The $4 trillion AI hype train is running on borrowed money

Companies are borrowing billions to build artificial intelligence (AI) infrastructure even though recent studies show just 3% of consumers pay for AI services, creating a dangerous gap between spending and demand.

Oracle exemplifies this risky bet. The database company signed a $300 billion contract with OpenAI this month, forcing it to borrow an estimated $25 billion yearly for the next four years. Oracle already carries $82 billion in long-term debt.

OpenAI needs to grow from $12 billion to more than $300 billion in annual revenue by 2030 to justify Oracle’s spending, according to D.A. Davidson analyst Gil Luria.

“A vast majority of Oracle’s data center capacity is now promised to one customer, OpenAI, who itself does not have the capital to afford its many obligations,” Luria said.

This build-first strategy spans the entire tech sector. Nvidia CEO Jensen Huang estimates companies will spend between $3 trillion and $4 trillion on AI infrastructure by 2030. President Trump referred to the Stargate project, a $500 billion joint venture among SoftBank, OpenAI, and Oracle, as “the largest AI infrastructure project in history.”

Smaller companies pile on more risk through creative financing. CoreWeave borrowed against an $11.9 billion OpenAI contract. Nebius Group did the same with a $17.4 billion Microsoft deal. Both companies bet their futures on contracts with companies that haven’t proven they can pay.

The borrowing spree marks a shift from the early AI boom. Tech giants initially funded AI development with cash from their advertising and cloud businesses. Now, companies are leveraging themselves heavily, hoping demand will catch up.

Oracle’s debt-to-equity ratio sits at 450%. Google-parent Alphabet’s ratio is 11.5%. Microsoft’s is 33%.

Real-world impacts mount alongside the debt. Meta’s planned $600 billion infrastructure spending includes a Louisiana data center requiring nuclear power. Elon Musk’s xAI facility in Tennessee became one of the county’s largest pollution sources, experts say.

Moody’s gave Oracle’s credit rating a negative outlook in July. The ratings agency warned that regardless of financing method, “the overall growth in balance sheet obligations will also be extremely large.”

The bet assumes AI will transform every business sector. Companies race to secure their positions before competitors lock up capacity. They’re building tomorrow’s infrastructure with today’s borrowed money.

Yet consumer adoption crawls along. Most people don’t pay for ChatGPT or similar services. Businesses experiment with AI but haven’t deployed it widely enough to justify trillion-dollar investments.

If demand doesn’t materialize, contracts could be postponed or renegotiated. Oracle might lease its infrastructure to other companies if OpenAI falters. But someone still needs to pay for all this computing power.

The AI infrastructure boom looks different from the dot-com bubble in one key way. Twenty-five years ago, companies built fiber optic networks that eventually found users. Whether AI infrastructure finds its customers remains an open question worth $4 trillion.