Last week, several major US technology companies saw their share prices fall sharply after reporting earnings. The primary concern was not weak financial results but investors’ growing unease over relentless increases in capital expenditure. The key question facing the market is no longer whether artificial intelligence has enormous potential, but when these massive AI investments will begin to generate meaningful returns.
Many technology executives, including Nvidia’s chief executive Jensen Huang, have urged investors not to become overly concerned about AI’s trajectory, arguing that the industry is unlikely to experience a collapse similar to the dot-com bubble in 2000. However, the lesson from the dot-com crash was not that the internet had failed. Rather, it demonstrated that valuations had become excessively inflated and ultimately required a painful market correction. Likewise, while AI remains a transformative technology, many of the challenges surrounding its development have yet to be resolved. As a result, the pace of progress may fall short of the lofty expectations previously priced into the market, making the recent pullback in large-cap technology stocks far from surprising.
One of the most significant obstacles is the growing global shortage of electricity, which is emerging as one of the biggest risks to the global economy over the next five to 10 years. The International Energy Agency forecasts that global electricity demand will accelerate by 3.6 percent in 2026 and 3.8 percent in 2027. The rapid expansion of AI data centers, together with rising electricity consumption from electric vehicles and industrial activity, is placing unprecedented pressure on power grids worldwide.
According to the IEA, global electricity consumption by data centers – including AI data centers – is expected to exceed 945 terawatt-hours by 2030, more than double the approximately 415 TWh consumed in 2024. To put this into perspective, that level of electricity demand would be roughly equivalent to Japan’s entire annual power consumption today. The critical question, however, is whether the world is prepared for such a surge in electricity demand. Unfortunately, the answer appears to be no.
The development of power infrastructure is lagging well behind the rapid expansion of AI. While a new data center can typically be built within two to three years, the supporting energy infrastructure requires significantly more time. A natural gas-fired power plant generally takes three to five years to construct, a nuclear power station typically requires eight to fifteen years, while high-voltage transmission lines often take five to 10 years to complete.
As a result, PJM, the largest regional power grid operator in the United States, estimates that the country could face an electricity supply shortfall of approximately 60 gigawatts around 2030. Closing such a gap would require generating capacity equivalent to roughly 60 large-scale nuclear reactors. Without sufficient new power generation, electricity shortages could affect as many as 50 million households.
Ultimately, the greatest constraint on AI’s future may not be computing power, semiconductor technology, or even innovation; it may simply be electricity. Without adequate and reliable power supplies, even the most advanced AI chips that Jensen Huang and Nvidia can develop will be unable to realize their full potential.
Andrew Wong is a veteran independent commentator
