Five AI CEOs Confirm the Power Bottleneck. Here's the Solution They Haven't Named Yet.

When the CEOs of NVIDIA, Tesla/xAI, OpenAI, Amazon and Microsoft all say the same thing in the same quarter, it is no longer a thesis. It is a market fact. And every one of them is describing the problem that GridReadiness has been documenting since 2025.

WHAT THEY SAID — VERBATIM

Jensen Huang — NVIDIA — CSIS Event "At the lowest level: energy. China has twice the amount of energy we have as a nation, and our economy is larger than theirs. It makes no sense to me... There are no new industries you can grow without energy."

Elon Musk — Tesla / xAI — Recent Podcast "Billions of dollars of the most advanced AI hardware ever built. Sitting dark. Not because the chips don't work. Because there isn't enough electricity to run them."

Sam Altman — OpenAI — Recent Interview "Meeting global AI demand will eventually require hundreds of gigawatts of power. Whoever controls power controls the AI value chain."

Andy Jassy — Amazon — Earnings Call "Our single biggest constraint is power. I don't believe that we will have fully resolved the amount of capacity we need for the demand that we have in a couple of quarters."

Satya Nadella — Microsoft — BG2 Pod "The biggest issue we are now having is not a compute glut, but it's power. It's sort of the ability to get the builds done fast enough close to power. So, if you can't do that, you may actually have a bunch of chips sitting in inventory that I can't plug in."

Five executives. Five independent statements. One identical conclusion: the US power grid cannot support the AI buildout.

JENSEN'S SMR VISION — RIGHT PROBLEM, WRONG TIMELINE

Jensen Huang has gone further than identifying the problem — he has proposed a solution. He recently stated that within 6–7 years, companies and potentially individuals will operate their own small modular reactors (SMRs), generating their own power independently of the grid.

This vision is coherent and may prove correct in the 2032–2035 timeframe. But it does not solve the problem for any project targeting commissioning before 2030. SMRs require:

Nuclear regulatory approval: 5–10 years in most jurisdictions
Manufacturing scale-up: first commercial SMR units still years from deployment at scale
Site preparation and grid integration: additional years
Cost normalisation: current SMR economics are not yet competitive at data center scale

The SMR vision is the long-term answer to the problem these five CEOs are describing. It is not the answer for a data center project that needs to be operational in 2027 or 2028.

THE PRESENT REALITY — WHAT THE NUMBERS SHOW

The CEO statements are confirmed by the procurement data that GridReadiness tracks monthly:

US AI Infrastructure — Current Constraints Transformer lead times (major US OEMs): 48–60 months
Grid connection queue, Northern Virginia: 7–10 years
US AI data center capacity blocked for 2026: 7 GW (Sightline Climate)
Projects with no disclosed power strategy: 25% of pipeline
Agentic AI token demand growth by 2030: projected 24× current levels

Andy Jassy's statement — "I don't believe we will have resolved this in a couple of quarters" — is an understatement. GridReadiness data shows this will not be resolved in a couple of years. The transformer manufacturing capacity being announced today will not come online until 2028–2029 at the earliest.

THE SOLUTION THEY HAVEN'T NAMED

None of the five executives named the solution that is available right now. They are describing a US-centric problem and implicitly seeking US-centric solutions — more domestic generation, SMRs, grid investment. These are correct long-term responses.

The immediate solution — the one that exists today, not in 2032 — is European power infrastructure:

European AI Infrastructure — Available Now France grid connection (brownfield brownfield industrial site): 18–36 months
EU second-tier transformer manufacturers: 20–32 months
French nuclear baseload: €50–70/MWh, stable, low-carbon
Available brownfield sites with existing HV infrastructure: confirmed
GOES supply (EU): ArcelorMittal + ThyssenKrupp — no monopoly risk

Nebius understood this. Their 240MW Béthune project — on a former Bridgestone industrial site in Hauts-de-France, already grid-connected — is the model. First capacity online by late summer 2026. That is what brownfield European infrastructure delivers against a US grid connection queue of 7–10 years.

SAM ALTMAN'S INSIGHT — AND ITS IMPLICATION

"Whoever controls power controls the AI value chain."

This statement, applied to geography rather than just generation capacity, has a specific implication: the companies that control European power infrastructure — grid-connected brownfield sites, secured transformer slots, long-term nuclear power contracts in France — control the AI value chain for any project that cannot wait for US grid capacity to be resolved.

IREN understood this when they ordered 12 transformers years in advance for Sweetwater 1. Nebius understood this when they chose Béthune. The window for the next wave of developers to make the same move — ordering from European manufacturers, identifying French brownfield sites — is narrowing as North American demand absorbs European manufacturing capacity.

WHAT TO DO WITH THIS INFORMATION

If you are a data center developer, infrastructure fund or hyperscaler procurement team reading the same CEO statements and asking what to do:

Verify your transformer status — if you have not placed orders with EU manufacturers, your 2027–2028 commissioning targets are at risk
Evaluate European sites — specifically brownfield industrial sites in France with existing 63kV or 225kV connections
Model French power economics — nuclear baseload at €50–70/MWh against US gas-peaker pricing is a material operating cost advantage
Act on the narrowing window — Efacec and Pauwels 2028 delivery slots are being absorbed by North American buyers now

"You may have a bunch of chips sitting in inventory that you can't plug in." — Satya Nadella, Microsoft

Five CEOs identified the problem. GridReadiness identifies where the solution exists right now — before the SMRs arrive, before the US grid is rebuilt, before the window closes.