Highlights

• Nvidia's $4.5 trillion valuation and 160% annual growth faces skepticism from Michael Burry ($1.1B in puts) and SoftBank's $5.8B exit, signaling potential AI bubble concerns.
• Over $5 trillion in AI infrastructure spending is being financed through bonds, REITs, and structured products, creating systemic risk if demand or execution falters.
• Critical minerals investors should respect bubble risk while focusing on structural demand—AI corrections create temporary air pockets, but electrification and digitalization megatrends remain intact.

---

Call it whatever suits your temperament—the AI revolution, a productivity supercycle, or just a very shiny, very modern bubble—but the numbers have stopped whispering and started shouting. Nvidia’s market cap now sits around $4.5 trillion, making it not just the most valuable company on Earth but a financial object comparable in size to entire national stock markets. Its three-year average market-cap growth rate—about 160% a year—belongs more to myth than to mature equity markets.

Meanwhile, the skeptics are no longer anonymous Twitter eggs.

Michael “Big Short” Burry’s Scion Asset Management disclosed roughly $1.1 billion in put options tied to Nvidia and Palantir—about $912 million notional against Palantir and $187 million against Nvidia—directly targeting the AI leaders he believes are priced for disappointment. Shortly thereafter, Burry deregistered his fund altogether, effectively slamming the door behindhis warning.

On the other side of the Pacific, SoftBank has sold its entire Nvidia stake for about $5.8 billion, not because Masayoshi Son lost faith in AI, but because he wants even more exposure—plowing capital into OpenAI and other AI platforms instead.  It’s the kind of move that can be read in two ways: savvy capital rotation inside a genuine boom… or a sign that the smartest players believe Nvidia itself is now the air at the top of the souffle, not the batter.

Debt, Data Centers, and the AI Carry Trade

Beneath the stock charts, the AI build-out is very real—and very expensive. JPMorgan’s latest deep dive estimates over $5 trillion will be spent on global data centers, AI hardware, and associated power infrastructure over the next five years, with as much as $1.5 trillion needing to come from investment-grade bond markets alone.

That wall of money is already reshaping credit. Recent reporting via Data Center Dynamics (opens in a new tab) and others reveal that U.S. AI/data-center–linked investment-grade issuance has surged, with roughly $75 billion of deals in just two months—about 5% of all U.S. IG issuance for 2025.

Asset-backed securities tied to data-center leases and digital infrastructure are proliferating. Private credit funds are being pitched as bespoke financiers to hyperscalers. The AI story is no longer just about overheated equities; it is woven into bond indices, structured products, REITs, utilities, and even sovereign power-investment plans.

Burry’s underlying fear—echoed quietly by more sober analysts—is straightforward: AI chips and data centers depreciate at warp speed. If end-demand, power capacity, or regulatory tolerance underwhelm the narrative, you’re left with trillions in sunk capex funded with long-dated debt and only PowerPoint slides to show for it.

Bubble Talk vs. Boring Reality

Some of the online commentary around AI—Nvidia “devouring the U.S. economy,” imminent systemic collapse—is clearly overcooked and should be received with skepticism. Nvidia is not literally consuming entire sectors, but it is priced for near-perfect execution: relentless revenue growth, sustained pricing power, and no serious competitive or regulatory shock.

Likewise, it’s not true as some anti-U.S. promoters (opens in a new tab) allege that AI-related bond issuance suddenly erupted to $130 billion in a matter of weeks; the actual figures are lower and more nuanced. But the direction of travel is not in dispute—AI infrastructure is being financed through every available channel, and the bill is large enough to leave a mark if expectations crack.

The uncomfortable, less viral truth looks like this:

• AI is not a fad. It is already changing code, chip design, logistics, and white-collar workflows.
• But that does not guarantee today’s winners deserve their current valuations, or that today’s capex pace is sustainable. Markets can overpay for real revolutions; ask anyone who bought fiber-optic stocks in 1999.

Defense, Not Drama: How to Invest When Everything Screams “AI”

This is where Rare Earth Exchanges plants a flag: ignore the YouTube apocalypse thumbnails and focus on positioning instead of prophecy.

First, separate the AI theme from the AI tickers. Nvidia, the hyperscalers, and a handful of glamour names may be priced as if every scenario is up-and-to-the-right forever. But the AI ecosystem is wider: power infrastructure, copper, rare-earth permanent magnets, grid hardware, cooling systems, and industrial software are all essential and often trade on more pedestrian multiples. The narrative premium is not evenly distributed.

Second, watch the credit plumbing. When credit spreads for heavy AI spenders widen, or when data-center REITs and utilities start to pay more to roll their debt, that’s the market whispering that the AI carry trade is getting crowded. Equity often listens late; bond investors are the nervous canaries.

Third, demand cash flow, not just “total addressable market” poetry. For AI-exposed names—whether chipmakers, data-center landlords, or software platforms—look at free cash flow after capex. Ask the brutal question: If they were forced to stop building tomorrow, would this still be a good business or just a half-finished monument to hype?

Finally, size positions as if the bubble might burst. Keep AI high-beta stocks as part of a diversified portfolio, not its spine. If you need leverage to make the trade “work,” you’re probably already too close to the edge.

Where Rare Earth and Critical Mineral Investors Actually Fit

For Rare Earth Exchanges readers, there’s an extra layer. AI is hardware and power-hungry. It pulls directly on NdPr magnets, Dy/Tb for high-temperature motors, copper for transmission, high-grade steel, and advanced semiconductor supply chains. A sharp AI correction could trigger a temporary demand air pocket, pressuring some producers. But the structural megatrends—electrification, automation, and digital infrastructure—do not vanish with a lower Nvidia P/E.

That means: don’t chase anything with “AI” in the ticker at any price—but also don’t confuse a correction in AI equities with the death of long-term demand for critical minerals. Focus on well-capitalized, technically credible projects and midstream refiners that can survive a cyclical wobble in data-center capex and still be standing when the next, more rational phase of the AI build-out begins.

In other words: respect the bubble risk, harvest the structural trend, and insist on balance sheets that can outlive the hype cycle.

© 2025 Rare Earth Exchanges™ – Accelerating Transparency, Accuracy, and Insight Across the Rare Earth & Critical Minerals Supply Chain.