2026 Is a Make-or-Break Year for the AI Trade

Demand for data center hardware from Nvidia and Micron Technology remains high. (CFOTO/Future Publishing via Getty Images)

The “AI trade”—investing in companies linked to the artificial-intelligence data-center buildout—is about to enter its fourth year, and that brings questions of durability to the fore. Yes, 2026 is a make-or-break year, and in the end, power supply may be what applies the brakes.

For now, demand for data center hardware from the likes of Nvidia and Micron Technology remains high, and that will continue at least through the first half of 2026. Memory-chip makers like Micron are seeing their order books stretching into 2027, and a memory shortage is on the horizon.

There are many things that can upend the AI trade. Big customers like Amazon .com, Microsoft , Alphabet , Meta Platforms, and Oracle —known as “hyperscalers”—may decide to slow down capital expenditures under pressure from shareholders. These giants also could also decide to take a pause while they digest all the data-center equipment they already have.

Private markets have become a big part of the picture, and that may also tip over the AI trade. In 2024 and 2025, AI start-ups raised $315 billion, according to Crunchbase, with much of that going to pay for cloud-computing services from the hyperscalers. A large part of OpenAI’s record $40 billion 2025 funding round winds up as revenue for Microsoft and Oracle. These AI start-ups are almost all unprofitable and will have to keep raising capital to pay their cloud bills. If, for whatever reason, they can’t, a big part of demand will fizzle.

Many data centers are being financed by a mix of private debt and equity. At some point, those investors may decide they have enough exposure to AI data centers and turn off the tap.

But ultimately, no matter what happens to demand and financing, it could be the power supply that will dictate how 2026 goes for the AI trade.

New data-center designs are cresting one gigawatt in power consumption, the entire capacity of a nuclear reactor. These giant projects can’t just hop on the electric grid in the U.S. There is a long interconnection process that takes years to complete, and there is an even more complex regime for new power plants. Similar issues exist in Europe, but authoritarian countries like China and the Persian Gulf kingdoms have more-compressed timelines.

Nevertheless, data-center builders in the developed world can’t wait, so they have turned to on-site power generation.

What are their choices? Nuclear reactors are expensive and have extended build times. Renewables are the least-expensive option, but they require a lot of land and need to be paired with enough energy storage to guarantee uninterrupted supply. New technologies aren’t ready for this scale of deployment. The conundrum has even led to talk of massive space data centers that can leverage all-day solar power.

Back on earth, developers who need power quickly have turned to gas turbine generators. They aren’t as efficient as renewables or combined cycle gas plants, but they can ramp up quickly.

But that is reaching its limits. GE Vernova is the biggest manufacturer of gas turbines, and it is sold out through mid-2029. There are similar dynamics at Siemens Energy and Mitsubishi Power, the second- and third-biggest turbine manufacturers.

Data center hardware names such as Nvidia and Micron, which worked so well in 2025, will continue to show strong earnings results at least through the first half of 2026. But beyond that, much more is up in the air.

Write to Adam Levine at adam.levine@barrons.com