You Won't Be Able to Unsee Elon's Master Plan: Why Energy Efficiency Compounds

Your brain costs about twenty dollars of electricity a year and produces roughly $65,000 of labor value. That 3,000-to-1 gap just became an energy race — and SpaceX is aiming Starship at the only power plant big enough to win it.

Stop. Look at this number before you scroll past.

Your brain runs on about twenty watts — less than the light in your fridge. Keep it on all year and you burn roughly 175 kilowatt hours. On a normal American electric bill, that is about twenty bucks. Maybe twenty-five if you live somewhere expensive.

The energy to run a human brain for a year costs about twenty dollars. That same brain, in the U.S., produces something like $65,000 a year of labor value. Napkin math: about 3,000 to 1. The market pays roughly three thousand times more for what a brain produces than it costs to power the brain.

Hold the ratio loosely. It is a frame, an order of magnitude. Even then, it is the largest arbitrage in the world economy — the price of intelligence versus the energy cost of intelligence, separated by three orders of magnitude. For all of human history that gap was untouchable. There was one way to make a thinking machine that ran on twenty watts: raise a human for twenty years. Make more humans. That just changed. And it is tied to SpaceX.

The Bottleneck Was Always Thinking

Every bridge, vaccine, movie, contract, and line of code started as a thought in a human head. There were only so many engineers and doctors. A wage, stripped down, is the price of scarce human thinking. For ten thousand years the world economy was one engine throttled by the same valve: capable minds and the hours in a day. People blame land or ore when they explain history. The real ceiling was intelligence.

AI is the machine that closes the gap. For the first time you can manufacture intelligence out of electricity instead of out of making a child. The economy stops being capped by how many smart humans exist and starts being capped by something physical: energy.

Do not hear "infinite money." Hear this: AI takes the historical bottleneck — intelligence — and turns it into an energy business.

Whoever Owns Cheap Power Owns Cognition

Picture an AI agent doing knowledge work: email, code, analysis, contracts. What does it cost to run? Electricity. Chips burn power. Power costs money. That is basically the whole bill. Profit, in the simplest terms, is the wage it replaces minus the electricity it burns.

If an AI worker's whole cost is energy, then the winner of the AI economy is whoever turns the cheapest electricity into the most cognition at the largest scale. Chip wars, data center buildouts, feuds, lawsuits, doomer talk, utopia talk — all of it collapses into one number: who has the cheapest energy and the cheapest way to put compute on top of it.

That is why the "unrelated" headlines suddenly rhyme. Why every tech giant looks like an energy company. Why Microsoft is reviving Three Mile Island — the nuclear plant whose 1979 partial meltdown panicked the country. Why there are year-long waits just to plug a new data center into the grid. Why nuclear restarts, gas turbines, and grid build-out are suddenly everywhere. The constraint was never going to be the software. It was going to be the power.

Trap One: The Arbitrage Gets Eaten. Trap Two: GDP Lies.

Can you mint endless profits by spinning up a million agents if energy is plentiful? Not the way the napkin suggests. You cannot multiply today's wages by a huge number, because a wage is a price, and prices move. A wage is the price of scarce thinking. When scarcity dies, the price falls.

The first company to replace a $65,000 worker with a few thousand dollars of electricity a year prints margin. Then the second company does it. Then a thousand more. They compete on price. Cognitive work gets driven down toward its true cost — energy. The 3,000-to-1 ratio does not survive contact with competition.

Who keeps the money? That is the real investing question. The surplus does not disappear. It migrates. First it lives in wages. Then, briefly, it goes to the AI owners who deploy first. Eventually it migrates to you as cheaper everything. The game is who holds the margin open longest before it slips to the next layer.

Second trap: GDP. GDP is prices times quantities. If AI drives the price of cognitive work toward zero while the amount of work explodes, the dollar number can stall or shrink while real abundance rockets. Calling this a "$40 trillion market" can be wildly off because the dollars are the measuring stick that breaks. Track purchasing power and standard of living, not the GDP print.

Floor: Every Human Brain. Ceiling: The Sun.

What stays real? The work still gets done. Cognition still happens. Value still gets created — more of it than ever. The dollar price falls toward the cost that does not vanish: energy.

