In this AI wave, everyone is discussing computing power, chips, and data centers. Very few people are seriously watching electricity.
But electricity is the real Achilles' heel of AI.
Training a large model can consume as much electricity as a mid-sized city uses in a day. Once an AI data center is built, it becomes an uninterrupted power consumption machine, with unstable and highly fluctuating loads.
This raises a question: who will supply the power?
Why Large Gas Turbines Don't Work
The traditional star of power generation has been large gas turbines. Single-unit capacity of 100 to 500+ megawatts, high output, high efficiency, the backbone of the grid.
But AI data centers aren't waiting for the grid.
Large gas turbines have a delivery cycle of 2 to 4 years. Data center construction schedules simply can't wait. Even more fatal is startup speed—cold starts take 5 to 60 minutes, while AI training loads fluctuate by the second. Gas turbines can't keep up at all.
Scale is also a mismatch. AI data centers prefer modular deployment—install a batch this year, expand next year. Large gas turbines are too big and inflexible.
So large gas turbines are basically out.
Small Gas Generator Sets: The Opportunity of Our Time
Small gas generator sets have a single-unit capacity of 2 to 8 megawatts, with medium-speed models reaching 20 to 25 megawatts. Compared to large gas turbines, they seem like "small stuff."
But it's precisely this "smallness" that matches every need of AI data centers:
First, fast delivery. 1 to 2 years, half the time of gas turbines. AI data center construction is a race against time—whoever can supply fastest gets the orders.
Second, fast startup. Cold start in 30 to 60 seconds, fully capable of keeping pace with data center load fluctuations. AI training tasks start and stop at any time; backup power must respond sensitively.
Third, modular flexibility. Install a few units first, add more when needed. Low expansion cost, short cycle, perfectly matching the "get running first" construction pace of AI data centers.
Fourth, a fallback. Around 2030, when grid infrastructure is built out, these units can be converted to backup power—no large-scale obsolescence.
Supply-Demand Imbalance: Prices Only Go Up
HSBC's research report provides data worth noting.
Top manufacturers' orders are already booked through the end of 2027. Equipment prices rose 15% to 20% in 2025, and the increase isn't over—prices are expected to rise another 10% to 15% annually before 2027-2028.
The supply-demand balance inflection point won't come until 2028-2029.
In other words, there are at least two to three more years of a seller's market.
Who's winning in this market?
High-speed gas generator sets (2-8 MW) are led by Caterpillar with about 35% market share, combined with INNIO Jenbacher totaling 65%. Medium-speed models (20-25 MW) are split between Wärtsilä and Everllence, together 75%, all running at full capacity.
Bare unit pricing: Caterpillar and CMI at $600-$650/kW, Jenbacher higher at $750-$800. Full data center deployment costs approximately $1,400-$1,700/kW.
Weichai Power: A Chinese Variable Worth Watching
The global landscape is dominated by Western manufacturers, but one Chinese player shouldn't be ignored—Weichai Power.
It has three differentiated advantages:
1. Much shorter delivery cycles. Caterpillar and Cummins need about 100 weeks; Weichai only needs 30 to 60 weeks, nearly twice as fast. In this order-packed market, speed is competitiveness.
2. New products coming soon. 2-3 MW models expected to launch in June 2026, 5 MW and 7 MW medium-speed models by year-end. The timing is right, catching the demand peak period.
3. SOFC business as a bonus. Solid oxide fuel cells are a long-term direction for future data center backup power. Weichai's capacity is still ramping up, but the direction is right.
Of course, Weichai still needs to prove product reliability, and brand recognition in overseas data center markets needs to be built—these are its challenges.
My Judgment
This is a supply-demand imbalance-driven short-to-medium-term opportunity, with a time window through around 2028.
The core logic is clear: AI construction speed has outpaced grid expansion speed. In this gap, small gas generator sets are the fastest and most suitable solution to fill it.
Large gas turbines—too slow, too big, too expensive—are not the choice for the AI era.
From an investment perspective, Caterpillar is the highest-certainty beneficiary, with a complete product line and #1 market share. Weichai Power is the potential high-elasticity play—if new products land smoothly and pass customer validation, there's room for upside.
But watch the rhythm—after 2028, supply and demand will balance out, and equipment prices will face downward pressure. This isn't a long-term holding thesis; it's about capturing this two-to-three-year window.
Behind the AI arms race for computing power, some are digging into electricity. Those selling generators may be the most understated winners of this market cycle.
This article is based on HSBC research reports and public information. For reference only, not investment advice.