The power grid can't wait. Goldman Sachs: Data center self-owned power plants may become mainstream; gas turbine orders are already booked until 2031.

The power crisis facing AI infrastructure is spurring an energy supply revolution. Goldman Sachs believes that, constrained by severely lagging grid expansion, the data center "self-owned power plant" model is moving from an edge option to the industry mainstream, and is expected to handle 25% of global data center power demand by 2030—a figure that will be almost zero by 2025.

On September 27, Goldman Sachs' Carbonomics team significantly raised its global data center behind-the-meter (BTM) capacity forecast from 40GW to 67GW in a 50-page report, and predicted that gas turbines, reciprocating engines and fuel cells will together supply 28% of the data center power demand in the United States and 25% globally .

At the same time, the report points out that the gas turbine market is in dire need of supply – GE Vernova’s backlog of orders and reserved capacity has reached 116GW, with all capacity sold out by 2030 and more than 50% of capacity sold by 2031. This supply bottleneck is pushing data center operators toward alternatives such as fuel cells, reciprocating engines and even industrial boilers.

Looking further ahead, Goldman Sachs believes that small modular nuclear reactors (SMRs) are the only scalable permanent solution after 2030, but due to the construction cycle, large-scale deployment is unlikely to be achieved before 2030.

Explosive demand: Data center power demand nearly triples in five years

The AI computing power arms race is pushing global power grids to their limits. Goldman Sachs' US technology team and GS SUSTAIN team predict that global data center capacity will increase from 101GW in 2025 to 217GW in 2030 , with the US alone reaching 108GW.

In terms of electricity demand, Goldman Sachs expects global data center electricity consumption to grow by 170% in 2030 compared to 2025, a significant upward revision from its previous forecast of 117%, with more than 60% of the increase coming from the United States.

This trend will push Goldman Sachs's compound annual growth rate forecast for total U.S. electricity demand to 3.5%—a figure that would have been considered a pipe dream at any utility investor day five years ago.

Power grid in crisis: new transmission line construction has plummeted, and grid connection wait times are approaching five years.

The pace of power grid expansion is far behind the growth in demand. A Goldman Sachs report shows that the average annual length of new high-voltage transmission lines built in the United States has plummeted from 1,700 miles between 2010 and 2014 to 350 miles between 2020 and 2023, with only 55 to 125 miles added between 2023 and 2024 .

Meanwhile, the median waiting time from submitting a grid connection application to commencing formal commercial operation has approached five years. Goldman Sachs believes the data from power equipment manufacturer INNIO is even more alarming: grid connection time has increased from approximately two years historically to over seven years today.

This reality has prompted hyperscale data center operators to begin signing 15-year BTM power contracts. When waiting for grid connection means that a multi-billion dollar campus won't be operational until the next decade, "temporary" backup power naturally evolves into a permanent solution.

Political pressure cannot be ignored either. The report cites data showing that 142 anti-data center protests erupted in 42 states across the US this summer. Every GW of self-generated power means one GW of electricity that won't be passed on to ordinary residents' bills; the political cost of the BTM model is continuously decreasing.

Fuel cells win: speed premium outweighs cost disadvantage

In Goldman Sachs' comprehensive rating system, fuel cells topped the list with 76.6 points, surpassing aero-derivative gas turbines (68.2 points), reciprocating engines (67.0 points), and heavy-duty gas turbine/combined cycle units (59.6 points) – despite having the highest levelized cost of electricity (LCOE) of all options.

The cost comparison is clear : for a 500MW data center, Goldman Sachs estimates that the LCOE of a reciprocating engine is about $80/MWh, a combined cycle gas turbine (CCGT) is about $81/MWh, an open cycle gas turbine (OCGT) is about $91/MWh, and a fuel cell is as high as $117/MWh—even with the 30% investment tax credit (ITC), it can only be reduced to $90/MWh.

