in power, cooling, networking, civil works or equipment pushes out capacity that can be leased or used.
The third is operating value and residual value. The guarantee threshold declines over time and applies only if conditions are met. The campus's long-term value will also depend on later accelerator power density, cooling fit, network architecture, workloads and energy efficiency. Facilities can outlive one server generation, but a facility unable to accommodate the next one can lose economic life earlier than its physical shell suggests.
The counter-case must remain visible. If the site energizes on schedule, Meta brings in long-lived workloads and later lease and guarantee disclosures match the announced structure, external infrastructure capital can advance a campus without Meta funding every asset directly. Current public records do not yet allow a conclusion between that outcome and the execution-risk case.
IDC ATLAS VIEWEl Paso moves the AI-campus question beyond who is spending the most. Outside equity and debt can hold long-lived facilities, while Meta's lease, management role and residual-value support keep demand and credit attached to the project. One gigawatt is a target, not operating capacity. The next proof points are financial close, energization, acceptance and real use—each one necessary, none sufficient on its own.