A contact attended a conference at ENS (École Normale Supérieure) in Paris last night, organised jointly with BCG. The subject: "AI, what's next." Energy cost was mentioned as a key KPI for AI infrastructure. Grid readiness did not come up. Neither did carbon intensity.

This is not a criticism of the conference. It is a diagnosis of where the mainstream conversation stands in June 2026 — and why the gap that GridReadiness exists to close is larger than it appears.

WHAT THE MAINSTREAM CONVERSATION GETS RIGHT

The cost of energy is a real constraint on AI infrastructure. A 100 MW data center consumes roughly 876 GWh per year. At €105/MWh (Germany average), that is €92 million in annual electricity costs. At €60/MWh (France nuclear baseload), it is €53 million. The €39 million annual difference is real and material over a 10-year asset life.

So the BCG/ENS focus on energy cost as a KPI is correct. It just stops two layers too early.

WHAT IT MISSES — LAYER 1: DELIVERABILITY

The cost of energy is irrelevant if you cannot access the energy. The constraint in 2026 is not the price of electricity — it is the physical delivery of electricity to the data center.

Energy Cost vs Energy Deliverability — June 2026 France nuclear: €50–70/MWh — but only accessible if you have a grid connection
Germany: €90–120/MWh — and Frankfurt new connections banned until 2030
Virginia: $60–90/MWh (variable) — and PJM queue is 7–10 years
Ireland: competitive pricing — but moratorium until 2028

The cheapest energy in the world is worthless if the interconnection queue is 10 years long.
France brownfield grid connection: 18–36 months.
That is the real KPI.

85% of major AI data center projects are currently delayed or cancelled — not because energy is too expensive, but because the physical connection to the grid cannot be secured in time. A $100 billion data center campus can sit idle waiting for a $40 million transformer order (Ashmore Group, May 2026). This is the layer the BCG/ENS conversation did not reach.

WHAT IT MISSES — LAYER 2: CARBON

The second missing layer is more surprising, because the hyperscalers have been public about their carbon commitments for years.

Carbon Intensity — AI Data Center Power by Market France (nuclear baseload): 58 gCO2/kWh
Sweden (hydro): ~45 gCO2/kWh (but limited capacity)
UK: ~180 gCO2/kWh
Germany: 400+ gCO2/kWh (coal + gas mix)
Virginia/PJM (gas-heavy): 350+ gCO2/kWh
Texas/ERCOT: 350+ gCO2/kWh

Source: European Environment Agency · US EPA eGRID 2024

Now map this against the hyperscaler commitments that are actually driving AI data center investment:

Hyperscaler Carbon Commitments Microsoft: carbon negative by 2030 — removing all historical emissions by 2050
Google: 24/7 carbon-free energy for all operations by 2030
Meta: net zero emissions across value chain by 2030
Amazon: net zero carbon by 2040 · The Climate Pledge
Apple: carbon neutral across entire supply chain by 2030

For a hyperscaler with a legally binding carbon commitment and a 2030 deadline, the carbon intensity of their power source is not a sustainability report footnote. It is an engineering constraint. A data center running on German grid power at 400+ gCO2/kWh cannot meet Microsoft's or Google's carbon commitments without enormous offsetting costs. A data center running on French nuclear at 58 gCO2/kWh can.

THE FRANCE ARGUMENT — COMPLETE VERSION

When the two missing layers are added to the energy cost discussion, France's position changes from "competitive" to "uniquely positioned."

France for AI Data Centers — The Full Picture Energy cost: €50–70/MWh · among the lowest in Europe · stable · not linked to gas prices
Carbon intensity: 58 gCO2/kWh · compatible with 2030 hyperscaler commitments
Grid connection: 18–36 months on brownfield sites · vs 7–10 years Virginia
Transformer sourcing: EU second-tier 20–32 months · vs US OEM 48–60 months
Sites: 63 government pre-qualified + 40+ proprietary brownfield database
Capital confirmed: SoftBank €75B · Ardian €5B · Nebius €8B · Choose France 2026

Nuclear power at €50/MWh and 58 gCO2/kWh. For a hyperscaler with a 2030 carbon commitment, France is not just cheaper. It is the only large-scale option that actually works.

WHY THE CONFERENCE CONVERSATION STOPPED TOO EARLY

The BCG/ENS discussion reflects where institutional thinking currently sits: aware of the energy cost problem, beginning to understand the scale of investment required, but not yet working through the physical infrastructure layers beneath the price signal.

This is the same gap that Ashmore Group's May 2026 report revealed — the most rigorous institutional analysis of AI bottlenecks to date, which identified France's nuclear and brownfield advantage but still left France off their release valve list. The mainstream research community is tracking the right problem from the wrong angle.

The cost of energy matters. The deliverability of energy matters more. The carbon intensity of that energy is now a hard constraint, not a soft preference. France addresses all three simultaneously. That is the conversation that has not yet reached the ENS amphitheatre — or most investment committees.

GridReadiness tracks the infrastructure layer monthly. The data is at gridreadiness.com/data.