Elon Musk brought a calendar to the Group of 20. Appearing at the summit this week, the founder of SpaceX and one of the artificial-intelligence industry’s most consequential financiers delivered a set of dated predictions about AI with the same confidence he applies to his own companies. His closest deadline is now roughly fifteen months away.
By the end of next year, Mr. Musk said, AI will have reached what he called Stockfish level in writing software, a reference to the open-source chess engine that has beaten every human willing to play it for years. His point was blunt: just as grandmasters no longer compete with engines, human developers will no longer be able to compete with machines at coding. He predicted that within the next twelve to eighteen months AI would reach very high capability across engineering and other digital fields.
By the end of 2027, he said, the technology would be able to perform all digital work, everything that does not require direct human manipulation of the physical world. The comparison carried a specific meaning for the executives and officials listening. Stockfish did not merely assist chess players; it settled the question of whether human calculation still mattered in the game’s highest reaches. Grandmasters now study engines the way an apprentice studies a master, and Mr. Musk’s implication for software was the same: coders will keep writing code, but they will write it in dialogue with machines that can outthink them.
The rest of his forecast followed a logic he has pressed for years. Once software can do all digital work, the constraint on AI stops being intelligence and becomes physics. Mr. Musk warned that the industry is heading into a power shortage, estimating that AI chips could face a shortfall of at least fifteen gigawatts in 2027. His reasoning: chip production is growing by roughly forty to fifty percent a year, while the electricity to run the chips cannot be brought online at anything like that speed.
To put the figure in context, fifteen gigawatts is on the order of the output of a dozen or more large power plants running at once for a single industry. Data centers under construction around the world are already competing for grid connections, transformers and natural gas, and the queues keep lengthening. Mr. Musk’s own ventures sit on the same side of that bottleneck, which gives his warning a particular weight.
Then the payoff, as he estimates it. Mr. Musk predicted that AI could expand the global economy by twenty to thirty percent, calling the figure a rough estimate that would translate into roughly twenty trillion to thirty trillion dollars of additional annual output. The arithmetic is simple: a fifth added to a world economy that produces in the neighborhood of one hundred trillion dollars a year. The claim is also, on its face, staggering, since growth on that scale would dwarf any single decade in modern economic history.
Economists who study technological change tend to push back on exactly this point. Productivity gains from earlier general-purpose technologies, from steam to the internet, took decades to show up in the national accounts, and skeptics note that the same lag could apply to AI even if the underlying promise is real. What Mr. Musk’s estimate lacks in caution it makes up in clarity, and executives are already planning as if something like it were true.
Robotics is where Mr. Musk is most optimistic, and his numbers there are the largest of all. He predicted that within ten years the world will hold one billion humanoid robots, roughly one for every eight people alive today. That forecast assumes the machines become cheap enough to deploy in factories, warehouses and homes, and that the economics of physical work follow the economics of digital work once the intelligence is available.
The robot forecast is also the one he has been closest to personally. The carmaker he runs has spent years developing a humanoid machine, and Mr. Musk has said before that robots could eventually matter more to his company than cars do. In his telling, the arc is consistent: AI masters digital work by the end of 2027, and the following decade belongs to machines that can do physical work, a billion of them needing chips, batteries, motors and electricity in quantities the supply chain does not yet produce.
Mr. Musk’s record with dates is uneven, and he has missed self-imposed deadlines before, sometimes by years. That record argues for reading his newest timetable as a direction rather than a schedule. Still, the G20 appearance put a precise set of claims on the record that other executives, policymakers and planners will now be asked about.
The part of his forecast most likely to survive contact with reality is the least dramatic. The power warning is already visible in the lengthening queues for grid connections and in the prices that data centers pay for electricity. If AI chips hit a fifteen-gigawatt shortfall in 2027, the shortage will not look like a technology problem. It will look like a construction problem.
