Select a result to preview

DL 414

Published: November 30, 2025 • 📧 Newsletter

Bubbles Go Pop

The conversation about AI has shifted from "is there a bubble?" to "how does it burst?" But this isn't just financial speculation. It's about what we're actually building and who's paying for it.

If you've found value in these issues, subscribe here or support me here on Ko-fi.

🔖 Key Takeaways

📚 Recent Work

This week I published the following:

🫧 The AI Bubble: When Infrastructure Outlasts Hype

You've probably heard whispers about an AI bubble. This comes up often in discussions with friends. I even have students talking about it in classes.

I think it's rooted in skepticism that the massive investments in artificial intelligence will ever pay off. But this isn't like the dot-com crash, where overvalued websites simply disappeared.

This bubble is being built in concrete and copper. US companies are spending more on AI data centers than on the entire American power sector. Individual facilities consume as much electricity as entire states. And the bill—running into the hundreds of billions—is increasingly being financed through complex off-balance-sheet structures that keep the real debt hidden.

The question isn't whether there's a bubble. It's what happens when you build massive physical infrastructure based on speculative returns, and then the speculation fizzles. Unlike stocks that can drop to zero, data centers don't vanish. They become infrastructure we're economically obligated to use, whether they serve human needs or not.

🏗️ The Material Reality Behind the Hype

When we talk about an "AI bubble," we're not just talking about overvalued stocks. We're talking about physical infrastructure being built at unprecedented scale:

This matters because unlike previous tech bubbles where the "bubble" was mostly inflated stock prices, this one is being built in concrete, copper, and power grids. When it pops, those don't disappear. They become infrastructure we're obligated to use simply because we built them.

These aren't abstract investments on a balance sheet. Northern Virginia has become the data center capital of the world, where gigantic, power-hungry facilities sit steps away from people's homes. This investigation shows what happens when your neighborhood becomes the world's digital backbone: constant construction noise, increased power demands, and entire communities reshaped by infrastructure built for someone else's speculation.

💸 The Financial Engineering Tell

Here's where it gets revealing. In October, Meta structured a $30 billion data center deal using a special purpose vehicle that keeps $27 billion in debt off its balance sheet. This is a company sitting on $65 billion in cash.

Why use complex financing structures reminiscent of Enron instead of just building it directly? Because when you can't document genuine confidence in returns, you document increasingly sophisticated ways to hide the risk.

Amazon, Google, Meta and Microsoft are set to collectively spend around $400 billion on AI this year, with some companies devoting about 50% of their current cash flow to data center construction. That's not investment. That's all-in gambling.

🌍 Two Civilizational Bets

The US and China are making radically different wagers on the future.

The US Bet: AI at Any Cost

The US has gone all-in on artificial intelligence infrastructure. In the first half of 2025, AI-related spending accounted for 1.1% of all US economic growth. Contributing more to the economy than consumer spending did in the same period.

The stock market tells the same story. Roughly 80% of S&P 500 gains come from AI stocks. Strip out the AI companies, and the market is barely growing.

The US economy is increasingly dependent on AI investment to show any growth at all. Without AI companies and AI infrastructure spending, these economic numbers would look anemic. That's not a sign of strength. It's a sign of concentration and risk.

The China Bet: Green Technology at Scale

China made a different choice. Clean energy (solar panels, wind turbines, electric vehicles, and batteries) now accounts for over 10% of China's entire GDP. That happened for the first time in 2024.

And they're building at unprecedented speed: installing wind and solar capacity equivalent to five nuclear power plants every single week. The rest of the world combined isn't building half that much.

The Reality Check

Then in late January 2025, something happened that exposed the assumptions behind the US bet.

A small Chinese company called DeepSeek announced they'd built an AI model rivaling OpenAI's ChatGPT for under $6 million. OpenAI had reportedly spent $600 million on theirs.

The stock market panicked. Nvidia, the company selling the expensive chips everyone assumed you had to buy to build competitive AI, lost $600 billion in market value in a single day. The largest one-day loss in US stock market history.

Why it matters: DeepSeek proved you don't need to spend 100 times more to get comparable results. US sanctions had forced Chinese researchers to innovate with cheaper, older chips. Constraint bred efficiency.

The message was clear. Efficiency through constraint beats scale through extraction.

⚡ The Energy Collision

Here's where speculation meets physical reality. A 5-gigawatt data center consumes roughly 40 million megawatt-hours of electricity per year—as much as the entire state of Nevada or Kansas.

Data centers could consume nearly 3% of global electricity by 2030. In the US, projections suggest 6-12% of the nation's electricity will power AI infrastructure.

And AI electricity consumption is scaling faster than utility electricity capacity. The Berkeley Lab warns this could trigger blackouts as data center demand outpaces grid expansion.

Meanwhile, after years of renewable energy rhetoric, Big Tech is pivoting to nuclear power to fuel their data center ambitions. Microsoft, Amazon, and Google have all announced nuclear energy deals because the grid can't handle both their AI speculation and actual human needs.

This isn't abstract. It's a material question about power, literally and figuratively.

🤔 Consider

The AI infrastructure build-out is so gigantic that in the past 6 months, it contributed more to the growth of the U.S. economy than all of consumer spending.
— Christopher Mims

Two nations. Two infrastructure bets.

The US is building data centers financed through structures reminiscent of Enron. China is building solar capacity at a rate equivalent to five nuclear plants per week.

History suggests that when speculation drives infrastructure buildout at this scale, what gets built outlasts the beliefs that justified it. We're left with physical systems we're economically obligated to use. Whether they serve human flourishing or corporate extraction.

The sunk costs become a trap. The infrastructure itself creates the dependency. We built it, so we must use it. That's not technology serving human needs. That's humans serving technology's financial demands.

The question isn't whether there's a bubble. It's what infrastructure we're locked into when it pops, and whose vision of the future becomes materially embedded in copper, concrete, and power grids.

⚡ What You Can Do This Week

Previous: DL 413 • Next: DL 415 • Archive: 📧 Newsletter

🌱 Connected Concepts: