Recently, I’ve been paying attention to an interesting phenomenon—the investment logic for new energy stocks has completely changed by 2026.

In the past, investing in new energy meant waiting for government subsidies and relying on economies of scale; electric vehicles and solar cells followed this approach. But now, the real demand driver has become AI. This is the most significant structural shift I’ve seen recently.

How much electricity do AI model training and data centers consume? According to the latest forecasts, global data center electricity use is expected to reach 1,050 TWh this year, a sharp increase from 460 TWh in 2022. Over half of this is contributed by AI-related activities; a large AI training task consumes as much power as tens of thousands of households’ annual electricity. Traditional data center rack power density is 5-15 kW, but AI clusters are directly soaring to 50-100+ kW, which is why Microsoft, Amazon, and Google are investing heavily in nuclear power.

Most importantly, solar and wind energy are intermittent and cannot meet the 24/7 stable power supply needs of data centers, so companies with nuclear and stable power assets are gaining the highest premiums. Amazon alone plans to deploy 12 small modular nuclear reactors with a total capacity of 960 MW.

But there’s a bottleneck many overlook—power generation is easy, but transmission is the real challenge. The global power grid is severely aging, with delivery times for high-voltage transformers and switchgear still at 2-3 years, and supply chains are already at full capacity. Data centers’ share of total U.S. electricity consumption is expected to rise from 4% in 2023 to over 8%, directly boosting power companies’ revenue growth rate from 1% to 4-6%. This is the so-called “selling shovels” opportunity.

Here in Taiwan, Delta Electronics, as a leader in power electronics, has seen a surge in orders driven by high-power density AI servers, expected to continue growing into 2026. Huasheng Electric, a long-term partner of Taiwan Power Company, benefits from the TWD 564.5 billion grid upgrade plan and is also a leader in the charging station industry. United Renewable Energy, as a solar cell leader, has seen gross margin recovery, benefiting from technological upgrades and anti-dumping tariffs, with overseas shipments expected to grow over 15%. Sunwoda’s wind turbine blade materials hold a high market share, with order backlog exceeding TWD 10 billion, and revenue growth projected at 18%. Yuanjing focuses on high-efficiency solar products, with high overseas order visibility, expecting annual growth of 12-15%.

On the U.S. stock side, it’s even more direct. Constellation Energy is the largest nuclear power operator in the U.S., having signed a 20-year contract to restart the Three Mile Island nuclear plant, with a major expansion of data center projects planned for 2026. Oklo’s micro nuclear reactors, supported by Sam Altman, have received NRC approval ahead of schedule; Amazon and Equinix are in talks. This company’s fast-fission technology is low-cost and quick to deploy, with explosive potential amid AI power shortages. Eaton, a leader in smart grids, already has transformer orders scheduled for 24 months out, and its grid business is expected to grow over 25% in 2026. GE Vernova covers high-voltage transformers and HVDC transmission, benefiting from global grid upgrade investments projected at $68 billion annually, with backlog reaching new highs. NextEra Energy, the largest green energy company in the U.S., leads in wind and solar capacity worldwide, with stable dividends and annual growth over 10%.

Honestly, green energy concept stocks and new energy stocks are quite volatile, but the current logic is no longer about hype but about real order certainty and rigid demand. For allocation, I suggest 50-60% in AI power stocks for high growth, 30-40% in traditional energy stocks for defensive support, and the remaining 10% kept liquid. In a long-term upward trend, short-term dips are opportunities to add positions. Key indicators to monitor include AI capital expenditure, grid investment scale, and technological iteration progress.

From 2026 to 2030, the structural opportunities in new energy stocks are truly not to be missed.