Three investors sent me the same message this week: “Are you worried?”

In January’s issue of Disruption Investor, we predicted another DeepSeek moment would shock Wall Street and send artificial intelligence (AI) stocks tumbling.

It was our top prediction for 2026.

And last week, it came true.

• A Chinese AI lab released Kimi K3.

It matched America's best models and topped them in blind coding tests, at roughly a third of the price. The real kicker? Kimi K3 is an open-source model. No $200-a-month subscription needed.

Wall Street took one look and hit the sell button on AI stocks.

The popular Roundhill Memory ETF (DRAM), which houses memory chip stocks, plunged more than 30% from its high.

CoreWeave (CRWV), the largest dedicated AI cloud, has been cut in half. And on Friday, Nvidia (NVDA) briefly lost its crown as the world’s most valuable company to Apple (AAPL).

• The logic behind the selloff goes like this…

If a Chinese lab can give AI away for free, nobody will pay American labs for it.

If nobody pays for AI, the roughly $750 billion Microsoft (MSFT), Google (GOOGL), Amazon (AMZN), and Meta Platforms (META) plan to spend this year on AI infrastructure is the biggest capital bonfire in history.

So sell the chips. Sell the data centers. Sell the big tech stocks. Sell anything even remotely related to AI.

I believe this will be the most expensive mistake investors make in 2026.

• January 27, 2025…

A little-known Chinese lab called DeepSeek released R1, a frontier-grade AI model it claimed to have trained for less than $6 million. Pocket change next to the billions American labs were spending.

The market concluded America was massively overspending on AI. Nvidia fell 17% in a single day and lost $600 billion worth of market value. It was the biggest one-day loss for any company in stock market history.

It looked like the AI spending boom was over. Can you guess what happened next?

Nvidia’s revenues more than doubled.

Two years ago, Google processed less than 10 trillion tokens a month. Today: 3.2 quadrillion. That's a thick novel's worth of AI text for every human alive, every month.

Anthropic went from $1 billion to $47 billion in annualized revenue in 17 months. No company in the history of capitalism has grown that fast.

Cheap intelligence ended up supercharging AI demand.

• If you sold on DeepSeek Monday, you sold the AI trade at the exact moment its fundamentals went vertical.

Here’s AI poster boy Nvidia since then:

The people selling this week are making the same mistake for the same reason.

• English economist William Stanley Jevons noticed something odd…

Steam engines were getting far more efficient, squeezing more work out of every lump of coal. Coal demand should have fallen.

Instead, it exploded. Cheaper steam power meant steam power was suddenly worth using everywhere. Total coal burned went up, not down.

The same thing happened with light. LED bulbs cut the cost of lighting by roughly 90%. So we lit up everything.

Economists call it Jevons Paradox. Make a useful thing cheaper, and the world finds a thousand new uses for it and spends more.

Same deal with AI.

GPT-4-level intelligence cost about $20 per million tokens in late 2022. Today, the same intelligence costs about 40 cents!

Did AI spending fall? Nope.

Uber’s (UBER) CTO recently admitted it burned through its entire 2026 AI budget in four months. Each Uber engineer now runs up $500–$2,000 a month in AI bills. Apple lets engineers burn $300 a day in tokens. The most AI-intensive companies now spend $7,500 per employee every month on AI.

I’m personally spending more than $200 per month on Claude, and it’s worth every penny.

This is the Jevons Paradox in action. The cheaper AI gets, the more we use it.

• Kimi K3 may be open-source…

But running it still requires expensive data centers filled with Nvidia GPUs and high-bandwidth memory from companies like Micron Technology (MU).

It also needs megawatts of electricity to power everything. Kimi K3's “recipe” will be free to download starting Monday. The kitchen you need to cook it—1.5 terabytes worth of scarce high-bandwidth memory across racks of Nvidia GPUs—is not.

Don't take my word for it. Within days of launching K3, the company behind it, Moonshot, stopped accepting new members.

Why? It ran out of GPUs. The lab that just “killed” demand for AI… can't get enough compute to serve its own free model.

So no, I’m not worried about this selloff.

I’m treating it the way DeepSeek should have been treated: as a gift. A chance to buy the toll roads of the AI boom at discounted prices—the bottlenecks that get paid no matter whose model tops the leaderboard.

Compute. Power. Memory. Chip-making machines. And all the other critical layers of AI infrastructure we’re investing in inside Disruption Investor and Disruption_X.

I suggest you do the same.

Stephen McBride

Chief Analyst, RiskHedge