Nobody's Ready for What Elon, Bezos & Palantir Are Building

At the G20, Elon put a number on the next 10 years: well over a billion humanoid robots, each one probably five times as productive as a human. He called that a conservative estimate.

Jeff Bezos told Fox we are going back to the moon to stay, and he wants 50,000 data center satellites in orbit. Palantir's Alex Karp says the only thing AI can't make worthless is the knowledge nobody else has: your own proprietary knowledge.

Three people. One bet. AI is leaving the digital world and moving into the physical one. Here is what they are building, how it connects, and who gets left behind.

A billion robots

A billion robots at five times human output means, in his words, "the billion humanoid robots will be more productive than all humans combined."

I have been talking about humanoid robots for years, and the pushback falls into three buckets. It's a gimmick, a PR stunt for headlines. It's decades away. Or Tesla specifically can't pull it off, because it's a car company.

I think all three views are wrong. Traditional robotics companies are great at building bodies that move. The intelligence layer is where things get stuck. Tesla comes from the opposite direction: start with AI that understands the world, then build a body for it to inhabit. Then scale puts it everywhere.

Intelligence gets cheap

Intelligence is the piece moving fastest. GPT-4 class inference, basically the first smart AI that blew everyone's minds, cost about $30 per million input tokens and $60 per million output tokens at launch in early 2023. By 2025, equivalent capability cost under $1 per million tokens. That is a 36 times reduction in 2 years. The cost curves on AI are steeper than Moore's law ever was.

My rough timeline: by the end of 2027, we'll have AI systems capable of performing 80% of digital tasks at or above the level of top 20% human capability.

You can already see it in an ordinary working day. Here is the example I keep using. You get into a self-driving car powered by AI and you talk to it. Grok is built into the Tesla, so that talking generates value. In a Cybercab, you couldn't take control of the car anyway. Meanwhile Grok bots work in the background: cutting videos, researching, buying things on your behalf, maintaining your website, running ad campaigns, prospecting leads and finalizing contracts.

Software goes headless

Software is going the same way. After developers rebuilt Adobe's core apps as open source, and as Elon moved to make Grok headless, Chamath Palihapitiya said this: "I think that IP and software was effectively rendered worthless 2 days ago." The value of IP no longer lives in software.

In February, Elon replied to a post on X that the system "won't use dollars as currency in the future, just mass and energy." When intelligence and labor are near free, the only things that cost anything are raw materials and the energy to power the machines. Everything else becomes software. Most people will think that is the most idiotic thing they've ever read. But it's physics.

The model is a supplier

So what happens to the model layer? The way I see it, the model becomes a supplier inside the product, and when suppliers are easy to switch, their profit margins usually fall. That is the threat facing a pure model company like Anthropic. It can spend billions training the smartest system in the world. But once a rival closes the gap, or an open model gets good enough, work moves somewhere cheaper, faster, or both. The price per token falls toward the cost of running the chip, which is electricity.

Palantir's CEO, Alex Karp, sees the same thing from the other side. What people overlooked, he said, is "the link between specificity and value."

Data is much harder to replicate than anything else in AI. Compute can be bought. Talent can be recruited. Algorithms get published. Data a company uniquely generates, or uniquely has access to, cannot be acquired on the open market. So the largest fortunes will not necessarily belong to whoever trained the smartest model this month. They will belong to whoever controls the scarce things abundance cannot erase: energy, chips, compute, distribution, proprietary workflows, real world data, and the physical machines where intelligence turns into action.

Atoms are slower

Intelligence on a screen is the easy part. Moving atoms is much slower, and David Friedberg explained why on the All-In podcast. In his words: "It's the jobs that are just a human running a loop with a computer that are most likely going to be accelerated or scaled up."

