The AI "arms race" among tech giants is evolving into a complex financial game.

When annual capital expenditures of hundreds of billions of dollars become the norm, even companies with cash reserves exceeding $340 billion like Amazon, Google, Meta, Microsoft, and Oracle are beginning to feel unprecedented financial pressure.

They are breaking away from the traditional approach of relying entirely on internal funds to build infrastructure, instead turning to Wall Street for more complex financial solutions, attempting to provide ammunition for this expensive competition without compromising their financial stability. However, risks are emerging alongside these strategies.

**Trillion-Dollar Giants' "Sweet Burden": AI Infrastructure Costs Drive Financing Innovation**

In the past, tech giants were accustomed to using their massive internal cash flows to build data centers, but the rise of AI has completely changed the game rules.

The speed and scale of this competition force them to seek external capital. Investors and credit rating agencies are closely monitoring how these tech giants will pay for AI data centers and whether these massive investments can be converted into new revenue streams.

To maintain healthy financial statements while pursuing aggressive expansion, tech giants have begun collaborating with bankers to design increasingly complex financial strategies with one core objective: transferring part of the costs and risks off their balance sheets.

Against this backdrop, financial instruments rarely heard of in the tech sector—such as joint ventures, backstop agreements, and syndicated debt—are being brought to the table.

**Three Financial "Strategies" for Risk Distribution**

In exploring the "externalization" of risks and costs, three innovative financial "strategies" have emerged, all centered on cleverly "externalizing" risks and liabilities.

**1. Meta's "Off-Balance-Sheet" Strategy: Joint Ventures**

Meta initiated a $29 billion financing for its data center project named "Hyperion" in Louisiana.

The core structure involves establishing a joint venture with investment company Blue Owl Capital. Blue Owl contributes $3 billion in equity, while the project's required $26 billion in massive debt is syndicated by bond giant Pimco with assistance from Morgan Stanley.

The key to this structure is that Meta will repay the debt in the form of lease payments in the future, thereby moving the entire project off its balance sheet and controlling debt levels.

**2. Oracle's "Risk Sharing": Syndicated Loans**

As the world's fourth-largest cloud service provider, Oracle recently agreed to become a tenant of a 1.4GW data center complex being developed by Vantage Data Centers, one of the largest projects under construction globally.

Due to the project's massive scale, developer Vantage is collaborating with six banks led by JPMorgan Chase and Mitsubishi UFJ Financial Group, including Goldman Sachs, to syndicate the required $22 billion in debt.

This model disperses risk among multiple lenders, reducing individual institutions' risk exposure and making massive financing possible.

**3. Alphabet's "Sophisticated Design": Backup Guarantee**

Alphabet's approach is the most complex and sophisticated: a "backup guarantee."

In this transaction, Alphabet provided up to $3.2 billion in backup guarantees for the lease contract between cloud startup Fluidstack and data center owner TeraWulf, obtaining a 14% stake in TeraWulf.

The sophisticated design lies in the fact that this guarantee is a "contingent liability" that only triggers if Fluidstack defaults, so Alphabet likely doesn't need to include it in current liabilities.

With Alphabet's support, TeraWulf raised $1 billion last month through convertible bonds underwritten by Morgan Stanley and Cantor Fitzgerald, more than double its initial financing target.

TeraWulf's Chief Financial Officer stated at a meeting last month:

"It's not easy to get a $2 trillion company, their management team, board, and everyone to agree to a novel concept, but I hope we've provided a roadmap."

**Hidden Concerns Under the Frenzy: Overheating, Concentration, and Leverage Risks**

Tech giants' massive financing needs coincide with a credit market flush with cash.

Private credit funds holding hundreds of billions in investment capital, along with banks increasingly comfortable with projects having "investment-grade" tenants, are actively entering the market. Loan-to-total-cost ratios for data center projects have risen significantly compared to the past.

Jason Tofsky, Global Head of Digital Infrastructure Banking at Goldman Sachs, stated that lending institutions agree to provide 80% to 90% of total project costs for data center projects. According to data from real estate company Jones Lang LaSalle, data center lenders typically provide 65% to 80% of total costs for new development projects. Tofsky noted:

"There's sufficient capital in the market to fund projects the market is familiar with. The market can digest these projects well."

However, capital frenzy is breeding new risks.

First is market overheating risk. UBS Group AG analysts warned in a report last month that while massive private credit inflows into the data center sector could drive AI development, they might also "increase the risk of market overheating."

Second is high concentration risk. Data center lease contracts are highly concentrated among a few creditworthy tech giants. This raises concerns that if any of these companies reduces spending due to strategic adjustments or faces credit rating impacts, the entire ecosystem could face enormous risks.

Finally, leverage risks for some companies are already apparent. Rating agencies Moody's and S&P warned Oracle in July that its leverage ratio (currently 4.3x) is much higher than other "hyperscale" providers as it enters the AI infrastructure construction phase. If it doesn't reduce its debt-to-earnings ratio below 3.5x, its credit rating faces potential downgrades.

Moody's analysts wrote in a credit report:

"While several other hyperscale providers are building AI infrastructure, none entered this phase with leverage as high as Oracle's, nor with cash flow as negative."