Have you ever thought that every day we interact with AI—asking it a question, generating a landscape image, running simple code—all quietly consuming massive amounts of electricity behind the scenes?

Today, the core bottleneck of the global AI industry is no longer the commonly mentioned GPU chip shortage, but the stable power supply unable to keep up with the explosive growth in computing power.

As traditional power grid expansion cannot match the increasing electricity demand of AI data centers; when many North American computing projects are delayed or halted due to power shortages, a device once considered an emergency backup power source has quietly become a hard currency sought after by global tech giants—gas turbines.

From being an emergency backup power source that few paid attention to in the past, it has now become the “powering heart” of AI computing centers. Currently, it is in an industry boom cycle of “full orders and scarce supply,” with even full prepayment only allowing a 3-7 year wait.

1. AI speeds up, but electricity can’t keep pace ==================

The explosive iteration of large AI models has completely rewritten the global power demand structure, and North America, as the core hub of the global computing industry, faces particularly acute power shortages.

AI’s electricity consumption is skyrocketing exponentially

The power intensity of AI computing centers and traditional data centers is worlds apart.

Traditional data centers typically have a single cabinet power of 5-10kW, roughly equivalent to the daily electricity use of 10 ordinary households; whereas the new generation of AI computing centers have single cabinets with 40-100kW, some supercomputing clusters even exceeding 150kW, with power density increasing 5-10 times, and electricity load growing exponentially.

A more intuitive comparison: a large model query consumes 3-10 times the electricity of a traditional search engine.

According to the U.S. Energy Information Administration (EIA), in 2023, U.S. data centers accounted for 4.4% of total societal electricity consumption; by 2028, this share is expected to surge to 12%, effectively doubling electricity use.

It’s not a lack of generation capacity, but the grid simply can’t handle the surge

More challenging than the surge in electricity demand is the fact that traditional power grids cannot keep up with the explosive growth in demand.

On one hand, North American power grids are already severely aging—70% of large transformers are over 25 years past their designed lifespan, and the pace of updating transmission and distribution infrastructure cannot keep up with the growth in computing load.

On the other hand, grid expansion cycles are extremely long. Data shows that from application to grid connection to commercial operation, the median cycle for U.S. data centers has extended to 6 years, with core clusters like Northern Virginia experiencing connection queues of up to 7 years.

As of October 2025, the total power capacity requested by data centers across the U.S. has reached 245GW—equivalent to building three new Yangtze River Delta core city clusters’ electricity loads.

On one side, AI projects are rushing to go online; on the other, grid connection takes six or seven years. The time gap between supply and demand has directly led to an unresolvable shortfall in power.

The power gap continues to widen, and power shortages have become industry norm

Due to supply-demand mismatch, the power shortage crisis in North America’s computing industry continues to intensify.

Morgan Stanley estimates that from 2025 to 2028, the cumulative power shortfall in U.S. data centers will reach 47GW—equivalent to the total electricity consumption of 15 Philadelphia-sized cities; North American Electric Reliability Corporation (NERC) also predicts that from 2027 to 2030, the annual peak power shortfall will remain above 20GW, with high-risk areas including Texas, California, and the Mid-Atlantic.

In today’s North American computing circle, there’s even a saying: “Getting a GPU isn’t real skill; securing stable power supply is the true key to AI project success.”

2. Power grid can’t wait, gas turbines become the optimal solution for AI power supply =======================

Faced with the “water from afar cannot quench the near drought” dilemma of the power grid, North American tech giants and data center operators are turning to “self-built distributed power sources,” with gas turbines being the current mainstream choice for distributed primary power.

Over 90% of new data centers in North America now adopt a “grid power + self-supplied gas turbines” configuration, with gas turbines evolving from emergency backup to the “main power source” supporting basic electricity loads.

Its three core advantages perfectly match the pain points of powering AI computing centers, with no easy short-term alternatives.

1. Rapid deployment solves the “waiting” problem

Traditional grid expansion and large power plant construction take 5-10 years, which AI projects cannot afford.

A gas turbine power generation system can go from procurement to commissioning in just 8-10 months; even more efficient combined cycle systems take only 16-20 months, quickly filling the power gap for computing projects and reducing project delivery times by years.

2. Stable and reliable power, fully compatible with AI’s power characteristics

AI large model training demands near-absolute power stability: even millisecond outages can cause weeks of training results to be lost, incurring huge losses.

Gas turbines start quickly, with millisecond response times, and have strong peak-shaving capabilities, precisely matching the sharp fluctuations in AI computing loads, providing 24/7 stable power.

Additionally, multiple units can be paralleled to flexibly meet different scale computing cluster demands— a 100MW AI data center can be fully powered independently with 70-80MW gas turbines, no longer limited by grid connection cycles.

3. Cost-effective and aligned with North America’s energy structure

North America, especially Texas, has abundant and stable natural gas resources. Combined cycle gas turbines with waste heat recovery can achieve efficiencies over 60%, with an average lifecycle cost of just 7-8 cents per kWh, ensuring long-term economic operation.

Moreover, gas turbines occupy small land areas, require no complex transmission infrastructure, and can be deployed directly within data center campuses, greatly reducing transmission losses and grid connection difficulties.

Major tech giants have already proven the necessity of gas turbines. Meta’s Hyperion data center in Louisiana has built three large natural gas power plants with multiple heavy-duty gas turbines, capable of providing 2.25GW at full load, expected to be operational by 2028-2029, with future capacity up to 5GW; Microsoft, Amazon, Google, and other cloud providers are also heavily procuring gas turbine units across North America. In Q3 2025 alone, new gas turbine orders in North America increased by 95% year-over-year, setting a record high.

