When Jensen Huang announced that Nvidia is shipping GPUs faster than ever, he was telling only half the story. The other half — the one that keeps data center developers awake at night — is that there is nowhere to plug them in.
The limiting constraint is not compute. It is high-voltage transformers, switchgear and grid connection timelines. And the situation is far worse than most analysts have modelled.
FROM 24 MONTHS TO 5 YEARS
Before 2020, ordering a high-power HV transformer meant waiting 24 to 30 months. Today, that lead time has stretched to 48 to 60 months — five years — at major manufacturers including ABB, Siemens Energy and Hitachi Energy.
Sources: GridReadiness manufacturer contacts, industry procurement data
This is not a temporary supply chain disruption. It is a structural imbalance between demand — which exploded with AI — and manufacturing capacity — which takes years to build.
WHY MANUFACTURING CAPACITY CANNOT CATCH UP QUICKLY
A high-power transformer is not mass-produced. It requires:
- Grain-oriented electrical steel, produced by a handful of mills worldwide
- High-purity copper windings assembled by hand by skilled technicians
- Mineral oil for insulation and cooling
- 6 to 12 months of testing and certification before shipment
Building a new transformer factory takes 3 to 5 years and costs hundreds of millions. The factories that are being announced today will not produce their first units until 2027 or 2028 at the earliest.
THE SCALE OF THE BLOCKAGE
Sightline Climate tracked 12 GW of US AI data center capacity announced for 2026 across 140 projects. Their findings are striking:
25% of projects have disclosed no power sourcing strategy
Source: Sightline Climate, 2025
That 25% figure is particularly damning. One in four announced data center projects — representing billions in capital commitments — has no answer to the question: where do the electrons come from?
WHY CAPITAL ALONE CANNOT SOLVE THIS
The standard analyst model treats hyperscaler capex as compute coming online within 18 months. When Microsoft or Google announces $50 billion in infrastructure spending, the assumption is that the data centers will be operational by the next earnings cycle.
This model is broken. If the transformer order was not placed in 2022, that capex commitment has no delivery date. You cannot pay for a transformer that has not yet entered the production queue.
The companies winning under this regime are those that placed equipment orders 3 to 4 years ago, before anyone was modelling hundreds of megawatts of inference load. Everyone else is in the queue behind them.
WILL EFFICIENCY GAINS SOLVE THE PROBLEM?
A common counterargument is that next-generation chips will consume less power, reducing demand for transformers. This reflects a misunderstanding of the Jevons paradox.
When compute efficiency increases, total deployment increases faster. The GB200 is more energy-efficient per token than the H100. But hyperscalers are buying ten times more of them. The IEA projects data center electricity consumption to triple by 2030 even accounting for efficiency improvements.
More efficient chips accelerate the demand for transformers. They do not reduce it.
THE EUROPEAN OPPORTUNITY
The transformer crisis is acute in the US. It is less severe in Europe for two reasons: European manufacturers retain some available capacity, and several EU countries — particularly France — offer grid connection timelines that are significantly shorter than equivalent US locations.
France's nuclear baseload means stable, low-carbon electricity at competitive prices. RTE, the French grid operator, has been processing data center connection requests faster than many US utilities. And industrial sites across France carry legacy HV infrastructure that can be repurposed.
For US developers looking to deploy capacity before 2027, Europe — and France specifically — may be the only viable option.
"The panic of 'we are six months from running out of compute' has become 'we are five years from running out of transformers.' Capital fixes one. Capital cannot manufacture a transformer." — GridReadiness Intelligence
WHAT TO WATCH
- Transformer manufacturer capacity announcements from ABB, Siemens Energy, Hitachi Energy
- US utility interconnection queue length — a leading indicator of project viability
- European grid connection timelines by country and region
- Second-hand HV transformer availability from decommissioned industrial sites