In 2027, the biggest bottleneck for the AI ​​sector may not be chips, but electrical infrastructure. Market estimates suggest that approximately 15 GW of the AI ​​processing capacity to be produced in 2027 may not be commissioned that same year. This roughly means that one-third of the new AI chips produced that year will remain in storage, be operated under power restrictions, or only be connected to the system in 2028.

In North America, the total new AI processing capacity that can be powered by electricity in 2027 is approximately 16.5–23.4 GW. Against a need of approximately 35 GW, the central estimate is only around 20 GW. The difference is approximately 15 GW. Moreover, the availability of electricity alone is not enough. According to analyses, the problem here is not the total electricity production capacity of the US, but rather whether this electricity can be delivered to data center sites by the 2027 timeline.

Grid connection queues, large gas turbines, transformers, switchgear, and permitting processes are delaying project commissioning. A new project applying for grid connection after mid-2025 is unlikely to be ready by 2027. For large gas turbines, the time from order to commissioning reaches four to five years, with production schedules reportedly full until 2031.

As Elon Musk recently pointed out, transformers, cabling, liquid cooling systems, massive cooling units, and complex network infrastructure also need to be installed at the same speed as AI chips.

According to the International Energy Agency, global electricity consumption for data centers will almost double from approximately 485 TWh in 2025 to 950 TWh in 2030. Consumption from AI-focused data centers is expected to nearly triple during the same period.

One of the most striking examples demonstrating the magnitude of this demand is $GEV. The company expects to contract for at least 125 GW of gas turbine equipment by the end of 2026. According to CEO Scott Strazik, the company's production capacity is already largely sold out until 2030.

Elon Musk says that one of the critical constraints in natural gas turbine production is the casting of the high-temperature resistant moving and fixed blades used in the turbines. By starting to produce these parts in-house, SpaceX aims to accelerate the commissioning of natural gas turbines to as little as 18 months.

$TSLA and $SPCX are separately trying to install 100 GW of solar power generation capacity per year. However, Musk notes that natural gas will be needed for several more years to support and initially feed the system as solar power capacity grows. Meanwhile, $VST, $CEG, and $BE, which claims it can connect data centers to electricity much faster than these companies with its existing power plants, are among the noteworthy stocks. So, the real question for 2027 is no longer just "How many AI chips can we produce?" No: How many of the chips we produce can we actually get working in the same year?