Goldman Sachs Raises Behind-the-Meter Data Center Power Forecast to 67GW; Fuel Cells Ranked Top
nashnova research
Goldman Sachs lifted its 2030 global behind-the-meter data center power forecast from 40GW to 67GW, with fuel cells scoring 76.6 points atop three technology paths — the grid cannot keep pace with AI-driven power demand, making self-generation a structural necessity.
Why did Goldman raise the forecast by nearly 70%?
The core driver is a widening gap between data center power demand and grid delivery. Global data center capacity is projected to grow from 101GW in 2025 to 217GW in 2030 — a 170% increase — with over 60% of the new demand in the U.S.
But the grid is slowing down. The median time from U.S. interconnection application to commercial operation now approaches five years. New high-voltage transmission line mileage fell from roughly 1,700 miles per year in 2010–2014 to about 350 miles in 2020–2023.
This means → demand is accelerating while grid delivery is deteriorating. Behind-the-meter power — generating electricity on-site rather than drawing from the public grid — has shifted from a nice-to-have to a structural necessity.
Why did fuel cells score highest?
In Goldman's framework, delivery speed (20% weight) and availability (15%) together account for over 30%. Cost ranks lower. In plain terms = the race is not about who is cheapest — it is about who can be installed and running first.
Fuel cells — modular devices that generate electricity through a direct chemical reaction using hydrogen or natural gas — can be deployed in 12–24 months, need no grid connection, and achieve the highest full-load run time. They scored 76.6 points. Aeroderivative gas turbines scored 68.2, reciprocating engines 67.0, and heavy-frame gas turbines with combined cycle 59.6.
The trade-off is clear: fuel cell levelized cost of energy runs about $117/MWh, roughly 45% higher than combined-cycle or reciprocating alternatives. This reflects a classic "pay more to move faster" calculus — expensive, but the fastest path to power-on.
Where is the bottleneck for conventional gas?
The constraint has shifted from technology readiness to capacity and lead times. Combined-cycle gas turbine delivery now stretches to four to seven years; heavy-frame units typically require five to seven years.
The major OEMs are heavily booked. GE Vernova will exceed 125GW in contracted orders by year-end, with 2030 capacity sold out and over half of 2031 already pre-sold. Siemens Energy carries a backlog near 69GW plus 27GW in reservations, implying lead times of three to four years. Mitsubishi Heavy Industries has project discussions extending into the 2030s.
This means → even companies willing to choose gas cannot secure delivery before 2030 — which is precisely why fuel cells' 12–24 month deployment window commands a premium.
How large is the fuel cell market, and who supplies it?
Goldman estimates fuel cell installations of roughly 12GW in the U.S. and 18GW globally by 2030, translating to a cumulative equipment market of about $35 billion (U.S.) and $55 billion (global). Gas-based behind-the-meter plus fuel cells together could cover roughly 28% of U.S. and 25% of global data center power by 2030 — up from near zero in 2025.
But a supply gap is evident. Bloom Energy, at full capacity and 85% utilization on its 2GW expansion, can cumulatively deliver only about 7.7GW of solid-oxide fuel cells by 2030 — well short of the 18GW global forecast.
This means → one supplier is far from enough. Licensees of Ceres Power — including Doosan, Delta Electronics, and Weichai Power — would need to scale simultaneously to meet the projection.
What are the key checkpoints for this forecast?
Goldman's beneficiary list splits by pathway: fuel cells — Ceres Power, Weichai Power, Delta Electronics; gas-based behind-the-meter — GE Vernova, Siemens Energy, Mitsubishi Heavy Industries, INNIO.
67GW is a forecast, not delivered capacity. Two variables will determine whether it materializes: whether gas turbine lead times continue to lengthen (which would make behind-the-meter demand even more rigid), and whether the fuel cell cost curve declines as projected (if not, the cost gap will cap scale).
In plain terms = the direction is almost certainly right — the grid falling behind compute power is structural. But whether the number lands at exactly 67GW depends on whether supply chain expansion and cost reduction can both outrun the clock.
市场有风险，内容仅供研究参考，不构成投资建议。