AI power crunch: Musk warns G20 of looming electricity shortage
AI power crunch could arrive as early as 2027, Elon Musk told G20 ministers, warning that chip production may grow faster than the electricity needed to run new data centres.
AI could run out of electricity before it runs out of chips!
Elon Musk has warned that the rapid growth of AI could run into a serious power shortage as early as 2027, as electricity supply struggles to keep pace with the growing number of AI chips being produced.
Speaking virtually at a G20 session on emerging technologies in Chapel Hill, North Carolina, the Tesla and SpaceX chief executive said the problem was “not the distant future”. He cited estimates of a potential 15 gigawatt power shortfall for AI chips in 2027.
AI chips are growing faster than power
Musk’s argument is fairly simple. AI chip production is growing much faster than the electricity infrastructure needed to run those chips. He said AI chip production is increasing by around 40% to 50% a year, while power availability outside China is growing at only 10% to 20% annually.
If that gap continues, electricity could become a bigger constraint for AI companies than access to processors. The focus around AI has largely been on getting enough powerful chips to train and run increasingly advanced models. But those chips eventually have to sit somewhere.
That means more data centres, which require enormous amounts of electricity not only for computing but also for cooling the equipment. And building that infrastructure takes considerably longer than producing another batch of chips.
The data centre problem is getting bigger
Elon Musk was not the only tech leader raising concerns about the infrastructure needed for AI. Meta CEO Mark Zuckerberg also addressed the G20 session virtually, pointing to the huge demand for new data centres.
He said building them could require hundreds of thousands, and potentially millions, of skilled trade workers.
Meta itself, Zuckerberg said, is already finding it difficult to hire enough people to build the facilities it needs.
That shows how the AI race is expanding beyond software and semiconductors.
Companies now need land, construction workers, electricity, cooling systems, grid connections and government approvals. The pressure is already being felt in communities where new data centres have sparked concerns over electricity prices, noise and strain on local infrastructure.
AI competition could become an energy race
Elon argued that countries able to rapidly expand their power generation could have an advantage in attracting AI investment. Large data centres can generate tax revenue and other economic benefits, but only if governments can provide reliable electricity at scale.
He also predicted that AI could eventually add 20% to 30% to the global economy, equivalent to roughly $20 trillion to $30 trillion a year, alongside major productivity gains in software and engineering. Those forecasts are ambitious, but the infrastructure challenge is harder to ignore.
The next stage of the AI race may not simply be about who has the fastest chips or smartest models. It could come down to a much more basic question: who has enough power to run them?