AI Selloff: Why the Demand Data Still Looks Strong

What did I talk about the AI selloff?

In my video published Thursday, November 5, 2026, I presented that the selloff in AI stocks wasn't evidence that AI demand was disappearing. The market was questioning how much investors should pay for that demand.

A viewer had just commented that the AI trade was over. I disagreed, and the rest of the video was about why, using numbers instead of headlines.

Is politics going to shut down existing data centers?

In the video, I said no, and that the panic was mixing up two different things. Restricting new data center development isn't the same as shutting down facilities that already run. If new capacity gets harder or more expensive to build, existing sites with power, permits and paying customers could become more valuable.

"There's a major difference between restricting new developments and shutting down facilities that are already operating."

The political risk is real. I cited the National Conference of State Legislatures: 16 states had considered moratorium legislation affecting data center development. But the same group tracked 38 states offering tax incentives to attract data centers.

It also isn't one party. Texas Governor Greg Abbott, a Republican, had halted new data center permits pending reviews of power demand, water and grid reliability. After the election, the incoming Democratic governor proposed requirements on union labor, electricity costs and water. Tougher rules aren't a switch that turns off the grid.

Is AI demand a bubble?

Not according to the numbers I walked through in the video. CBRE data showed vacancy across eight major North American data center markets at just 1.4%, even though supply grew almost 34% in one year. New capacity was getting absorbed about as fast as it was built. That isn't what disappearing demand looks like.

Asking rents for certain deployments rose 8.3%, and Northern Virginia, the world's largest data center market, had vacancy of just 2%. Of the capacity under construction, 80.4% was already pre-leased. I also cited the International Energy Agency: global data center electricity use was about 485 terawatt hours in 2025, with demand projected around 950 by 2030.

Those CBRE figures cover eight markets, not the whole industry, and strong demand doesn't guarantee every company makes money:

"AI infrastructure demand is still strong. The real question is whether the companies that are building it can turn that demand into profits."

Why doesn't announced capacity equal a working data center?

Because announcing capacity and running profitable capacity are two completely different things. A real data center needs land, permits, financing, equipment and a grid connection, then customers paying enough to cover operating costs, financing and depreciation. A lot of investors buy the announcement.

AI chips, memory, cooling and transformers can take months or years to arrive while everyone competes for the same supplies. And the biggest question is simple: do you actually have the electricity to turn it on?

That cuts both ways. The same bottlenecks that make developers risky also make it hard for anyone to flood the market with new supply.

Why are power stocks falling if AI needs so much electricity?

In the video, I called this one mind-blowing. AI can't run without electricity, yet established producers were selling off alongside the speculative names. Constellation Energy was down roughly 38% from its 52-week high and Vistra about 43%, even though both own operating plants generating power today.

Many of these stocks had big rallies that priced in years of future demand. Higher rates cut the returns on billion-dollar power projects. And if hyperscalers slow their buildouts, expected demand could take longer to show up.

Some speculative nuclear names were down more than 80% from their highs. The market was lumping everything together. A company earning profits from electricity today isn't the same as one promising a reactor five or ten years from now.

What did I say about semiconductors and Amazon?

In the video, I said semiconductors were my preferred area for selective buying in the dip. I cited the Semiconductor Industry Association: global chip sales reached about $792 billion in 2025, up more than 25% from the prior year, with growth continuing into 2026. Customers hadn't stopped buying chips.

That pace won't necessarily last. Semiconductors are cyclical and supply eventually catches up. But we're still early in companies working AI into automation, software, healthcare, financial services and manufacturing.

I also said Amazon was moving toward the top of my watch list because of its AI infrastructure business and custom chips. I wasn't trying to call the bottom. Our breakdown of why Amazon stock jumped during the AI rout covers an earlier swing.

What's the real risk in all this AI spending?

