Major technology firms including Microsoft, Amazon, Alphabet, and Meta are planning to invest approximately $635 billion in AI infrastructure development by 2026, a substantial increase from the $383 billion planned for 2025. However, this "adapt or perish" technological arms race is encountering an unexpected obstacle: electricity supply.
The energy bottleneck is shifting the industry's primary constraint from semiconductor shortages to power availability, measured in kilowatts. Training and operating AI models consumes a staggering amount of electricity. The largest data center sites in the United States now consistently draw over one gigawatt of power, enough to supply 850,000 households. As computational demands explode, the critical limitation is becoming electrical capacity rather than processing chips.
U.S. grid operator PJM has warned that by 2027, the electrical grid may lack sufficient capacity and reserves, increasing the risk of blackouts. Compounding the problem, delivery timelines for large gas turbines have extended to 2029 and 2030.
These energy constraints are creating significant cost pressures that could force a reassessment of the planned $635 billion investment. Melissa Otto, Head of Research at S&P Global Visible Alpha, cautioned that if energy prices remain persistently high, tech giants might be compelled to revise their capital expenditure plans in the first two quarters, potentially triggering a "meaningful correction across all equity markets."
Electricity costs constitute 60% to 70% of total data center operating expenses. A more severe challenge is the imbalance between investment and return: for every dollar invested in GPUs, applications need to generate approximately four dollars in revenue. The current actual revenue gap exceeds $600 billion.
In response to the power bottleneck, technology companies are accelerating their transition from asset-light cloud service providers to asset-heavy energy operators. The White House has convened seven major tech firms to sign the "Electricity User Protection Pledge," committing them to build their own power generation resources.
This shift is creating an unprecedented, extended profit cycle for major gas turbine manufacturers including GE Vernova, Siemens Energy, and Mitsubishi Heavy Industries, whose order backlogs now stretch to 2030. This development suggests that the next winners in the AI competition may not be software companies, but rather energy equipment manufacturers.
Comments