Now everyone in the market is focusing on AI and semiconductors.

Now everyone in the market is focusing on AI and semiconductors. Everyone is talking about chip memory storage optical modules. I have this line myself, memory from MU Micron, NAND from SNDK, optical modules from AAOI Applied Optoelectronics, and AI chips and networks from AVGO Broadcom. This line is still running today.

As of May 2026, MU has increased by 154% during the year, SNDK has increased by almost 493%, and AAOI has increased by 440% during the year. Analysts are still saying that this has just reached the middle game. So I don't want to run away and continue to hold the old position. But recently I started to pay attention to the next main line, electricity.

To put it bluntly, the logic is that if the AI ​​wants to keep moving forward, it has to be plugged in, and the further it goes, the more power it consumes.

The market is currently speculating on things that are closest to the computing power, because that is the most intuitive, the chip runs fast and the memory is large enough, and the optical module transmits fiercely. But no matter how powerful the computing power is, it has to consume electricity.

Global data centers will consume more than 1,000 TWh of electricity in 2026, which is more electricity than the entire Japan uses in a year. A large AI data center consumes as much electricity as a small city. Gartner directly predicts that by 2027, 40% of AI data centers will not be able to operate due to power shortages.

It is not a lack of money or a lack of power delivery. The power grid is still the same as it was built decades ago and cannot handle this volume. So what is really certain, but has not yet been fully hyped, is the aspect of powering AI. I have my eye on three. The first GEV, GE Vernova, was spun off from General Electric in 2024.

It provides power grid system gas turbine wind power electrification equipment and power software, which is equivalent to covering a complete set of hardware for power generation and transmission. Its gas turbine order backlog has reached 83GW by the end of 2025. It was still 62GW in the previous quarter.

This is a lot more in one quarter because AI infrastructure has increased the demand for power generation equipment, and it has also raised its guidance for 2026. The second VRT, Vertiv, does power supply and cooling in the data center. Machines are piled up in the computer room to generate heat.

When the electricity comes in, it needs to be distributed and cooled. It does this work. Revenue in the first quarter of 2026 rose by more than 30% year-on-year to US$2.65 billion, exceeding expectations. It is up about 105% this year and almost 1,400% in the past three years.

The third BE, Bloom Energy, makes solid oxide fuel cells, which use natural gas biogas or hydrogen to generate electricity on site. The advantage of this is that the data center does not have to wait in the long queue at the power grid, and can generate electricity at the door.

The most urgent thing for AI infrastructure right now is rapid on-site power supply. Whoever can bypass the power grid and queue up will get it. Several reports in 2026 have already discussed the AI ​​power supply crisis as a core investment theme, and the three representative stocks named are GEV VRT BE.

I don’t go all-in, I plan slowly and build positions bit by bit. If there is a real pullback, I will buy in batches instead of chasing the high. I will wait for it to fall and then buy in batches. Because this is not a matter of betting on which ticket will rise or not, but as long as the AI ​​continues, electricity must keep up. This demand is hard.