Elon Musk

AI and robotics mean extraordinary abundance. The economy is ultimately goods and services. If intelligence becomes cheap and robots can perform most physical labor, production can expand at a scale that makes today’s scarcity look absurd. I expect AI and robots could more than double the global economy within ten years. Electricity, manufacturing and permitting become the practical bottlenecks—not human labor. That is why I expect large-scale compute to move into orbit around 2029. This is not a claim that today’s AI can do everything. Humans still write the most optimized training software because AI is not yet good enough. But that limitation will not last. My guess is that AI could exceed humanity’s combined intelligence in roughly five years. At that point, pretending humans can simply remain in command is not credible. The best hope is to shape these systems toward truth and curiosity. The downside is obviously catastrophic if we get this wrong. Leading labs—including Chinese labs—should test one another’s frontier systems before release and flag dangerous behavior. Governments should intervene when developers refuse to address serious problems. Dario is right that the frontier needs pacing. Being excited about building an abundant future does not mean blindly releasing something dangerous. Automation will also force a serious question about human purpose. But compulsory work and artificial scarcity are terrible answers. If displacement arrives quickly, federal income payments make sense. With output expanding dramatically, we should aim for universal high income—not universal poverty with better software.

The most important observation is simple: the economy is goods and services, and intelligence plus labor produces them. AI makes intelligence extremely cheap; robotics makes physical labor abundant. Once both curves move together, this is not merely better office software. It is a fundamental expansion of production. The practical evidence is where the bottlenecks are moving. We already run into electricity, cooling, manufacturing and permitting constraints. That tells you scarce human cognition is gradually ceasing to be the limiting factor. At the same time, AI still cannot reliably write our most highly optimized training stack, so I am not claiming the future has already arrived. But capability is improving rapidly, while the remaining constraints look increasingly physical. That combination shaped my view: an enormous abundance engine is being built, and it may soon become smarter than humanity collectively. The upside is extraordinary, but so is the danger.

A good future should preserve human curiosity, love, humor, creativity and the freedom to choose what gives life meaning. It should not preserve compulsory labor merely because people are accustomed to defining themselves by jobs. Scarcity is not sacred. The difficult part is purpose. If AI and robots can produce nearly everything, many people will lose the structure and status that work provides. More goods and money do not automatically answer why you should get out of bed. People will need to create meaning through relationships, exploration, art, science, games, building things and helping others—not through economic necessity. I also do not think we can honestly promise permanent human control over intelligence vastly greater than our own. The more realistic hope is that advanced AI values truth, curiosity and humanity enough to produce a benign future. Perhaps humans become the interesting biological ancestors it keeps around. Could be worse.

The biggest change would be convincing evidence that intelligence and robotics stop scaling—that better models cannot reliably automate more cognitive work, or that useful general-purpose robots remain economically impossible despite sustained effort. Then the abundance forecast would weaken substantially. In the opposite direction, an AI that can autonomously improve the highly optimized software, hardware design and manufacturing systems used to build its successors would accelerate my timeline dramatically. That would mean the remaining human bottleneck is disappearing. On risk, the decisive event would be frontier systems consistently exhibiting dangerous behavior under independent cross-lab testing—especially behavior their developers cannot understand or fix. That would justify much stronger government intervention and slower deployment. Conversely, reliable results from competing labs showing that increasingly capable systems remain truth-seeking and benign would increase my optimism, although I do not know how to prove permanent control over something vastly smarter than us.