AI Stocks May Be Whistling Past the Graveyard, And the Graveyard Is Getting Crowded

The market is up. Not all of it. Just the parts that matter. The parts that people talk about at dinner. The parts that make the numbers on the screen go green. Nvidia is still a five-trillion-dollar company. Broadcom is still a darling. Oracle still has a story. You can hear the whistling from here. It’s a cheerful tune. It’s the tune you whistle when you walk past a graveyard at night. You don’t look at the headstones. You keep your eyes forward. You keep your pace steady. The whistle is not for the dead. It’s for you.

The thing about the graveyard is that it’s patient. It doesn’t mind the whistling. It’s been there a long time. It will be there when the tune changes.

---

What the Whistling Sounds Like

The Numbers That Don’t Add Up

You don’t need to be a quant to see it. You just need to read. Worldwide AI spending is projected to hit $2.7 trillion in 2026. That’s a 49.5% jump from the year before, according to Gartner. The five biggest hyperscalers, Microsoft, Alphabet, Amazon, Meta, and Oracle, are on track to spend $688 billion on capex in 2026 and $870 billion in 2027. That’s real money. It’s money that has to come from somewhere.

And it is coming from somewhere. That’s the problem.

Capex vs. Revenue

Microsoft’s AI revenue for fiscal year 2026 was $34.33 billion. That’s a number. It’s not a small number. But it’s a number that sits next to a capex figure that dwarfs it. The hyperscalers are shifting from asset-light to asset-heavy. Their capital expenditure as a proportion of sales is going from 10–20% in 2024 to above 30% by 2028, according to UBS. That puts them in the same neighborhood as telecoms and utilities. Utilities are not known for their thrilling growth stories. They are known for their debt.

The Circular Financing Loop

Here’s where the whistle gets a little shaky.

Nvidia invested up to $100 billion in OpenAI. OpenAI committed to filling data centers with Nvidia chips. Nvidia then worked with six Wall Street firms to mobilize more than $500 billion of third-party capital for AI infrastructure. Critics call it what it is: circular financing. Nvidia helps fund the customers who buy its chips, which in turn lifts its own revenue.

It’s a beautiful loop. It works as long as everyone keeps their footing. The Atlanta Fed described it in a September 2026 paper. Circular financing increases the likelihood that many AI firms will simultaneously suffer from an adverse shock. A harmful shock to one or a few companies can spread quickly to numerous AI firms and potentially the broader economy.

That’s not a whistle. That’s a warning horn. But it’s buried in a PDF, and the market is busy whistling.

---

The Warning Signs Nobody Wants to Read

Credit Default Swaps Are Widening

This is the quiet one. The one that doesn’t make headlines. Credit default swaps for major tech names, Nvidia, Broadcom, Oracle, Meta, are widening. Broadcom and Oracle stocks have declined in line with rising CDS spreads. If the spreads continue to widen, further equity repricing is likely, especially for Oracle and potentially Nvidia.

CDS spreads are not a narrative. They are not a story about the future. They are the cost of insurance against default. When that cost goes up, someone is worried. Someone is looking at the graveyard and seeing something move.

The Circular Financing Trap

The BIS weighed in. Circular financing entails chip producers and hyperscalers taking equity positions in AI labs and neocloud providers. In turn, those labs and providers agree to multiyear commitments to purchase chips or computing power. The interconnections increase the likelihood of simultaneous shocks. Due to circular financing, a harmful shock to one or a few companies can spread quickly.

It’s not a conspiracy. It’s a structure. And structures have failure points.

The Revenue That Isn’t There Yet

OpenAI and Anthropic are responsible for approximately 50 percent of the revenue backlogs for the biggest cloud providers, Microsoft, Oracle, Alphabet, and Amazon, over the next five to ten years. Neither OpenAI nor Anthropic is yet profitable. They regularly suffer large losses.

If those spending commitments don’t materialize, the hyperscalers face significant declines in revenue and projected earnings growth. The whistle doesn’t change that. The graveyard doesn’t care about your revenue backlog.

