AI Giants Are Bleeding Money—Why Oracle and Broadcom's Latest Numbers Should Make You Think Twice

The AI boom narrative is cracking, and the cracks are showing up in some unlikely places. After years of unflinching enthusiasm about artificial intelligence stocks, even Wall Street is starting to ask uncomfortable questions: What if all these massive investments don’t actually pan out?

The Red Flags Keep Multiplying

Let’s start with Oracle, which delivered what many considered a stellar earnings quarter back in September. The company reported over $450 billion in remaining performance obligations fueled by its fast-growing AI cloud services business. Investors loved it—the stock shot up 40%. But here’s where the story gets interesting.

Just months later, the same company that looked unstoppable started showing cracks. Oracle’s November earnings report revealed revenue of nearly $16.1 billion, which missed consensus estimates. More concerning: the company reported a staggering negative $10 billion in free cash flow for the quarter. The market didn’t take kindly to this, and the stock declined.

But the real eye-opener came when media reports surfaced that Oracle would need to raise $38 billion in debt just to build the data centers necessary to meet demand. Even worse, The Information revealed that Oracle’s AI data center business, despite explosive growth, is only generating profit margins between 10% and 20%—far below what investors had hoped for.

The fallout was swift. The price of five-year credit default swaps on Oracle’s debt (essentially insurance against default) hit record highs. This isn’t just a stock price issue; it’s a credit concern.

Broadcom’s Numbers Can’t Hide the Underlying Problem

When chipmaker Broadcom reported earnings for its fiscal Q4 2025, investors initially seemed pleased—numbers came in above estimates with solid forward guidance. Then came the guidance: the company projected its gross margin would decline next quarter. Not expand. Decline. The market also felt that Broadcom’s AI product backlog was significantly understated.

Two major AI infrastructure plays, both stumbling on margin and growth narratives simultaneously? That’s not random. That’s a pattern.

The Math Doesn’t Add Up—And Even Big Tech Admits It

IBM’s CEO Arvind Krishna recently made a startling statement that captures the core problem: there is “no way” the hyperscalers will see meaningful returns on their staggering AI infrastructure spending.

Here’s the math: approximately $80 billion is needed to build a single 1-gigawatt data center. If a company commits to 20-30 gigawatts—which several are doing—that’s $1.5 trillion in capital expenditures from a single entity.

Oracle itself raised its full-year capital expenditure guidance from $35 billion to $50 billion. That’s a 43% increase. Broadcom, Nvidia, and others are on similar trajectories. When you layer in the resource constraints—the sheer power demand required to run these data centers, the water needed for liquid cooling solutions, the geopolitical complexities of building infrastructure globally—the feasibility question becomes harder to ignore.

The Uncomfortable Question Investors Need to Ask

Nobody disputes that artificial intelligence can reshape industries. The skepticism isn’t about AI’s potential; it’s about the price tag and timeline for realizing that potential.

The hyperscalers are collectively projected to spend trillions in coming years on AI-related capital expenditures. Meanwhile, profit margins on these massive infrastructure projects are coming in at 10-20%, companies are taking on unprecedented levels of debt, and free cash flow is turning negative even as growth accelerates.

It bears repeating: the internet bubble had to pop before the internet actually changed the world. We’re potentially at a similar inflection point with AI.

What Should You Do?

If you’re holding AI infrastructure stocks or companies heavily exposed to data center buildouts, this is a good moment to reassess. Look at the valuations. Examine the assumptions underlying those valuations. Run the numbers on return on invested capital—not just top-line growth, but actual returns.

If the returns seem too good to be true, they probably are. That might mean trimming some positions and taking gains off the table. The upside opportunity in AI is real, but so are the execution risks and capital constraints that are now becoming impossible to ignore.

The warning signs are there. The question is whether you’ll listen.