The AI industry is consuming electricity at a rate that exceeds global production capabilities, leading to simultaneous transformations in energy policy, geopolitics, and the global economy.
Sam Altman from OpenAI stated that his company requires a gigawatt of electricity daily. To provide some context, the total amount of new power generation added across the United States in the previous year was approximately 53 gigawatts.
The numbers pertaining to the expansion of AI are astounding. It is expected that in 2026, Amazon, Microsoft, Alphabet, and Meta will invest around $630 billion in data centers and chips. When you add the next seven largest cloud and infrastructure providers, that figure increases to $811 billion. The four biggest tech firms run about 600 data centers worldwide, with another 544 either in the planning phase or under construction.
Building delays and broken supply chains
Building one of these facilities is no small task. A modern 100-megawatt AI data centre costs upward of $4 billion. Around 70% of that goes toward servers and processors. But the real headache is not money, it is getting the thing connected and running.
In major cities like London, securing a grid connection can take up to ten years. Companies are moving to rural areas, particularly parts of Texas, where permits are easier to get, but skilled workers are scarce. In some places, tech companies have had to help build entire communities to house their staff.
Supply chains are buckling under the pressure. Transformers now have lead times of up to 100 weeks in Europe. Nearly 60% of data centre projects were delayed by more than three months last year. About 88% ran into problems just pouring concrete foundations. Cooling systems and fire alarm installation delayed 78% of projects.
The latest Nvidia chips, the Blackwell series and the forthcoming Rubin architecture, generate significantly more heat than previous models. This has led operators to swap out air cooling for liquid cooling systems that require plumbing and water treatment infrastructure. Standard electrical delivery systems are unable to handle the power demands of next-generation server racks, prompting companies to turn to solid-state transformers and placing them in direct competition with the automotive industry for components.
Geopolitical risks and declining returns
The instability in the Middle East is contributing to the risk factor. Most data centers rely on diesel generators for backup power. Due to regional conflict, fuel supplies are threatened, and oil executives attending the CERAWeek conference in Houston cautioned that supply risks have yet to be factored into market prices. According to Melissa Otto, who leads research at S&P Global Visible Alpha, equity markets could face a significant correction if oil prices remain high over an extended period. The S&P 500 is heading for its worst quarterly performance in about four years, with the tech sector down nearly 8%.
The financial returns that tech giants expected from their infrastructure spending are also slipping. Alphabet’s return on invested capital is projected to fall from 51% last year to around 36% by 2030. Microsoft’s share is anticipated to decrease from 95% in 2020 to 36% by 2030.
According to some experts, the solution is not to construct additional power infrastructure but rather to utilize the existing resources more effectively. In established Western economies, electricity grids operate at an average utilization rate of about 30%. They reach their limits for only around 100 hours each year. It could offer an extra 100 gigawatts of power without the necessity for any new power plants by managing that spare capacity more effectively.
A project between GridCARE and Portland General Electric is already testing this idea, using AI to predict renewable output and shift data centre workloads to times and places where power is available. Analysis suggests that a 1-gigawatt data centre using off-peak grid capacity could cut electricity bills for average consumers by as much as 5%.
The message is simple: whoever figures out the power problem first will likely lead the next decade of technological and economic competition.
Your keys, your card. Spend without giving up custody and earn 8%+ yield on your balance with Ether.fi Cash.
Source: https://www.cryptopolitan.com/ai-power-demand-global-energy-grids-risks/