Elon Musk stated at the G20 Innovation Ministers' Meeting that AI could increase the global economy by 20% to 30%, equivalent to an annual addition of about $20 trillion to $30 trillion in economic output; he explicitly defined this figure as a rough personal estimate.

Musk referred to the incremental gains brought by "digital AI," which does not require humans to physically alter materials, including software development, financial analysis, administrative processes, content production, research, design, customer service, compliance documentation, and some business operations. He predicts that by the end of next year, AI could complete nearly all tasks in the digital domain.

The 20% to 30% increase in the global economy is not a forecast of model company revenues nor an official GDP prediction for the coming year. This figure implies several stringent prerequisites: model capabilities can reliably replace or amplify knowledge work; companies can restructure processes rather than just purchase chat tools; computing power and electricity are sufficient; regulations allow deployment; and labor, capital, and demand can adapt to the new production capacity in a timely manner.

If AI first increases the output of the same group of employees, it may initially manifest as expanded corporate profit margins, lower service prices, or increased investment; if AI directly replaces some copywriting, customer service, basic programming, research, legal review, and operational positions, the new output may not necessarily translate into job growth, and income distribution will depend on how wages, capital returns, product pricing, and tax redistribution change.

Musk also predicts that AI software will become "extremely capable" within 12 to 18 months, achieving a competitive advantage in software writing similar to Stockfish's advantage over human chess players. The Stockfish analogy emphasizes machines surpassing in closed, clearly defined tasks, but software delivery also involves demand definition, product trade-offs, safety responsibilities, cross-team collaboration, production operations, and legal responsibilities, which cannot be measured solely by code generation capabilities.

He pointed out that the physical constraint on AI economic expansion will be electricity. Musk stated that by 2027, about 15GW of manufactured AI computing power may be unable to be powered on because the growth rate of AI chip supply outpaces that of the grid, transformers, power generation, cooling, and data center delivery; thus, the $30 trillion output scenario also depends on the synchronized expansion of energy and infrastructure.

In market mechanisms, AI application companies, cloud service providers, model companies, and customers using AI to reduce costs are the direct buyers of productivity gains; GPUs, data centers, electricity, transformers, networks, liquid cooling, model APIs, and high-quality data are on the supply side. If productivity gains are realized, capital will shift from repetitive digital labor and traditional outsourcing to AI infrastructure, software automation, energy, and companies owning customer workflows; however, if electricity, compliance, data permissions, or corporate adoption speed are limited, capital expenditure may first concentrate on computing power assets, while income and GDP gains will lag significantly.

Source: Public Information