The race for sovereignty in artificial intelligence has caused a competition for US power plants. The acquisition of a fleet of natural gas units by NRG Energy Inc. It comes a few months after the even greater agreement by Constellation Energy Corp. For Calpine Corp. The US power plant agreements announced by mid -May amount to $ 51 billion, more than any other year of this century except one.

The NRG $ 12 billion agreement is a tangible proof that the explosion of data centers by technological giants, among other things, requires huge amounts of energy. It also emphasizes a relevant fact: History shows that the US finds it difficult to build the required production units to meet this demand.

It is not often the case for a company to announce a acquisition that corresponds to about a third of its own value and investors welcome it with enthusiasm. The fact that NRG’s stock jumped by 26% on Monday partly reflects that the company closed the previous administration’s errors. The failed and untimely shift to energy retailing and services had left NRG with inadequate capacity just when the demand for electricity began to grow again after years of stagnation. The assets acquired by LS Power LLC correct this situation and also restructure the NRG business of Texas, expanding its presence in the “heart” of Data Centers, that is, the PJM network, which covers a large part of Midwestern and Middle Atlantic states. In addition, as Contellation with Calpine did, NRG is partially paying in its shares, with the future involvement of LS Power by 11% marking a vote of confidence.

NRG estimates that it pays about half of what the construction of new production units would cost. These costs are a bit unclear, but it has certainly increased. Nextera Energy Inc. CEO John Ketzam estimates that the cost of a new natural gas plant has increased from less than $ 800 per kilowatt of $ 2,400 in just a few years.

This reflects the conflict of two branches that operate with completely different timetables. Technological companies see themselves participating in a race for sovereignty in artificial intelligence, especially against their Chinese competitors, which requires mass investment in computing power, even before business viable applications of this power are completely clarified. But they are based on the energy they need in an electricity industry that has not seen increased demand for over a decade and whose creation of new units takes years, with an additional problem with the limited capacity of the US transport network.

That’s why we see companies like Microsoft Corp. and Meta Platforms Inc. Pay roughly to reopen old nuclear units or build new gas units as quickly as possible.

As the regional electricity networks are constantly issuing increasing estimates of how much load is expected to be added to the system, the question arises as to whether the power sector can respond to the challenge. Based on its historical performance. The short answer is: No.

This comes from a recent resolution by the Hugh Wynne and Eric Selmon of Sector and Sovereign Research LLC. According to their calculations, forecasts imply that peak demand in the US will increase by 139 gigavat by 2030. Although batteries can help cover cutting -edge conditions, it is more likely to require a new dismantling power. Removing already scheduled power additions for the next two years, it will take about 38.5 new distribution capacity every year from 2027 to 2030, approximately quadruple of the average annual rate of the past decade.

One way to interpret it is that these predictions for higher demand for electricity, based on the explosion of artificial intelligence, will simply not be realized because the energy required will not exist. The real limitations of the natural world will be imposed on the ambitions of the virtual. This is sure to play a role in what we see in the next five years. However, another consequence is that data centers will need to become more “smart” in the way they consume energy, managing their load to reduce the pressure on the network, which sounds reasonable, as we are talking about the dawn of the supposed “supernatural”.

In addition to traditional efficiency improvements in newer chips and cooling systems, this effort can also be extended to management of demand itself. A study of flexible demand by major energy consumers, including Data Centers, published in March by researchers at the Nicholas Institute for Energy, Environment and Sustainability of the University of Duke, has attracted great attention to the industry. The study found that if these users could reduce their consumption by about the equivalent of a day of use annually, it would essentially release space for another 76 cargo gigs, significantly contributing to the fulfillment of these forecasts for high demand.

As happened, the NRG did not just get 12.9 LS Power production capacity. It also acquired CPower, a Virtual Power Production Platform (VPP) with six power gigs, distributed to 2,000 commercial and industrial customers. VPPs collect smaller, distributed power plants and storage units from many locations and operate as single larger units. They also incorporate complex demand management mechanisms, fluctuating it based on value signals. These VPPs can offer balancing services to the network against payment.

They are thus the point of intersection between electrical power and computational power. Overall, the NRG agreement reflects the enormous challenge of creating adequate power production and examines what could be done differently to keep up with demand.