“Market Jitters Over AI Debt” Cash-Strained Meta Joins Forces With BlackRock to Build Data Centers

Meta’s cash burden intensifies amid astronomical AI capital expenditure
Forms $14 billion joint venture with BlackRock to tap external capital
Risk premium spreads across AI data-center debt market

Meta has agreed to establish a joint venture with BlackRock, the world’s largest asset manager, to build artificial intelligence (AI) data centers. As the cost of AI infrastructure investment reaches astronomical levels, Meta has opted to build and lease data centers with outside capital rather than shoulder the entire burden itself. However, if the data centers secured with vast external funding fail to generate new cash flows, long-term lease payments, refinancing costs, and declining residual values could simultaneously increase the burden on Big Tech and financial investors.

Meta-BlackRock $14 Billion Joint Venture

According to Reuters and other reports on the 30th, local time, Meta will establish a joint venture with BlackRock to build a $14 billion AI data center in El Paso, Texas. Funds managed by BlackRock will hold an 80% stake in the venture, while Meta will own the remaining 20%. Meta will contribute $2.3 billion worth of land and assets under construction in kind, while BlackRock plans to inject cash through debt financing and other means. In principle, the two companies will also bear the costs of data-center development and infrastructure construction in proportion to their stakes. The structure places a substantial share of the large-scale investment burden on BlackRock.

Once the data center is completed, Meta plans to lease the entire facility from the joint venture. The arrangement resembles a sale-and-leaseback structure, under which Meta does not directly own the data center but effectively uses it as a dedicated long-term facility. The El Paso data center will have 1 gigawatt (GW) of computing capacity, equivalent to the output of a large nuclear reactor. It is scheduled to begin full-scale operations in 2028, supporting Meta’s training and inference for frontier AI models as well as its AI service operations.

Meta’s Interest Costs Rise by $1 Billion Over Previous Project

Meta’s decision to adopt the joint-venture structure appears to reflect the growing strain that AI investment is placing on its cash flow and financial structure. According to The Wall Street Journal, the interest rate on $12.55 billion in bonds issued to finance Meta’s lease of the data center under construction in El Paso reportedly climbed to 7.5% annually, a level associated with speculative-grade, or high-yield, debt. The coupon on the bonds is approximately 2.875 percentage points above U.S. Treasury yields. Bond yields for blue-chip companies such as Meta typically trade only modestly above the U.S. 10-year Treasury rate, but this issuance was priced at interest rates more commonly seen in the riskier end of the debt market.

As a result, Meta’s data-center costs are also set to rise. During the Hyperion data-center project last October, Meta issued $27 billion in data-center bonds. The interest rate on the latest issuance is about 0.4 percentage points higher than that transaction. Meta’s additional annual interest expense will rise by more than $50 million each year. The newly issued bonds mature in 2048, meaning that Meta will pay an additional cumulative $1 billion over the life of the project.

This also means that the financing cost of building a single AI data center has become far more expensive than in the past. Financial markets are now awash with large-scale bond issuances launched simultaneously to fund AI infrastructure investment. In addition to Meta, major Big Tech companies are drawing not only on corporate bond markets but also on private equity, insurers, and infrastructure investors to secure the vast sums needed for AI infrastructure and power-grid capacity. Concerns are mounting that AI demand may fail to keep pace with the expansion of computing capacity, yet companies continue to compete intensely to secure computing-power capacity in an effort to dominate the future market.

Table 1. Meta-BlackRock El Paso AI Data Center Investment Structure

BlackRock Becomes the Financial Lifeline for AI Data Centers

In this process, BlackRock is serving as a critical funding source. The world’s largest asset manager acquired infrastructure specialist Global Infrastructure Partners (GIP) in 2024, followed by private-credit manager HPS Investment Partners last year. BlackRock’s integrated private-finance platform, including HPS, had $190 billion in client assets at the time of the acquisition, while GIP’s assets under management now exceed $200 billion. Infrastructure-investment capabilities to develop and own data centers and private-finance capabilities to raise long-term debt have thus been combined within a single organization.