The companies paying for it. Microsoft, Amazon, Google, Meta and other big tech companies are spending enormous sums on AI capacity, and Wall Street is fairly asking when that turns into cash flow. If they decide they've built enough, chip orders, leasing and power demand could all slow.

But these aren't startups borrowing to survive, which is why I stayed bullish on profitable hyperscalers. Look at how Amazon's free cash flow narrows in the most recent year as spending ramps.

According to the company's annual filings, Amazon generated $139.51 billion of operating cash flow in fiscal 2025 but spent $131.82 billion on capital expenditures. That's the gap the market is asking about. Our guide to free cash flow and why investors trust it explains why that line matters.

What would change my mind?

In the video, I said I'm bullish on AI but not blindly bullish. Rising data center vacancy and falling rents would be a warning sign. So would chipmakers reporting declining margins and repeated cuts to earnings forecasts. Neither had shown up in the data I walked through.

"But slowing growth and collapsing demand are two completely different things."

Financing at developers like IREN is the other watch item: costlier debt, customers cutting prepayments, or large stock issuance to keep expanding. IREN's diluted share count went from 99.6 million in fiscal 2024 to 316.1 million in fiscal 2026, and debt-to-equity rose to 1.87 from 0.53 a year earlier. Our guide to reading the debt-to-equity ratio puts that in context, and you can run the free IREN Stock Audit for the full picture.

What was I buying after the selloff?

Four areas, in order: semiconductors, then existing electricity producers like Constellation and Vistra, then profitable hyperscalers like Amazon, then data center developers like IREN and APLD (Applied Digital). For the developers, I saw the opportunity but also execution risk in financing, equipment costs and construction timelines.

On chips, I used an ETF to spread the risk instead of guessing which single stock recovers first. When I made the video on November 5, 2026, I owned shares of Micron, the SOXX semiconductor ETF, IREN and APLD. I may have bought or sold since.

"I don't believe we're watching the end of the AI boom. I believe we are watching the market question how much investors should pay for that boom."

What's next for IREN and the AI trade?

The nearest catalyst was IREN's earnings report, due the evening the video came out, Thursday, November 5, 2026, with financing terms the key thing to watch. Beyond that, the signals to track are data center vacancy, rents and chipmaker margins.

Going into that report, IREN's latest quarter showed revenue down 26.7% from a year earlier, and the consensus estimate for next fiscal year was a loss of $3.98 per share. Its beta of 4.33 explains a lot of the swing. Our guide to what beta measures shows why a stock like that can fall 76.4% in a year while the S&P 500 rises 8.2%.

Frequently Asked Questions

What is data center vacancy, and why does it matter for AI stocks?

Vacancy is the share of built data center capacity that isn't leased to a customer. Low vacancy means tenants are absorbing space as fast as it comes online, which supports rents and gives operators pricing power. Rising vacancy would signal that supply is outrunning demand. That's why it's one of the cleanest real-world checks on whether AI spending is turning into paying customers rather than just announcements.

What does pre-leased mean for a data center under construction?

Pre-leased capacity is space a customer has already contracted to rent before the facility is finished. For a developer, it lowers the risk of building something nobody uses, and it can make financing easier because lenders can see future rent. It doesn't remove execution risk, though. The building still has to get power, equipment and permits on schedule for that lease to turn into revenue.

Why does share dilution matter for data center developers like IREN?

When a company issues new stock to fund growth, each existing share owns a smaller slice of the business. According to IREN's annual filings, its diluted weighted share count went from 40.9 million in fiscal 2022 to 316.1 million in fiscal 2026. Growth funded that way only helps shareholders if profit per share eventually rises faster than the share count does.

How has Amazon stock done compared with the S&P 500?

Over the year to the close on November 5, 2026, Amazon shares rose 4.0% while the S&P 500 gained 8.2%, per Chart Equity's data. The stock traded at 21.0 times trailing earnings and sat about 9.7% below its 52-week high. So it lagged the market even with a 16.9% trailing net margin, which fits the debate over how much it's spending.