technology-power

Artificial Superintelligence: Power’s Next Battleground

Leandro Chique

Leandro Chique

— 4 min read
Artificial Superintelligence Power's Next Battleground

Key Takeaways

The signal: energy, not code

In Moonshots with Peter Diamandis, Eric Schmidt frames the near-term AI story in one line: the natural limit isn’t chips — it’s electricity. The episode sets the stakes for an AI era where “superintelligence” becomes a practical infrastructure problem long before it becomes a purely algorithmic one. YouTubeFortune

How big is the load?

Two datapoints anchor the energy shock:

  • Data centres’ electricity use will more than double by 2030 to ~945 TWh, with AI the main driver — a Japan-sized footprint. IEA+1
  • The U.S. alone may need an extra ~92 GW to power AI build-out — roughly a hundred large nuclear units worth of capacity. FortuneQuartz

This isn’t about model size; it’s about megawatts. The noise says “chips.” The signal says “the grid.” Fortune

United States: innovation meets interconnection

What’s working. The U.S. is adding generation at speed: developers plan ~64 GW of new capacity in 2025, with solar contributing over half (12 GW online in H1, ~33 GW targeted for the year). U.S. Energy Information AdministrationNevada Current

What’s not. Interconnection queues and permitting slow the scale-up just as AI demand spikes. Independent reporting shows regional grids scrambling — from Texas manufacturing “burn-in” power surges to East-Coast data hubs hunting firm supply. Houston Chronicle

Bottom line. America has the capital, technology, and project pipeline — but the binding constraint is still the speed of transmission and firm capacity coming online where AI wants to build. U.S. Energy Information Administration

European Union: green momentum, grid friction

Rapid build-out. The EU expects record renewable additions in 2025 (~89 GW) — ~70 GW solar and ~19 GW wind — on top of 2024’s solar surge (now larger than coal in the EU mix). Reuterspower-technology.comEmber

But the grid lags. Europe’s power system faces connection delays and congestion: connecting large data centres commonly takes 7–10 years in legacy hubs, reshaping where AI infrastructure goes. The European Court of Auditors likewise flags grid readiness as a strategic risk for electrification. EmberReuters

Case in point (UK). Developers are exploring direct hookups to gas pipelines to bypass lengthy grid waits — a stark signal that firm power trumps ideals when timelines bite. Financial Times

Bottom line. The EU leads on climate ambition and adds capacity fast, but without baseload and faster interconnects, AI at scale will drift toward regions that can energize it sooner. ReutersEmber

China: the scale play

Supply at speed. China’s build-rate is extraordinary. In 2024 it added an all-time record of new capacity, and H1 2025 alone saw ~268 GW of new renewables (an eye-popping ~212 GW solar). Wind+solar are already shouldering over a quarter of monthly generation at times. Climate Energy Financepv magazine InternationalEmber

Storage and baseload. New-energy storage hit ~74 GW/168 GWh by end-2024, while nuclear continues to expand with dozens of additional GW under construction. Together, that’s a foundation for 24/7 compute. China Energy Storage Alliance

Bottom line. If advanced chips flow, China’s grid scale becomes a force multiplier — enabling AI capacity build-outs at a speed the West struggles to match. Climate Energy Finance

The Energy Cold War: deterrence and hardening

Schmidt and co-authors have floated Mutual Assured AI Malfunction (MAIM) — a strategy space that borrows from nuclear deterrence: sabotage destabilizing AI projects rather than rush an uncontested “Manhattan Project.” Expect data centres to be militarized assets — sited near firm power, ring-fenced by national security. Business InsiderRAND Corporation

The policy logic follows the electrons. Regions that can guarantee firm, proximate power will pull the AI stack — chips, labs, model training, and downstream industries. Regions that can’t will rent it. Ember

Strategy map: what matters now

For the U.S.

For the EU

  • Marry ambition to baseload. Record-pace renewables won’t alone feed 24/7 AI load. Faster grid build-outs and re-opened nuclear options are the lever. ReutersEmber
  • Fix the queue. A 7–10 year connection cycle cedes the initiative; policy must compress that to ≈1–3 years in priority zones. Ember

For China

Why it matters (for readers)

  • Investors: Grid-adjacent assets (transmission, firming, storage, nuclear/SMR) are now AI-beta. Energy is the new compute. IEA
  • Operators: Site selection should follow firm MW first, fiber second; permitting timelines are a competitive moat (or trap). Ember
  • Policymakers: Infrastructure is destiny. Without fast, firm, local power, you export your AI future to someone else’s grid. Reuters

Sources (selected)

Power, not policy, will decide who leads the AI century.

Read more