In an interview with A16Z, Sam Altman did something that left a deep impression on me: a CEO of an AI company who, throughout the interview, kept returning to the topic of energy.
Not talking about how impressive GPT-5 is, not talking about when AGI will arrive, but talking about electricity, nuclear fusion, and infrastructure.
This points to something many people have yet to realize: the ceiling of the AI revolution isn’t in the algorithms, but in the infrastructure. More precisely, in electricity.
The Flywheel: Driving Down the Cost of Intelligence
Altman described OpenAI’s business logic as, in essence, a three-stage flywheel.
The first stage is a frontier research lab — building the most advanced AI models. The second stage is large-scale infrastructure — deploying the models at a scale that vast numbers of users can access. The third stage is personalized applications — enabling everyone to use AI for their own purposes.
These three stages form a self-reinforcing cycle: research produces better models, models bring in more users, users generate revenue, and revenue is invested into larger infrastructure and more research.
What is the core driving force of this flywheel? Continuously driving down “the cost of intelligence.”
Think about it: ten years ago, to obtain an in-depth analysis in a specialized field, you needed to pay for a consultant. Today, you can use ChatGPT to get an answer that’s at least an 80 out of 100 within minutes. The cost of intelligence is falling at an astonishing pace.
The lower the cost, the more widespread the use. The more widespread the use, the greater the returns. The greater the returns, the more infrastructure that can be invested. The more infrastructure, the more the cost continues to fall.
That is the power of the flywheel.
Sora: Not Just a Video Generator
Altman’s positioning of Sora in the interview differs from how most people understand it.
Most people treat Sora as an “AI video generation tool” — you input text, it produces video. But Altman believes Sora’s true significance lies in this: it is the starting point of a world simulator.
Why? Because to generate realistic video, the AI can’t just “paint” the scenes — it must understand the causal laws of the physical world. How will a ball move after it’s thrown? How does light refract when it hits the surface of water? How does a person’s center of gravity shift as they walk?
These seemingly simple things mean that the AI must construct an internal model of the physical world. It isn’t “generating video” — it’s “simulating a world.”
What is the extension of this direction? Altman mentioned “AI scientists” — future AI won’t merely analyze data; it could participate in the process of scientific discovery: proposing hypotheses, designing experiments, predicting outcomes. If AI truly achieves this, the pace of scientific progress will undergo a qualitative transformation.
In AI Never Shuts Down: The Economic Order Being Reorganized, I discussed how AI is restructuring the operational logic of the economy. But Altman’s vision reaches further — what he sees isn’t just an economic restructuring, but a restructuring of the scientific method itself.
Energy: The Ultimate Bottleneck
This is the part of the interview that alerted me most.
AI’s demand for compute is growing exponentially. And what lies behind compute? Electricity.
The electricity consumed in training a large language model is equivalent to several months’ worth of power for a small city. And as models grow larger and users multiply, the demand for electricity will only continue to soar.
Altman’s energy strategy spans three time scales. In the short term, it relies on natural gas — not environmentally friendly, but reliable. In the medium term, it relies on solar power plus storage technology — increasingly cheap, but with the problem of intermittency. The ultimate long-term bet is nuclear fusion.
Why nuclear fusion? Because fusion has the potential to bring an order-of-magnitude drop in energy costs — Helion Energy’s long-term goal is 1 cent per kilowatt-hour, compared to the current U.S. average of about 12 to 15 cents, which means the cost could fall to less than a tenth of what it is now. If nuclear fusion is truly realized, the constraints on AI’s compute expansion would be completely lifted.
That is why Altman personally invested in the nuclear fusion company Helion Energy. He isn’t doing charity — he’s buying insurance for the future of AI.
In A Comprehensive Reading of Sovereign AI, I discussed how the autonomous development of AI requires technological sovereignty and data sovereignty. But Altman reveals an even more fundamental sovereignty: energy sovereignty. Without a stable and abundant supply of electricity, all the dreams of AI are mere empty talk.
A Sunrise, Not an Explosion
At the end of the interview, Altman used a metaphor: the arrival of AGI won’t be a sudden, instantaneous explosion, but more like a sunrise.
Light doesn’t illuminate the whole world in a single “snap.” It seeps out gradually, slowly, bit by bit from the horizon. You won’t notice the exact moment when “the sun rose” — only when you look back will you realize that the world has already grown bright.
This metaphor is important. Because many people’s imagination of AI is either the panic that “it will instantly replace everyone,” or the dismissiveness that “it’s just a tool, no big deal.”
The truth probably lies somewhere in between: AI’s impact is gradual, continuous, and irreversible. Society will have time to adapt, but “having time” doesn’t equal “adapting automatically.” You must actively adjust your position and align with the rhythm of this transformation.
Aligning, Not Resisting
My greatest takeaway from Altman’s interview wasn’t any specific technical detail, but a way of thinking: the AI revolution isn’t an event, but a process. What it requires isn’t just technological breakthroughs, but also energy, infrastructure, social institutions, talent cultivation — the supporting framework of an entire system.
A thriving AI ecosystem requires a platform that is stable, predictable, and trustworthy. This is a project of trust, not a technological arms race.
Before the sunrise, the most important thing isn’t predicting when the sun will fully rise — it’s adjusting your position: making sure that when the light comes through, you’re standing on the side that meets it, not with your back turned to it.
💬 Comments
Loading...