Key Takeaways

  • A single query on ChatGPT-4 or Claude draws up to 30 times more electricity than a standard Google search, burning the equivalent of running a 100-watt light bulb for two to three minutes.
  • Modern autonomous warfare relies on low-latency compute: a $12 drone can destroy a multi-million-dollar Abrams or Leopard tank by locking onto the heat signature of its tailpipe.
  • Hyperscalers face hard physics constraints because data centers require 24/7 uninterruptible baseline power that solar and wind alone cannot provide without massive storage or backup.
  • Advanced chip manufacturing and tech hardware depend entirely on rare inputs like gallium, scandium, rhenium, dysprosium, niobium, and tantalum, with supply chains heavily dominated by China.

The Energy Bill of the AI Boom

When you type a query into Google, the servers consume enough electrical energy to illuminate a 100-watt light bulb for about 12 seconds. When you send a prompt to Claude or ChatGPT-4, that number jumps thirtyfold. As Robert Friedland pointed out, an AI query burns enough juice to power that same 100-watt bulb for two to three full minutes.

Silicon Valley built its wealth on the assumption that software scales with near-zero marginal cost. That illusion is colliding with the physical power grid. Every frontier model, every context window expansion, and every autonomous agent cluster requires physical megawatts. Friedland made the limitation blunt: “A data center needs uninterruptible power. It can't work on solar, because the sun doesn't shine all the time. It can't work on wind. The wind doesn't blow all the time.”

If you run an AI company today, your real cost floor is not API token pricing from OpenAI or Anthropic. Your cost floor is the physical availability of baseload electricity and the raw copper wiring required to distribute it across the grid.

Why Software Cannot Ignore the Periodic Table

Software founders often treat AI as a digital tool for writing code or summarizing text. But it is also direct military infrastructure. Friedland highlighted how low-cost autonomous hardware is rewriting battlefield economics: “Drone technology is so efficacious that the Leopard tanks and the Abrams tanks we were giving to Ukraine are parked and no longer used, because a $12 drone can find them from the heat signature from the tailpipe and just kill them.”

Building the next generation of defense tech and AI silicon requires physical elements that software engineers rarely think about. “Without gallium, there is no Nvidia,” Friedland warned. “No Nvidia chips. Without scandium, rhenium, dysprosium, niobium, tantalum, these are metals you don't even know about, we can't build this studio, we can't develop AI as fast as the Chinese.”

The entire stack, from low-latency satellite links to high-density data centers, rests on physical supply chains concentrated in adversarial jurisdictions. When software depends on physical hardware, software risk becomes supply chain risk.

What to Do With This

Audit your compute unit economics against rising power costs instead of assuming inference prices will drop forever. If you build hardware or physical infrastructure, map your tier-one and tier-two component dependencies against critical mineral bottlenecks like gallium and copper today. Build supplier redundancy before geopolitical export restrictions force your hand.