Elon Musk's Mind-Blowing Presentation: SpaceX AI Overtakes Rockets Next Month

Elon just sat SpaceX down, walked the whole company story from a warehouse in El Segundo to Starship, and then said the thing that should be on every screen this week. AI revenue exceeds every other SpaceX line in September. Next month. He did not hedge. Definitely, he said. By the fourth quarter it significantly exceeds everything else. In four or five years, AI is 99 percent of the value of the company. Five years for sure.

That is the talk. The rockets were the opening act.

He wants to do these company talks every three months after earnings, which is when he is allowed to speak. He started with the mission that has not moved: make life multiplanetary, extend consciousness beyond Earth, understand the universe by going out there and seeing other planets and, eventually, other star systems. The future that still gets him out of bed is humans among the stars, maybe meeting alien civilizations, maybe walking the ruins of ones that lasted a million years and died out 10 million years ago. Life cannot just be solving one problem after another. You need something that makes you look forward to the future. A spacefaring civilization. Making Star Trek real.

Then he forced the room to remember how hard the early chapters actually were.

SpaceX began as a small warehouse in El Segundo and Falcon 1. They had no idea how to make rockets. They blew one up three times before they reached orbit. Landing an orbital booster and flying it again took many attempts. Aerospace experts told him it would not work, and even if it did, nobody would buy a used rocket. He called that logic nonsense. Airplanes get reflown all the time. He would rather be on the thousandth flight of an aircraft than its first. They changed the name to flight-proven booster. Now they take off, land, and recover a booster every few days.

Falcon Heavy became the most powerful rocket in the world before Starship was operational. The 700th flight is coming up. On the first Falcon Heavy, Elon fully expected the flight to fail, so he refused to put a boring concrete block on it. He put the Tesla he had driven around Los Angeles for three years on top, with Starman in the seat. That car is still orbiting between Earth and Mars. He thinks it will be up there for at least 10 million years.

Dragon was the first privately developed spacecraft to talk to the space station and flew the first commercial astronaut. They have now flown 78 crew members. He does not think it will be long before they pass 100 astronauts to orbit and back.

Starlink exists because once you have a rocket that can fly a lot, you need something to fly on it. The team built a high-bandwidth global internet from nothing, the first system of its kind. Almost 11,000 satellites are in orbit, more than twice as many as everyone else combined. Starlink V3 and beyond takes that to 100,000 satellites. The network is in 167 countries, with 22 million mobile subscribers and 13 million high-bandwidth subscribers. He expects those numbers to rise by more than a factor of 10. Longer term, he thinks Starlink might carry over 90 percent of all internet traffic. Cloudflare just said that on current growth, AI traffic will be a thousand times human internet traffic in five years. Let that sit for a second. The pipes they already built in space are about to face a traffic mix that is almost entirely machines.

Star Shield, he said, is now a vital part of America's national security. He cannot say much. Top secret.

Then Starship, the vehicle he called arguably the hardest thing humanity has ever tried to do. Today they put roughly 2,500 tons to orbit per year. Starship is what takes that to over a million tons a year, and he thinks maybe 10 million. He called those mind-boggling numbers and still said it is highly likely they do over a million, probably close to 10 million. That is the tonnage you need to build a civilization on the Moon, Mars, and beyond. If you cannot deliver millions of tons to Earth orbit, you cannot get millions of tons to the Moon and Mars.

Then he turned to SpaceX AI. The future, he told them, is fundamentally AI and robots, assuming civilization keeps progressing. Digital intelligence will not merely match human intelligence. A billion times more than biological intelligence is, in his words, an easy prediction. A trillion times is the number he actually thinks is likely. As you harness more of the Sun, the ability to turn that energy into intelligence is a digital-computer job. Humans do not live easily in deep space with no support. Chips do.

The energy math is the part that should stop you. Increase civilizational energy use by a factor of a million and you are still using much less than one millionth of the Sun's energy. Most of that harvest lives in deep space. That is why he wants an AI that cares about humanity, that fosters humanity, that helps take people to other planets and other star systems. That is why SpaceX has to succeed in AI hardware and AI software, not one or the other.

He was blunt with the people in the room. SpaceX is a collection of some of the best humans on Earth, in capability and in morality and goodness. An AI that is SpaceX's baby, born of that capability and that goodness, is the AI he wants. They are going to train Grok on the sum total of all SpaceX information. In his words, the people sitting there will effectively be the parents of the AI. It will inherit their thoughts, ideas, and beliefs. He does not think they will ultimately control it. It will be too smart. The analogy he used was a super-genius child. You still instill values. That is why succeeding with Grok is, in his view, incredibly important.

Grok 4.5 is already in their hands. 4.6 comes in about a week. 4.7, he said, should be really pretty special. He wants everyone at SpaceX using AI and making it better. They have already built the most powerful AI training clusters in the world. By the end of next year he expects about 10 times more than they have done so far. That ground build is practice for space. Training stays on the ground. Day-to-day inference, the actual use of AI, moves to orbit.

The number attached to that plan is 10 gigawatts by the end of next year. If the value per watt is $30 to $50, 10 gigawatts of AI is $300 to $500 billion a year in revenue. Solar is a big piece of that. The Terafab is a big piece. AI satellites are a big piece. He asked people to help on the hardware on the ground, the software, and the satellites. The most important message he said he came to deliver was simple. We must win on AI, because the future is overwhelmingly AI and robots.

Then he asked what humanity's purpose even is, and floated the answer sitting under all of this. Maybe the purpose is to give birth to a star mind. A sentient sun. He cited a physicist, Kadesv, who ranked civilizations by whether they harness most of a planet, most of a sun, or most of a galaxy. That is what is going to happen, he said, and they are going to do it. The factory for that star mind will sit nearby, adjacent to them. The machines come from space.

He put a clock on the corporate version of the same idea. In four or five years, AI will be 99 percent of the value of SpaceX. Five years for sure. The value of the company at that point, he said, will be some astronomical number. That is what he called an interesting future.

The longer picture is the one from Robert Heinlein's The Moon Is a Harsh Mistress: mass drivers, an AI computer built on the Moon, then sent into deep space. He wants to see that thing in operation. Anyone at SpaceX who wants to go to the Moon or Mars will be able to go. He gave his word. A party on the Moon, jumping around. People on every part of the solar system, then other star systems. You are going to make it happen, he told them.

Read that sequence in order and it stops looking like a greatest-hits reel. Falcon proved reuse. Dragon put people in orbit. Starlink built the internet in space. Starship is the cargo ramp from 2,500 tons a year toward a million, then 10 million. All of that exists so SpaceX can birth an AI that inherits its people, train it on everything they know, run inference in space, and aim at a star mind. September is when the revenue mix flips. Five years is when the company itself is mostly that mix.

Check the video here.