3. Demand explodes, supply can’t keep up: a “one machine hard to find” industry reality =======================

Demand is growing explosively, but supply faces rigid bottlenecks that cannot be short-term broken through, turning the “one machine hard to find” rumor into a real industry norm.

Global market highly monopolized, technological barriers insurmountable in the short term

The global gas turbine market is a typical oligopoly, with GE Vernova, Siemens Energy, and Mitsubishi Heavy Industries holding over 85% of the market share. Core technologies, capacity, and supply chains are highly concentrated.

Gas turbines are called the “crown jewel” of high-end equipment manufacturing, with extremely high manufacturing thresholds involving single-crystal high-temperature alloys, precision casting, special coatings, and automatic control—dozens of cutting-edge technologies.

Especially the hot-end blades, which must operate stably at over 1600°C and tens of thousands of RPM for decades, are produced by only a few companies worldwide. Production yield is low, and expansion cycles are long. A complete gas turbine production line, from factory setup, debugging, to mass delivery, takes at least 3-5 years, making short-term volume increases impossible.

Capacity shrank over the past decade, unable to meet surging orders

Adding to the difficulty, the past ten years saw the gas turbine industry in a cyclical downturn, with the three major giants reducing capital expenditure, cautiously expanding capacity, or even shutting down some production lines. When sudden order surges occur, supply is unprepared and cannot respond quickly.

Furthermore, the core component supply chains of the three giants overlap heavily. If any link faces shortages, the entire machine delivery is affected. For example, high-temperature alloy blades, accounting for 35% of the total machine value, have only a handful of qualified global suppliers, with capacity already fully booked by the big three, leaving no room for new entrants.

Delivery cycles extend to 7 years, with orders scheduled into 2032

Severe supply-demand imbalance results in prolonged delivery times and shifting order schedules.

Originally, heavy-duty gas turbines had a 12-18 month delivery cycle, now generally extended to 3-5 years; some customized models take 6-7 years, with schedules reaching 2032.

By early 2026, Siemens Energy’s backlog of gas turbines reached €146 billion, with main models scheduled for delivery around 2029-2030; GE Vernova’s total unfulfilled orders exceed 80GW, with new contracts in 2025 doubling year-over-year, and capacity essentially locked in for 2029; Mitsubishi Heavy Industries’ orders are also scheduled beyond 2030. Even with plans to double capacity, new capacity won’t be available until after 2028.

Industry estimates suggest that in 2026, global gas turbine annual capacity will be about 60GW. The additional installation demand in North America’s AI sector alone exceeds 40GW. Coupled with global energy transition and grid peak-shaving needs, the global gas turbine supply-demand gap ratio exceeds 40%. This tight balance is unlikely to see substantial relief before 2029.

4. Under supply-demand imbalance, new industry changes are emerging ====================

Severe capacity shortages prevent rapid relief of current power shortages, prompting two major new trends: accelerated adoption of alternative solutions and a complete overhaul of procurement rules.

Gas generators become the preferred substitute, with orders booming

Many may confuse gas turbines with gas generators; here’s a simple distinction:

Heavy-duty gas turbines are like “main force vehicles,” suitable for ultra-large-scale AI centers, but current capacity is extremely tight; gas generators are like “flexible small units,” with lower technical barriers, delivery in 6-12 months, much faster than 3-5 years for turbines, and can directly serve as main power sources for data centers via parallel operation—becoming the best choice to fill power gaps.

Large-scale orders are already landing in North America. Industry leader Caterpillar recently secured a 2GW natural gas generator set order, with potential additional capacity up to 8GW, scheduled for delivery from September 2026 to August 2027; another top manufacturer also won a 507MW gas generator order for North American data centers in Q4 2024, used as main power.

Estimates show that by 2026, the additional demand for gas generator sets in North American data centers will reach 9-12GW, requiring over 3,000 units, with annual demand growth exceeding 20%, entering a boom period.

Procurement rules are being completely overhauled: full prepayment, locking in capacity first

The industry’s extreme shortage has completely overturned previous procurement norms.

In the past, projects would proceed after environmental assessments and site selection; now, the trend is “pre-lock capacity, prepay in full, then push the project forward.”

Today, North American data center operators often pay full upfront before environmental approvals or site finalization, locking in gas turbine and generator capacity, even risking project changes, just to secure a place in the queue.

In the current environment, whether you get power equipment has become the key factor in whether AI projects can be implemented and delivered on time. Without equipment, even with land and GPUs, projects can only stall.

Conclusion

In the long term, explosive growth in AI computing power is not a short-term trend but a certainty over the next 5-10 years; aging power grids and rigid energy transition demands are also long-cycle sectors.

As the core equipment of distributed power supply, the demand for gas turbines and gas generator sets will remain high for the foreseeable future. This “scarcity of machines” supply-demand balance is far from resolution.

Amid this subtle reshaping of the global energy landscape, Chinese companies are also making their mark. Previously leading in ultra-high voltage and grid equipment, Chinese manufacturers are now entering a historic opportunity. Companies like Dongfang Electric, Shanghai Electric, and China National Aero-Engine Corporation have achieved independent mass production of heavy and medium-sized gas turbines. With cost advantages, faster delivery, and full industry chain support, they are rapidly expanding into domestic and Belt and Road markets, poised to capture China’s share of the global power shortage wave.