Woofun AI reports that a $12.3 billion bond issuance for Meta’s El Paso, Texas data center project was executed on July 27 by entities linked to BlackRock, marking a distinct departure from traditional corporate debt structures. The transaction, centered on the Sopaipilla Investor vehicle, highlights a growing divergence between tech giant credit ratings and the specific risk profiles of artificial intelligence infrastructure. This deal underscores how capital markets are increasingly pricing AI data centers as standalone assets rather than mere extensions of parent company balance sheets.

Pricing dynamics revealed immediate market friction, with Bloomberg noting that the bond’s yield had already escalated in the gray market prior to formal pricing. At the time of finalization, the yield stood approximately 2.60 percentage points higher than U.S. Treasuries, a spread that suggests investors demanded compensation for perceived structural risks. Despite this premium, the bond did not fail to attract buyers, indicating that while the cost of capital rose, the underlying demand for AI infrastructure exposure remained robust among institutional participants.

However, the order book’s weakness provided a less enthusiastic counter-narrative to the successful issuance. The final order book totaled around $20 billion, representing only 1.6 times the issuance size, a figure significantly lower than the Bloomberg-reported average of roughly 4 times subscription levels for 2026 bonds. More critically, the formal pricing was approximately 40 basis points higher than the Beignet 2049 bond issued last year for Meta’s Louisiana data center financing, signaling a tangible increase in the cost of capital for similar projects.

This 40 basis point premium, though seemingly minor, carries profound implications in the context of a multi-decade project bond. It indicates that the market is willing to lend but requires a higher price to account for the unique uncertainties of AI data centers, which are no longer viewed simply as capital expenditures for tech companies. Instead, these facilities are emerging as a distinct class of infrastructure debt that necessitates separate valuation, decoupled from the broader corporate credit profile of the sponsor.

Woofun AI data shows that the bookbuilding process itself illustrated how the market negotiated this new risk paradigm. Initial guidance from Sopaipilla Investor was set at around 2.875 percentage points above U.S. Treasuries, a rate designed to attract risk-taking funds. Through the bookbuilding process, this spread narrowed to about 2.60 percentage points, reflecting a gradual adjustment based on order size and investor appetite. This progression resembled a roadshow for scoring the project, where underwriters started with a higher rate and lowered it as confidence in the order book grew.

Sources of this additional cost and risk are multifaceted, stemming from both supply-side constraints and project-specific variables. Bloomberg states that large tech companies issuing bonds are tightening the available funds for asset managers, creating competitive pressure.

Furthermore, data centers require upfront investments in land, electricity, construction, and equipment, followed by a wait for tenants to fulfill long-term contracts. Creditors are thus facing a set of arrangements where future cash flows may or may not materialize as planned, rather than a business already reflected in Meta’s financials.

The deal structure and ownership split further clarify why Meta did not issue the bonds directly. The Texas project is financed by a Belelde-related entity called Sopaipilla Investor, with Belelde-related entities holding an 80% stake in the project and Meta holding the remaining 20%. The project, expected to be around 1GW and come online in 2028, is structured so that the bonds are issued by a holding company, with funds flowing into the project rather than Meta’s balance sheet, effectively isolating the debt from the parent company’s direct liabilities.

This structure is not unprecedented, as evidenced by the precedent of the Louisiana Hyperion project. The initial joint venture scale of the Hyperion project in Louisiana is around $27 billion, also following an 80% financial capital and 20% Meta arrangement, with some funding coming from private debt offerings targeted at institutional investors. The two structures look almost like copies of each other, suggesting a standardized approach to financing large-scale AI infrastructure through specialized project companies rather than direct corporate borrowing.

The macro context reveals why this structure is now commonplace: Meta’s capex surge is outstripping traditional cash flow rhythms. According to the U.S. Securities and Exchange Commission, Meta’s 2025 capital expenditure was $69.691 billion, approximately 4.6 times that of 2020. Consequently, free cash flow decreased from $54.072 billion in 2024 to $46.109 billion. With the Hyperion scale projected to exceed $500 billion and 5GW, project financing allows Meta to spread the load, shifting risk from corporate balance sheets to project-specific cash flows.

Ultimately, Sopaipilla serves as a mini stress test for the market, demonstrating that AI infrastructure funds are differentiating 'big tech participation' from 'worthy of borrowing against big tech credit.' As data centers move from corporate budgets to the project bond market, risk transitions from financial reports to longer contracts. For further insights, join the official BlockBeats community: Telegram Subscription Group: https://t.me/theblockbeats, Telegram Discussion Group: https://t.me/BlockBeats_App, Official Twitter Account: https://twitter.com/BlockBeatsAsia.