Every one of these machines runs on the same thing, and the AI industry has an energy problem. The compute needed to train and run models is scaling so fast that Earth-based infrastructure can't keep up. Energy on the ground is finite, expensive, and politically contentious. The winning company will be the one that turns chips into useful computation the fastest. If the missing input is power, you build power plants. If the plants are waiting on turbines, you buy turbines. If the turbine makers are waiting on blades, you make blades. You follow the constraint until you own the constraint.

Why the moon

In January 2025, Elon said, "We're going straight to Mars. The moon is a distraction." By February 2026, Elon and SpaceX had made the moon their biggest priority. Everyone covered it as a space story. It is an AI and energy story that happens to take place on the moon.

Mars is a 6-month trip with launch windows every 26 months. In technology, how fast you learn is how fast you win. Blow it on Mars and you wait 26 months for another shot. Blow it on the moon and you go back in 10 days maximum. China is also targeting a crewed lunar landing before 2030, and whoever establishes a permanent lunar presence first gets resource access, space law precedent, and the manufacturing platform for space-based AI infrastructure.

At Davos in January 2026, Elon said something that got buried under all the noise: "The lowest cost place to put AI will be in space." Put the data centers in Starship, power them with the sun, and skip the fights with local governments on Earth. My view is that SpaceX over time is going to be the largest provider of compute on the planet and off of the planet.

Bezos wants that market too. Blue Origin is working on something called Project Sunrise. In his words: "We have a FCC application in to launch over 50,000 satellites that are orbital data centers."

The rest of us

So the richest people on the planet are racing to build robots, energy and compute on Earth and in orbit. What happens to the rest of us?

Elon has talked about universal high income, the idea that robotic abundance produces so much surplus value that every person can live at a standard we would consider wealthy today.

What concerns me is the road between here and there. The guy who drove the truck for 30 years can't eat a PowerPoint slide about the abundant future. He needs to know what he does next Tuesday.

Picture a barbell at the gym: two heavy weights on the ends, a thin bar in the middle. That is the economic shape I think AI is creating. The top 20% will thrive. The bottom 20% will actually be lifted up. The middle 60%, the vast majority of people in developed nations, will get crushed during the transition.

Unlike previous technological disruptions, this one does not mainly hit blue-collar workers, at least at first. It hits knowledge workers: the accountant reviewing financial statements, the lawyer doing document review, the marketer writing copy, the analyst building spreadsheets. For the middle class, work is a source of identity as much as income. A lawyer who spent 7 years in school and another 10 building a career loses a job and an identity at the same time. I'm aware that telling a 45-year-old lawyer to reinvent themselves is easy for me to say. This transition will likely devastate many people who did nothing wrong.

Jeff Bezos sees it very differently. "I think you're going to have a labor shortage," he said. "It's going to actually be difficult to hire people."

What chess shows

I'm optimistic about the destination, pessimistic about the journey, uncertain about the outcome. Every generation has its prophets of transformation who turned out to be early or wrong. Nuclear power was supposed to make electricity too cheap to meter. Flying cars were perpetually 5 years away. I could be adding to this list.

Here is why I think this time is different. The robots are not theoretical. The cost curves are not assumptions. The AI capabilities are demonstrable today. Still, history is full of exponential curves that ended up being S-curves and plateaued earlier than expected.

Think about chess. IBM's Deep Blue defeated Garry Kasparov in 1997. By the 2010s, AI chess engines could obliterate any human player on Earth. Chess must be dead, right? The opposite happened. Chess has never been more popular. Those same engines became the most powerful teaching tools chess has ever seen. A learning curve that used to take decades can now be compressed into years. Professional players are wealthier. They are more famous. Their audience is larger. The AI amplified them.

I think AI will master cognitive tasks the way it mastered chess. Mastering the task does not necessarily eliminate the human. It can amplify the human.

Go back to Elon at the G20. A billion robots, each doing five times the work of a person. Add the moon, the orbital data centers and the knowledge nobody else has. The age of abundance is coming. The transition to get there will be brutal. And on the other side is a question no generation of humans has ever had to seriously answer. What do we do when we do not have to work?