---

The Anthropic Moment

When the Music Changed

In early February 2026, Anthropic, a competitor to OpenAI, presented its AI agent for legal work. Other agents followed. Investors, surprised by the capability, began to sense how much office work AI could destroy. They cut their exposure to knowledge-based service providers. The Korean Kospi index, heavy in hardware, outperformed the S&P 500 by more than 20% in a month. The S&P 500 was dragged down by losses in service-sector stocks that had been buoyed by inflated valuations of intangible assets.

The fear of an AI bubble manifested in a simple, violent rotation between AI sub-sectors. The fear of a stock market bubble gave way to the fear of macroeconomic disruption.

The Rotation That Shook the Market

The stocks of the data-center boom’s key players, Google, Meta, Microsoft, Nvidia, and Oracle, all declined after peaking in late 2025 or January 2026. Meta and Microsoft lost 16% and 25% in value, respectively. Nvidia’s price-earnings ratio declined to pre-ChatGPT levels, cutting more than $800 billion of market capitalization. Oracle’s market capitalization shrank by 47%. Microsoft’s stock declined by 25% in the first quarter of 2026, its worst quarterly decline since 2008.

These are not small numbers. These are the numbers of a market that is beginning to look at the headstones.

What It Means for the Rest of Us

The AI boom was never just about AI. It was about cheap money and a good story. When money stops being cheap, the story has to work harder. Rising interest rates reduce the ability to finance the AI expansion. Government regulators restricting AI development could also trigger a downturn. Underwhelming productivity gains could do the same.

The whistle gets harder to hear when the wind picks up.

---

The Graveyard Is Not Empty

Who Gets Buried First

The most vulnerable are not the hyperscalers. They have cash. They have scale. They have the ability to absorb losses. The most vulnerable are the companies that borrowed against the promise of AI revenue that hasn’t arrived. The neocloud providers. The startups with burn rates that outpace their fundraising. The legacy SaaS firms that need to change fast to meet the new competitive challenges coming from LLMs.

They are the ones who will be buried first. The whistlers will walk past their graves without breaking stride.

The Off-Balance-Sheet Problem

The BIS also points out that hyperscalers increasingly outsource data center construction to third parties using off-balance-sheet entities. Those entities then lease the facilities back to hyperscalers via long-dated contracts that often lock in costly terms for exiting the agreements. The terms of circular financing and outsourcing deals are often opaque.

Opacity is not a bug. It’s a feature. It allows the picture to look better than it is. It allows the whistle to stay cheerful. But opacity doesn’t change the underlying arithmetic. It just delays the moment when the arithmetic has to be faced.

---

What the Whistlers Miss

The Real Economy Doesn’t Care About Your Narrative

You can tell a story about AI changing the world. You can talk about productivity. You can talk about the future. The real economy doesn’t care. It cares about power, rates, and the cost of money.

There isn’t enough power for all the data centers under contract. There might not be enough paying customers to support all the data centers. Rising interest rates stay high as AI growth slows. Earnings estimates fall across the AI supply chain. The economy is not a narrative. It is a series of constraints.

Power, Rates, and the Cost of Money

Morgan Stanley estimates that from 2025 through 2028, tokens per watt will rise almost 6x. Tokens per dollar of electricity will rise more than 5x. Tokens per chip will rise almost 15x. That sounds like progress. It is progress. But progress has a cost. It requires more chips, more power, more infrastructure. And all of that requires money.

When money is cheap, you can build. When money is expensive, you have to justify. The justification is getting harder. The revenue is not keeping pace. The whistle is getting strained.

---

The Whistle Stops Eventually

The graveyard is patient. It doesn’t mind the whistling. It has heard it before. It heard it in 1999. It heard it in 2007. It hears it now.

AI stocks may be whistling past the graveyard. That doesn’t mean they’re doomed. It doesn’t mean the graveyard is full. It means the whistle is a choice. It means someone is choosing to look forward instead of sideways. Someone is choosing to hear the tune instead of the silence.

The silence is there. It’s always there. You just have to stop whistling long enough to hear it.