Key Takeaways

  • Total electricity consumption across the United States sits at roughly 500 gigawatts.
  • Elon Musk posits that a 1 percent increase in US power capacity (around 5 gigawatts) generates a direct 1 percent lift in gross domestic product.
  • The US economy produces roughly $60 billion to $65 billion of GDP per continuously consumed gigawatt of power.
  • OpenAI and Anthropic match national macroeconomic averages almost exactly, generating a combined $60 billion to $70 billion in annual run-rate revenue across roughly one gigawatt of live power.
  • AI compute capacity has been tripling annually while frontier lab revenues have grown tenfold each year, creating an aggressive divergence in unit productivity.

The Real Ceiling on Tech Growth Is Measured in Megawatts

Software founders used to think compute was an infinite abstraction sitting in an AWS region. Silicon Valley treated electricity as a fixed utility bill rather than a hard constraint on economic output. That assumption is dead.

On TBPN, John Coogan and Jordi Hays unpacked discussions involving Elon Musk, Jensen Huang, and Gavin Baker regarding the tight physical link between power grids and macroeconomics. As Coogan pointed out, “Consumption in the United States is about 500 gaw.” When you look at total national economic output across that 500-gigawatt baseline, the math yields a clean conversion rate: “a single gigawatt generates about 60 to 65 billion dollars of GDP per continuously consumed gigawatt.”

Musk made a public bet on this exact ratio. He claimed that expanding power supply unlocks economic expansion in lockstep: “I would bet anyone that 1% increase in power usage corresponds to roughly 1% increase in GDP and and so 10 10 gawatt would be a 2% increase in in GDP.”

For decades, tech companies scaled without worrying about base-load generation. Today, building the next tier of intelligence requires negotiating directly with utilities, buying nuclear plant output, and building dedicated substations.

Frontier AI Labs Are Replicating the Entire Macroeconomy

The surprising discovery in Coogan's analysis is how closely modern foundation model companies match national output ratios. Tech startups typically generate extreme software margins with negligible physical footprints. Frontier AI labs look far more like industrial economies.

Coogan noted the numbers: “Anthrop and OpenAI are reportedly sitting between 60 and 70 billion dollars in terms of ARR and both of them have around one gigawatt live.” When you measure their collective revenue against their physical power footprint, their economic productivity mirrors the rest of the United States economy almost dollar for dollar.

That dynamic creates a new law of growth. If your product requires continuous inference and massive training clusters, your revenue growth curve will eventually hit a wall dictated by turbine manufacturing, transmission lines, and grid permits. However, software efficiency creates rapid leverage on top of that baseline. As Coogan observed, “compute has been tripling every year, but revenue has been 10xing every year. And so there's actually a divergence there.”

Frontier labs are squeezing ten times more revenue out of three times more compute each cycle. If that efficiency multiplier slows down, your capacity to buy and power physical gigawatts becomes the sole determinant of your company size.

What to Do With This

Audit your compute roadmap tomorrow morning. Calculate your exact infrastructure cost per active user against your gross margins, then stress-test what happens if GPU hosting prices rise 25% due to data center power bottlenecks. If your product relies on heavy inference, start testing smaller, specialized fine-tuned models today to reduce token energy burn before grid constraints squeeze your margins.