However, the core competitive advantage of fuel cells lies in "time." In Goldman Sachs' rating system, "time to production" accounts for as much as 20%, "availability" accounts for 15%, while LCOE and initial capital expenditure combined only account for 25%. Solid oxide fuel cell (SOFC) manufacturers can deliver power from order placement to electricity in just 6 to 12 months; reciprocating engines require 1.5 to 2.5 years; and heavy-duty gas turbines require 5 to 7 years, far exceeding the normal market time of 2 to 3 years.

Bloom Energy's statement during its Q2 earnings call perfectly illustrates this logic: customers now measure ROI in terms of "total electricity cost per unit of computing power," and a 1GW data center that goes online a month earlier could potentially generate $1 billion to $2 billion in revenue. At this scale, the $37 premium per MWh is almost negligible.

Fuel cells also have two additional advantages: First, they have the lowest redundancy requirements, requiring only about 9% over-capacity (about 725MW) to serve a 500MW IT load, while CCGT requires as much as 47% redundancy (about 979MW); Second, fuel cells directly output DC power, which can be seamlessly connected to NVIDIA's 800V high-voltage DC rack architecture, eliminating the need for transformers and inverters and reducing power loss from about 10 to 12% to about 3%.

The entire supply chain is in crisis: from gas turbines to fuel cells, nothing is spared.

The supply shortage in the gas turbine market has reached an all-time high. Goldman Sachs data shows that global gas turbine orders will reach 100GW in 2025, a significant jump from 55GW in 2024; the momentum is even stronger in 2026, with 67GW already signed in the first half of the year alone, including 38GW in the second quarter alone .

The order book situation for major manufacturers is alarming: GE Vernova has a backlog of orders plus reserved capacity of 116GW, and is expected to exceed 125GW by the end of the year, with all of its 2030 capacity sold out and more than 50% of its 2031 capacity already sold; Siemens Energy has committed a cumulative capacity of 87GW, of which 24GW is related to data centers, and even if all announced capacity expansions are included, there will still be a supply-demand gap of about 10% in 2030; Mitsubishi Heavy Industries is already in talks about projects to be delivered in the 2030s.

Engine manufacturers are also overwhelmed: Wärtsilä's 2028 capacity is sold out, and it is negotiating slots for 2029-2030; INNIO's backlog of orders and reservations exceeds 15GW, more than four times the deliveries of the past 12 months, and it has just signed a 1.1GW base power order for a hyperscale data center; Caterpillar's large engine backlog has increased more than 3.5 times; Cummins' orders are booked until 2028; Rolls-Royce said that its data center business now accounts for more than 80% of its power generation sales.

The pressure of supply bottlenecks has spread to extreme measures: SpaceX, owned by Musk, has reportedly begun building its own turbine blade factory in Texas to break through the bottleneck; some AI developers have even restarted industrial boilers and steam turbines just to get the machines running by 2032.

Fuel cells also face production capacity constraints . Goldman Sachs estimates that even if Bloom Energy's new 2GW production line operates at full capacity, its cumulative deliveries by 2030 will only be about 7.7GW, far below its projected global demand of 18GW. To fill this gap, Ceres Power's licensees—including Doosan, Delta, and Weichai—need to simultaneously expand production on a large scale, while Doosan's current annual capacity is only 50MW.

Long-term solution: Small modular nuclear reactors are the only permanent solution

Goldman Sachs also included a key assessment in its report: Small modular reactors (SMRs) are a viable BTM option, "reliable and relatively economical in the long run," but "most data center investments are unlikely to be realized before 2030 due to the long investment cycle."

This timeline aligns perfectly with the inherent logic of the entire BTM narrative: gas turbines, reciprocating engines, and fuel cells are essentially transitional solutions , built on the assumptions of a 5-7 year turbine backlog, limited fuel cell production lines, and persistently low natural gas prices. The only solution that can provide permanent, scalable, fuel price-independent, zero-emission, and truly on-site power generation is a modular nuclear reactor deployed within the industrial park.

Goldman Sachs' timeline almost naturally aligns with the SMR developers' plans: gas-fired BTMs and fuel cells will handle the transitional load before 2030, while modular nuclear energy will gradually scale up in the 2030s, coinciding with the expiration of the first batch of on-site gas assets and the uncertainty of natural gas price trends.

Risk Warning and DisclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.