The United States (like Europe, China and the giants of the Middle East) are in the midst of a real estate race: that of data centers. Nobody wants to be left behind AI race and, for that, they need gigantic enclosures in which to train it. The big problem is that these facilities They consume a lot of energyand there China seems to have the upper hand facing a United States that does not have the best face. In fact, it is estimated that half of its data centers scheduled for 2026 will be delayed not canceled.
And it is something they cannot allow.
It’s not a money problem. Meta, Amazon, Microsoft and Google are going to pool more than $650 billion this year to build artificial intelligence infrastructure. To put it in context, It’s more than the Apollo program cost. that took us to the Moon for the first time or to the great railway expansion of the 19th century.
It is private capital that is doing the trick, but although the State does not pay the main bill, it facilitates operations and influences the pace and deployment of massive data centers through regulatory decisions, permitting and energy planning. And the latter is vital.
The tyranny of 24/7. My partner Alba coined that term a few days ago to describe the current situation in which companies focused on AI find themselves. AI is intended to help us optimize our electricity consumption by the 2030s, but right now it is only achieving one thing: collapsing the traditional grid. This technology needs a lot of energy and, furthermore, constant, which is causing collapses in the network.
The estimate is that the energy consumption of these data centers will increase by 175% between now and 2030. And not only consumption: Google’s emissions have increased by 48% in the last five years and Microsoft’s by another 31%. They were two of the most committed companies with ‘net zero’ by 2050.
The other bottleneck. With this in mind, and knowing that the industry is devouring resources such as NAND memories To feed the AMD and NVIDIA platforms used by hyperscalers, we must talk about the other bottleneck in the sector: the energy. On the one hand, there are the plants themselves and we already know that companies have plans for private nuclear plants, gas is booming and coal is used in peaks of demand.
On the other hand, there is the equipment that is installed in the data centers themselves. We are talking about transformers, switches, dissipation equipment and batteries. Panasonic is one of the largest manufacturers of batteries for racks of data centers. They are “packages” of batteries that are inserted between the equipment so that, in the event of a blackout or maximum demand, they provide specific energy support. A few days ago they commented that its annual production had already been soldbut the problem is that orders keep coming.
Bad forecasts. And there is that bottleneck that we mentioned. As they point out in Bloombergthere are analyzes that already suggest that half of the data centers planned for the United States throughout 2026 will be delayed or canceled. It will be a blow to an industry that cannot stop because there is a lot of money at stake (and even more so the year Anthropic and OpenAI want to become public companies) and where they compete against a China that does not seem to take its foot off the accelerator.
The solution is to electrify the grid using renewables, but the problem is that these solutions can provide constant energy, but they are not the best to provide a lot of energy during peak periods. workouts. Large batteries would be needed and, with the parallel rise of electric cars, there are none. The Wood Mackenzie analyst group points out that the United States “does not have enough capacity to stand on its own, so its companies are forced to go to the export market.”
Geopolitical paradox. And therein lies the problem. The United States and China are immersed in a technological war, but also a commercial one. This makes it difficult for American companies to buy what they need from the Chinese industry, which is what leads the way in batteries and solar panels. Jensen Huang -CEO of NVIDIA- already commented a few months ago that the international conflict was fine, but that there was no need to be short-sighted and they had to take advantage of what China has to offer.
The reality is that data centers are estimated to consume up to 12 GW of energy in 2026 in the United States alone, more than ten million American homes need. And, although the electrical infrastructure represents less 10% of the total cost of a data center, it is impossible for the facility to start operating without it. Now, The US has room for maneuveranother thing is that they activate the levers.
Images | campact, Florian Hirzinger (edited)
GIPHY App Key not set. Please check settings