AI Race Turns into Bottomless Money Pit! Tech Giants Spend Heavily, Wall Street Finally Says 'Too Expensive'

The AI race continues to heat up, as tech giants invest astronomical sums to gain a computational advantage by expanding data centers. However, this financing frenzy is beginning to strain Wall Street. As companies like Meta and Google continue to increase their AI capital expenditures, demand for related bonds is showing signs of fatigue, and financing costs are rising—indicating that the AI construction boom is driving up capital costs across the industry.

Since the beginning of this year, major tech companies have aggressively issued bonds to fund AI infrastructure projects, rapidly increasing supply in the bond market. Meta’s latest financing deal for a data center in El Paso, Texas, had to offer a higher yield than similar transactions last year to attract investors.

Markets are becoming increasingly sensitive to the ballooning scale of AI investments. After Alphabet’s Google announced a more aggressive AI spending plan last week, tech stocks came under selling pressure, dragging down bond prices of companies like Microsoft and Amazon—reflecting investor concerns about a flood of future AI-related debt entering the market.

According to Bank of America Global Research, AI-related corporate bond issuance reached $270 billion by early July—nearly double the total amount issued in all of 2025. Still, despite rising financing costs, the market widely believes AI data center construction will continue, as no tech giant wants to fall behind in the AI race.

Meta plans to raise hundreds of billions of dollars to continuously expand its AI infrastructure. Sources indicate that Meta has informed investment banks and major asset managers that it will need to raise hundreds of billions of dollars to fund AI infrastructure development and is currently negotiating additional financing schemes with investment firms like Blackstone.

The Wall Street Journal recently reported that Nvidia is in talks with OpenAI to provide around $250 billion in funding to support the construction of a large-scale data center project in Ohio.

Neha Khoda, head of U.S. credit strategy at Bank of America, said the market expects the AI construction boom to continue, but “the price of capital will be higher than in the past.”

On Monday, Meta issued $12.55 billion in long-term bonds through Sopaipilla Investor, a company tied to its El Paso data center. The company holds 80% ownership by a fund managed by BlackRock. The bond issuance was led by Morgan Stanley and JPMorgan Chase and matures in 2048.

Sources said the bonds yielded about 2.875 percentage points above the 10-year U.S. Treasury, reaching approximately 7.5%, about 0.5 percentage points higher than Meta’s similar Louisiana data center financing last year—indicating investors now demand a higher risk premium.

Market participants expect Meta to launch more such deals. The company has been locking in data center capacity early and has been issuing tenders for land, energy, and other AI infrastructure projects.

The Wall Street Journal original: AI data center financing is getting more expensive—just look at Meta

Dina Powell McCormick, a former Goldman Sachs executive and former Trump administration official who joined Meta in January as President and Vice Chair, is leading this massive financing initiative.

Sources said she has met in recent months with heavyweight investors including BlackRock Chairman Larry Fink, Blackstone CEO Stephen Schwarzman, President Jon Gray, and Brookfield CEO Bruce Flatt to discuss funding sources for AI infrastructure.

She currently co-leads Meta’s Compute division with Daniel Gross and Santosh Janardhan—a team established by CEO Mark Zuckerberg to plan the company’s massive future computing capacity—and works with CFO Susan Li on financing strategy.

From zero debt to aggressive bond issuance—Meta’s debt has rapidly increased within a year. In the past, Meta’s highly profitable ad business generated ample cash flow, and the company had almost no need to borrow—only issuing corporate bonds for the first time in 2022, a decade after its IPO.

But over the past nine months, Meta’s financing pace has clearly accelerated. In October last year, the company issued $30 billion in corporate bonds, roughly doubling its total debt. In April this year, it issued another $25 billion in new bonds to fund AI investments.

In addition to direct bond issuance, Meta uses joint venture structures to reduce balance sheet liabilities. For example, in Louisiana’s Hyperion data center, Meta partnered with Blue Owl Capital to form a joint venture. A fund under Blue Owl invested about $3 billion for an 80% stake, and the holding company Beignet Investor issued $27 billion in bonds to fund the construction of a roughly 2GW-scale data center.

Moreover, Meta provides residual value guarantees for these bonds, promising to cover investor losses if leases are terminated early or not renewed. As a result, both financing deals received investment-grade ratings—Sopaipilla earned an A+ from S&P and AA- from Fitch.

Companies prefer to pay more rather than delay construction. Beyond Meta itself, data center developers are also increasing borrowing to meet rapidly growing AI demand.

Meta recently signed a long-term lease to occupy a large data center in Shippingport, Pennsylvania, with over 3GW of capacity. The project was developed by Aligned Data Centers, which was recently acquired for $20 billion by a consortium led by BlackRock. Sources said multiple banks are providing about $10 billion in project financing for the deal.

It remains unclear when interest rates will fall, forcing large AI companies to reassess overall infrastructure costs. Just as surging demand for construction materials and semiconductors has driven up data center construction costs, the massive demand for capital is now pushing up overall financing rates.

However, the market widely believes tech giants have no intention of slowing down. As the AI race intensifies, most companies prefer to accept higher financing costs rather than delay construction by waiting for lower rates, ensuring their computing capacity deployment stays ahead.