Evan Caron is co-founder and chief investment officer at Montauk Capital.
OpenAI’s letter to the White House in October frames America’s AI challenge as an energy shortage: build 100 GW of new generation every year or lose to China. It’s a seductive, simple message. It’s also incomplete.
Framing the problem as a lack of electrons only tells a small part of the whole story. The real issue is the failure to coordinate the ones we already have. How we define a problem shapes what we build to solve it. If we treat energy as a capacity race, we’ll waste trillions addressing the wrong bottleneck.
The tragedy of thinking like engineers, not ecologists
Elinor Ostrom won the Nobel Prize for showing that shared resources thrive when governance aligns individual and collective interests.
OpenAI’s plan for behind-the-meter generation repeats the classic error: extracting private reliability from a shared system instead of strengthening the system itself. Every data center that isolates from grid signals makes the grid weaker for everyone else.
China is doing the same thing at national scale; building 400 GW of new capacity but still curtailing 20% of renewables in some regions. Capacity without coordination is just an expensive paperweight.
Norbert Wiener’s Cybernetics showed that control comes from feedback, not commands. Many modern systems, from autopilots to internet routing, work through distributed intelligence. The grid can too. Millions of devices could self regulate power flows if they were connected by the right feedback protocols.
OpenAI’s plan, by contrast, doubles down on centralization. Build giant power plants to serve giant loads. But real resilience comes from distributed coordination — autonomous agents balancing local conditions the same way the Border Gateway Protocol routing protocol directs internet traffic without a central controller.
Energy systems, like software systems, fail when coordination costs outweigh scale benefits. America’s grid already operates below 50% efficiency (after accounting for losses and idle capacity). Adding 100 GW without fixing the coordination layer is like adding servers to an unoptimized app.
Network economics tells us that value scales with connections, not capacity. Every EV, battery, and device is a potential grid node that could stabilize the system. The more we isolate these resources with private generation, the weaker the collective network becomes.
Network effects in energy systems
In technology, progress comes from protocols, not raw infrastructure. Energy needs the same leap. Standards for communication, pricing for flexibility, and data systems for real-time optimization. That’s where AI can transform energy not by consuming more power, but by orchestrating it.
Resilience comes from diversity, not redundancy. The 2021 Texas freeze proved that centralization fails catastrophically. Distributed, intelligent coordination like the internet’s decentralized architecture keeps systems alive under stress.
Complex systems improve through iteration, not replacement. Software learned this with agile development. The grid can too. Dynamic line ratings, demand response, and AI-driven optimization improve efficiency in months, not decades. Revolutionary rebuilds like OpenAI’s $500 billion plan can stall for years due to regulation.
Claude Shannon’s information theory explains why. More bandwidth (generation) doesn’t equal more throughput unless you improve encoding (coordination). The U.S. grid doesn’t need more signal; it needs less noise.
OpenAI’s own breakthroughs prove this. GPT-4 wasn’t 10 times bigger than GPT- 3, it was smarter about allocating compute. The same logic applies to power. Intelligence, not scale, drives exponential efficiency.
If America wants to lead in AI sustainably, it must build coordination intelligence:
- Protocols for autonomous energy coordination.
- Markets that value flexibility and responsiveness.
- AI systems that turn data centers into grid assets, not isolated islands.
- Governance that aligns private reliability with public stability.
AI shouldn’t treat energy as a constraint. It’s the key to solving energy’s coordination problem. OpenAI could lead this transformation, but instead it’s proposing to build power plants.
The future won’t belong to the nations with the most capacity. It will belong to those with the best coordination.