Human brains are the floor, not the ceiling. About 3.5 billion workers do maybe $60 trillion a year of cognitive labor at today's prices. AI will arbitrage that — and that math is still the floor, because it assumes AI only substitutes for work humans already do. Once agents run projects end to end and spin up other agents, they stop being replacements and start being creators. Ten thousand drug candidates in parallel. A tutor for every kid on Earth, all day. Simulations no firm could staff. Value creation decouples from headcount. Sixty trillion is where you start.

The cap is energy. Human brains are the floor. Sunlight is the ceiling.

The sun hits Earth at about 170,000 trillion watts. All of human civilization — cars, factories, data centers, lights — runs on roughly 20 trillion watts. The sun already delivers something like 10,000 times more energy to this planet than our species uses. We run civilization on about one ten-thousandth of the free energy already landing here. That ignores the rest of the solar system.

If intelligence is an energy business, real output scales with how much energy you turn into thinking. Climb even a sliver — say toward one-thousandth of sunlight instead of one ten-thousandth — and you are looking at roughly a thousand times more real output per person. A thousand times richer in what money buys. The ceiling is the sun. We are nowhere near it.

Orbit Is The Prize. AI-1 Is The Pitch.

Cheapest abundant sunlight is not on Earth. Night, clouds, seasons, atmosphere — half the planet is dark because it spins. In the right orbit the sun never sets. Stronger light. No weather. Vacuum as a heat dump. A power plant with no off switch. That is energy-to-intelligence at solar-system scale. That is SpaceX's real ceiling: Starship as access to the largest power generator in the system.

Alongside the IPO, SpaceX announced AI-1 — a data center in space. A satellite that is a rack of computers fed by a giant solar array, dumping heat into vacuum, hauled up by Starship. About 150 kilowatts of computing payload per bird, with solar and radiators. Sunlight in. Usable intelligence out.

Grant The Skeptics. Then Look At The Stack.

The skeptics are right about a lot.

One AI-1 is a few racks of GPUs — roughly a thousandth of one big hyperscale data center on the ground. A rounding error alone. The bet only works as thousands of satellites, Starlink-style: one bird is useless; ten thousand is a business.

Heat is the hard problem, not chips. Radiating 150 kilowatts in vacuum with only radiators is brutal.

Latency hurts snappy chat. Batch jobs, training, offline workloads fit better — lasers or not.

SpaceX is not alone. Google has Suncatcher. Starcloud and others exist.

Economics today look something like four times a warehouse in Texas. Parity may wait until the 2030s. The whole thesis gates on Starship driving launch costs down. This is a 2030s idea at the earliest. Not a 2026 trade.

So why does it still matter? The vertical stack. Google can have the idea. Who builds the cheap rocket? SpaceX. Who builds the satellite bus at practiced scale? SpaceX. Who makes solar cells in Bastrop with Tesla? SpaceX. Who needs to train enormous models as the anchor customer? xAI — Grok folded into SpaceX before the IPO. Anthropic and Google are already renting compute from that operation for over $20 billion a year. Launch, satellite, power, demand — one roof. That stack does not exist anywhere else.

The option nobody prices cleanly is becoming the company that manufactures intelligence from the cheapest energy in the solar system. Floor: every human brain. Ceiling: the sun. Rocket companies do not get priced like that. That is why the IPO felt historic.

What To Do With This

One: energy is the new oil and the new real estate. Smart people watched rates because rates priced money. Watch dollars per kilowatt hour, interconnect queues, nuclear restarts, grid build-outs, solar deployment. Whoever controls cheap energy controls the cost of intelligence — and the cost of intelligence is about to set the price of almost everything. That is the macro variable of the next decade. Watch it like the Fed.

Two: AI-1 is a 2030s thesis sold inside an IPO pitch. Keep the skepticism. It can work and still take long enough to crush people who bought the near-term hype. Both can be true.

Three — the part that actually matters: closing this arbitrage is not a story about a few billionaires pocketing $65,000 per replaced brain. Competition drives cognition toward its energy cost. That is how everyone gets effectively richer — not overnight, and maybe not in dollar terms, but in what money buys. Cheap intelligence means a cheaper doctor in your pocket, lawyer, tutor, analyst, self-driving car, humanoid robot, factory. Scarce human thinking was the expensive input. When that scarcity ends, the price collapses toward sunlight.

The twenty-dollar brain in your skull is the blueprint. Intelligence was always energy in an expensive costume — a body we feed to keep it running. We built a machine that takes the costume off. Now the question is what we do with it.

Check the video here.