BlackRock also provided more than $3 billion in debt financing for Meta’s Hyperion data-center project in Louisiana last year. Of the $27 billion in bonds issued at the time, global asset manager PIMCO purchased $18 billion, while BlackRock also participated as a key investor. Its role has expanded further in the El Paso project. BlackRock has moved beyond bond investing to become the joint venture’s largest shareholder, infrastructure manager, and debt-financing vehicle.

Its investment footprint is also broadening rapidly. A consortium including BlackRock’s GIP and Abu Dhabi investment company MGX completed its acquisition of Aligned Data Centers on the 21st at an enterprise value of $40 billion, while also committing an additional $5 billion in growth capital. Aligned owns or is developing 51 data-center campuses with total capacity of 6.4GW. The BlackRock-led AI Infrastructure Partnership has also set a target of raising $30 billion in equity capital and combining it with debt to secure up to $100 billion in investment capacity.

The Shadow of Private Credit

Private credit, which BlackRock is actively deploying, is a market that connects institutional capital outside the banking system with corporate loans and project bonds. It encompasses a range of products, from corporate loans carrying elevated default risk to investment-grade infrastructure bonds backed by long-term lease agreements from high-quality companies. The El Paso bonds, for instance, were arranged with participation from BlackRock subsidiary HPS, but are investment-grade project bonds structured around Meta’s lease payments and residual-value guarantees.

Nevertheless, it is clear that anxiety is growing across the private-credit market. A $26 billion corporate-loan fund managed by BlackRock’s HPS received redemption requests totaling $1.2 billion, or 9.3% of net asset value, in the first quarter of this year. HPS paid only $620 million, reaching its contractual quarterly redemption limit of 5%. Average redemption requests at 16 publicly traded business-development companies tracked by Fitch Ratings rose to 10.3% of shares outstanding in the second quarter.

Redemption pressure is also undermining confidence in the valuation of private assets. In some secondary transactions, stakes in private-credit funds managed by HPS, Apollo, and Ares were offered at prices 15% to 30% below net asset value. A considerable gap has emerged between the asset values recorded on fund books and the prices investors are willing to accept when converting holdings into cash. Given the long maturities and low trading frequency of private assets, a rush of redemption requests can force managers either to liquidate assets at steep discounts or restrict withdrawals.

Such concerns were also reflected in the elevated interest rate on the El Paso bonds. Orders of up to $20 billion were received for more than $12.5 billion in bonds, yet the spread did not tighten during the bookbuilding process. Investment-grade bonds typically see yields fall below initial guidance when demand exceeds the issue size, but the initial terms remained unchanged in this transaction. Institutional investors have less capacity to absorb additional AI-related debt amid continued large-scale bond issuance by Big Tech, while uncertainty over investment returns has increased.

AI Monetization Will Determine Investment Success

The AI data-center market will continue to require enormous amounts of outside capital. McKinsey expects cumulative global investment in data centers, excluding information-technology equipment, to exceed $1.7 trillion by 2030. Data-center capacity measured by power demand is projected to rise from 82GW in 2025 to 220GW in 2030, with AI-related facilities accounting for roughly 70% of the total. As the scale of investment has moved beyond what Big Tech can fund solely with its own cash, participation by private credit, infrastructure funds, and insurance capital is bound to increase further.

The problem is the gap between investment speed and revenue generation. Meta’s second-quarter revenue rose 28% year on year, but operating profit fell 8%. Cash flow from operations reached $31.86 billion during the same period, but free cash flow dropped sharply to $784 million as capital expenditures surged to $31.08 billion. Its capital-expenditure outlook for the full year stands at $130 billion to $145 billion. Meta’s cash-generation capacity remains solid, but AI capital expenditure is rapidly consuming available cash.

For expanding AI investment to remain merely an industry’s growing pain, improvements in advertising efficiency and user engagement alone will not suffice. The key lies in whether enterprise AI services, paid agents, and cloud-computing sales translate into meaningful cash inflows, while data centers maintain high utilization rates. Only then will companies have room to absorb financing costs as the price of securing market leadership. Conversely, if monetization takes longer than expected, long-term lease payments and refinancing costs will weigh on cash flow, while falling residual values could further increase the risk of losses for both Big Tech and financial investors.