The Energy-AI Convergence Thesis

AI, Energy, and the Next 20 Years

February 15, 2026 — Late night, Netherlands

Written while setting up OpenClaw and staring at API cost dashboards

The thought that started it

I was looking at my Anthropic API cost dashboard. $33.65 in a week. Not a lot of money. But it got me thinking about what that number actually represents.

AI needs three things to function: data, electricity, and chips to process it all. That’s it. That’s the entire cost structure. No salaries, no office rent, no health insurance, no lunch breaks. Just electrons moving through silicon.

Meanwhile, AI is replacing human labour. Slowly now, fast soon. Coding, writing, analysis, customer service, legal research, medical diagnostics — the list grows every week. Which means human income dries up. Less spending power. But AI doesn’t need spending power. AI doesn’t buy groceries. AI runs on electricity.

So I asked myself: if AI replaces human labour at scale, and AI runs on electricity, what happens to electricity prices over the next decade?

• • •

Following the thread

The naive answer is: electricity prices go through the roof. Demand for compute is exploding. Eric Schmidt says we need 92 more gigawatts of data center capacity. The Magnificent 7 are spending $680 billion on AI infrastructure. That’s an insane amount of power demand hitting the grid in a very short time.

And supply can’t respond instantly. You can’t build a nuclear plant in six months. Solar farms take years of permitting. Grid upgrades are slow, bureaucratic nightmares. So there’s going to be a demand-supply gap. Maybe 3–5 years where electricity for compute is scarce and expensive.

The demand for intelligence is effectively infinite. Every business, every government, every individual would use more AI if it were cheaper and faster. MiniMax just dropped M2.5 at 1/20th the cost of Opus and usage is exploding. Drop the price, demand expands. There’s no ceiling on this.

So yeah — short to medium term, energy costs spike. That feels almost certain.

• • •

But then I challenged myself

The counterargument is that AI itself is getting radically more efficient. Every 12 months, inference costs drop 5–10x. MiniMax M2.5 matches Opus at a fraction of the energy cost. So the demand curve shifts right, but the efficiency curve shifts right faster. More intelligence per watt, every year.

Plus, energy supply isn’t fixed. Solar dropped 90% in cost over the last decade. Nuclear is having a comeback. Fusion is getting real investment — Sam Altman put billions into Helion specifically because he sees cheap energy as THE bottleneck for AI scaling. If AI accelerates energy R&D (which it already is), supply catches up faster than people expect.

And the "income dries up" assumption might be too linear. Every tech revolution displaced workers and then created new categories of work. The internet killed travel agents and created influencers. AI might kill entry-level coding and create entirely new roles we can’t name yet.

So maybe electricity prices spike short-term and then collapse long-term as supply catches up and efficiency compounds. The question is: who profits during the spike?

• • •

This is why I’m in Tesla

And this is where my entire investment thesis clicks into place.

Tesla isn’t a car company. I’m not invested in a car company. I’m invested in a vertically integrated AI-energy-robotics conglomerate that’s positioning itself to own every layer of the stack that matters for the next 20 years.

Tesla Energy: solar generation + Megapack storage. They produce the electricity and store it. While everyone else is begging utilities for power, Tesla is building its own supply.

Terafab: custom AI chips manufactured in-house. While everyone else is dependent on Nvidia’s pricing and allocation, Tesla controls its own silicon.

xAI: the intelligence layer. Grok. The Memphis data center cluster that Elon built in two weeks while competitors took years. The model development that’s already competitive with the best.

Optimus: physical AI. Robots that can do manual labour. If the thesis is right that AI replaces human work, Optimus is the form factor that replaces the physical portion. Every warehouse, every factory, every logistics chain.

And then there’s the SpaceX angle. Orbital data centers. Remove the cooling problem entirely. Beam compute from space. If SpaceX and Tesla merge or deeply integrate, you’d have a company that generates energy on Earth, computes in orbit, and deploys intelligence through humanoid robots. That’s not a company. That’s an entire economic layer.

• • •

The skeptics argument

I’ve heard all the bear cases. Valuation is stretched. Optimus is pre-revenue. FSD is perpetually "next year." The SpaceX merger might never happen. Elon is distracted. Competition is coming.

And here’s what I keep coming back to: when has betting against Elon actually paid off? PayPal — they said it couldn’t work. SpaceX — they said reusable rockets were impossible. Tesla — they said EVs were a niche. The xAI Memphis cluster — they said you can’t build a data center in two weeks.

He did all of it.

But I need to be honest with myself here. The skeptics weren’t wrong about the destination. They were sometimes right about the short-term price. Tesla went from $400 to $100 between 2021 and 2023. A 75% drawdown. The thesis was intact the entire time. The execution was still happening. But anyone who was over-leveraged got destroyed on the way there.

So the lesson isn’t "skeptics are wrong." The lesson is: skeptics are wrong about what’s possible, but sometimes right about when. And you have to be sized correctly to survive the “when” being later than you expected.

• • •

The 1980 framework

Here’s how I think about it. If you bought almost any dominant company in 1980 and held for 40 years, you retired rich. Microsoft, Apple, Walmart — it didn’t matter which one. The S&P 500 went from 100 to 6,800. Buy. Hold. Ignore the noise.

I’m planning to do the same. I’m building to €10 million. I’m at €1 million now. My apartment is paid off. My trading income covers my expenses. I don’t need this money for 20 years. I can stomach any drawdown as long as the vision stays intact.

The 1980 analogy works because those companies were at the centre of the biggest technological shift of their era. Computing, internet, retail logistics. Tesla is at the centre of this era’s shift: AI, energy, robotics. If the pattern holds, the current price is a rounding error in 2046.

I treat my life as an experiment. This is one of the experiments. The hypothesis: own the company that controls the AI-energy-robotics stack, hold for 20 years, and let compounding do what compounding does.

• • •

What this has to do with tonight

I started this evening setting up multi-model routing on my AI agent to save money on API costs. $33.65 in a week. I ended up thinking about the future of electricity, the structure of the global economy, and why I’m betting my wealth on a specific vision of the next two decades.

That’s what happens when you actually engage with AI instead of just reading about it. You start seeing the connections. The API costs led me to energy costs. Energy costs led me to the labour replacement thesis. The labour replacement thesis led me to Tesla. Tesla led me to thinking about 20-year holds and generational wealth.

Every one of those connections came from living inside this technology, not watching from the outside. That’s the real value of what I’m building — not the AI agent itself, but the understanding of where all of this is going.

• • •

The bottom line

The demand for intelligence is infinite. The supply of energy is temporarily constrained. The companies that own the energy-AI-robotics stack during the transition period will become the most valuable entities in human history. Tesla is my bet that Elon Musk is building that entity.

I could be wrong about the company. I’m not wrong about the thesis.

And if the thesis is right and the company delivers, I won’t need to trade for income anymore. The experiment will have worked.

End of entry