Gerard Reid is one of the sharpest energy analysts writing about AI infrastructure. His December 2025 piece on Substack — "The Real Bottleneck for AI Is Not Electricity" — is worth reading in full. It is well-reasoned, data-rich, and largely correct about the US problem.
But it contains one significant error. And that error matters for anyone making capital allocation decisions about European AI infrastructure in 2026.
WHERE REID IS RIGHT
Reid's core argument holds. The US faces compounding constraints on AI infrastructure deployment:
"Inability to get grid connections, supply bottlenecks such as transformers" ✓
"Multi-year waits are a competitive death sentence for AI companies" ✓
"AI companies will make their investments somewhere else" ✓
A power industry practitioner confirmed Reid's thesis independently just this week: 85% of major player data center projects they work on are delayed or cancelled due to grid interconnect, power equipment or permitting. That is not a forecast. That is a field report from 2026.
WHERE REID IS WRONG — THE EUROPE PARAGRAPH
Here is the paragraph that requires correction:
"Europe has strong electrical infrastructure and frequent periods of very low or even negative power prices thanks to the buildout of renewables. But complex regulation, fragmented capital markets and slow execution mean that little frontier AI training happens in Europe and even inference adoption is slower than it should be."
Reid diagnoses Europe accurately as of 2022 or 2023. He has not updated the diagnosis for what has happened in 2025 and 2026. The execution he says is absent is already underway.
WHAT THE DATA ACTUALLY SHOWS — FRANCE 2025–2026
Mistral AI: €4 billion in data centers announced France + Sweden, May 28, 2026. Europe's leading AI company is building its own infrastructure in Europe.
French government: 63 sites pre-qualified for data center deployment, 4 RTE fast-track sites at 400–1,000 MW each. CRE-approved process in place since May 7, 2025.
EDF brownfield programme: Loire-sur-Rhône, Moselle, Seine-et-Marne — former thermal plants converted to data center infrastructure with existing HV connections.
RTE investment: 300M EUR committed for grid reinforcement for data center deployment in Hauts-de-France alone (Data4 Escaudain announcement, 2026).
This is not "slow execution." This is a coordinated national programme delivering grid-connected AI infrastructure at a speed that is impossible in the United States in 2026.
THE TIMELINE COMPARISON REID MISSES
Reid argues that AI companies need speed above all else — and he is correct. But he applies that logic only to the US, implicitly assuming that Europe is slower. The timeline data says the opposite:
US HV transformer (major OEM): 48–60 months
US greenfield combined timeline: 8–12 years to first power
France brownfield RTE connection: 18–36 months
EU second-tier transformer (Efacec, Pauwels): 20–32 months
France brownfield combined timeline: 24–42 months to first power
Speed advantage: France is 5–8 years faster than US greenfield
If speed is the criterion — and Reid correctly argues it is — then France brownfield sites are not the slow option. They are the fastest viable option for any project targeting commissioning before 2030.
THE REGULATORY ARGUMENT — WHAT HAS CHANGED
Reid's point about "complex regulation" reflects the pre-2025 reality. The French government has moved deliberately to address this:
- Loi Industrie Verte (2023) — accelerated permitting for industrial sites including data centers
- RTE fast-track process (2025) — CRE-approved dedicated connection procedure for AI data center sites, announced at Choose France summit
- Government site programme — 63 pre-qualified sites with coordinated planning approval support
- National AI strategy — France 2030 programme with explicit data center infrastructure targets
These are not announcements. They are operational changes that are already compressing deployment timelines in 2025 and 2026.
WHERE REID'S INSIGHT LANDS CORRECTLY FOR GRIDREADINESS
Reid's conclusion deserves to be quoted in full:
"AI does not operate on textbook timelines. Multi-year waits are a competitive death sentence for AI companies and if a state or country does not enable the necessary speed of AI and related electrical infrastructure deployment, then AI companies will make their investments somewhere else."
This is exactly right. And "somewhere else" — for projects targeting 2027–2028 commissioning — is France brownfield. Not because of sentiment or policy. Because of the only metric that matters: how fast can you get grid-connected power to a GPU rack.
The US answer in 2026 is: not before 2030 on most sites. The France brownfield answer is: 24–42 months from decision to first power.
Gerard Reid is right that speed determines the winner. The data shows that France is currently winning on speed for any project that cannot wait for the US grid to be rebuilt.
GridReadiness tracks French brownfield AI infrastructure sites, RTE connection timelines and EU transformer lead times monthly. The full database of 40+ grid-connected French industrial sites is available to clients. gridreadiness.com