Elon Musk: AI Could Increase Global Economy by 20% to 30% Annually
Elon Musk stated during a remote speech at the G20 Innovation Ministers' Meeting in Chapel Hill, North Carolina, that artificial intelligence could increase the global economy by approximately 20% to 30%, equivalent to an annual increase of $20 trillion to $30 trillion, and noted that this is a rough estimate.
He mentioned that by the end of next year, AI will be able to perform any digital work and any tasks that do not require manipulating atoms by hand; software writing will reach the level of the chess engine Stockfish, making it impossible for humans to compete in programming. Humanoid robots will exceed 1 billion units in ten years, which is a conservative estimate; robots will begin manufacturing other robots, leading to recursive expansion. He is willing to bet that a single unit's output will be approximately five times that of an adult, and the total output of 1 billion units will surpass that of all humanity. Digital AI is the appetizer, while robots are the main course, potentially increasing the economic scale by more than tenfold in the long term.
Regarding electricity, he stated that the industry consensus is that by 2027, there will be at least a 15 GW shortage of electricity for AI chips, with chip production capacity increasing by about 40% to 50% annually, while power supply outside of China is expected to grow by about 10% to 20% annually, with faster curves outpacing slower ones. The gap is "next year, not in the distant future." He mentioned that companies like Google and Anthropic have rented computing power from SpaceX, which has established its own power plant, and stated that selling electricity to data centers is an opportunity for countries to capture tax revenue and service fees. Other speakers included Sam Altman, Jensen Huang, Mark Zuckerberg, and Demis Hassabis.
The above figures are predictions from the speaker, not realized national accounts. The 15 GW figure is based on industry gap metrics he referenced. The establishment of power plants and renting computing power points to the integration of energy and computing, which does not prove that a global electricity shortage has materialized at that figure.
The ministerial meeting will include growth narratives and power plant licensing in the same agenda. He equated the regulatory environment and China's power expansion as competitive variables.
In market mechanisms, buyers are model companies seeking growth narratives and site selection benefits, while sellers are power generation and grid companies that can connect to the grid before 2027. The driving force is the G20 policy window, which is an anticipated event. Funds are flowing from chip orders to power plants, gas turbines, and substations. Beneficiaries are clusters and power generation equipment manufacturers that have secured electricity, while those under pressure are economies where grid growth lags behind chip growth, and purely digital jobs covered by the narrative of being replaced by robots.
Source: Public Information
ABAB AI Insight
Musk treats the G20 as a joint roadshow for Tesla and SpaceX: digital intelligence contributes 20% to 30%, while robots in the atomic world double that. The comparison to Stockfish frames programming as a solved chess game. One billion humanoids, five times the output, and robots making robots project the Optimus production line into a macro production function. The electricity segment is the part that requires signatures from various countries: chip growth outpaces power supply, and whoever connects to the grid first collects taxes. SpaceX's power plant and rented computing power turn competitors into customers. China has electricity but lacks chips, while the U.S. has chips but lacks electricity, which he depicts as two asymmetric constraints.
The capital path shares the same "constraints on electricity" narrative for cars, Starship, xAI, and power plants. The Optimus production line in Fremont is the only physical anchor for a decade of one billion. The motivation is to have ministers view data centers as infrastructure rather than as NIMBY; the strategy is to keep regulation light and pile up permits on the generation side. The $20 trillion to $30 trillion figure is a nominal addition and subtraction of global output, not accounting for energy inflation and grid investment.
Analogous to the competition for power plants during electrification, the fiber surplus during the internet age, and Nvidia framing supply as national strength. Hassabis mentioned that human-level intelligence is equivalent to ten times the Industrial Revolution, and the scale competition has already begun. The stage is that digital capabilities are close to being outsourced, while physical capabilities are still constrained by electricity and manual labor.
This is a combination of technological substitution and capital concentration. The substitution involves the valuation of purely mental labor; the concentration involves groups that can simultaneously master chips, electricity, and robot production lines. The mechanism is: software is first consumed by models, factories are then consumed by humanoids, and there must be a transformer in between. Whoever connects the power plant to the cabinet first will be the one to invoice that 20% to 30%.