What happened
At the G20 Innovation Ministers meeting in Chapel Hill, North Carolina, on Sept. 2, Nvidia founder and CEO Jensen Huang discussed his 'AI economics' with U.S. Commerce Secretary Howard Lutnick. He told officials that AI is evolving from a software technology into national-scale economic infrastructure, like roads, electricity and the internet, and that the worst mistake would be failing to use it and being left behind.
Huang said chips and computers should no longer be viewed only as end products such as PCs and phones; instead, computing power is becoming a production asset that generates economic value. He described data centers as 'AI factories' that consume electricity, chips and data while producing tokens of intelligence—a shift from pricing energy by kilowatt-hour to pricing intelligence by token.
He also argued that this next phase is highly capital-intensive, estimating that building one gigawatt of AI infrastructure currently costs about $50 billion to $60 billion, roughly twice the cost of a $25 billion wafer fabrication plant. Huang predicted that meeting global demand could mean building around 100 GW of AI infrastructure between now and 2030, potentially involving trillions of dollars, though the final figure would depend on project structure, equipment prices and construction schedules.
Why it matters
Huang's framing makes AI a matter of national industrial strategy rather than just a software trend. If his view holds, countries that invest in AI factories and the energy systems behind them could gain an edge, while those that hesitate may fall behind in the next wave of economic transformation.
The conversation also revealed tensions in the AI policy debate: OpenAI CEO Sam Altman said some worry about AI risks is necessary, even as Huang urged governments to avoid regulating based on theoretical harms. Existing data center projects have already sparked disputes in parts of the U.S. over electricity prices, water, land and environmental effects, and major questions remain about who will finance, own and profit from AI infrastructure and who will bear its financial and social costs.
Key facts
Huang and U.S. Commerce Secretary Howard Lutnick spoke at the G20 Innovation Ministers meeting in Chapel Hill, North Carolina, on Sept. 2, according to the source.
Huang said building 1 GW of AI infrastructure requires about $50 billion to $60 billion in investment, about twice the cost of a $25 billion wafer fab.
He forecast that global AI infrastructure construction could reach about 100 GW from now to 2030.
The International Energy Agency projects global data center electricity use will rise from about 485 TWh in 2025 to about 950 TWh in 2030, the source says.
Huang predicted AI will largely achieve artificial general intelligence within a few years, but added that companies still need business-specific goals, knowledge, context and access permissions.
What to watch next
Whether countries, acting on Huang's advice, begin designing national AI infrastructure strategies that include energy, grid, chip and site plans.
How AI regulation evolves: Huang is pushing for rules that target real, present harms, while OpenAI's Sam Altman stresses the ongoing need to take AI risks seriously.
How the burden of AI's massive buildout is shared—especially whether communities and governments that absorb environmental and grid pressures will see benefits broad enough to support projects like public utilities.