usatoday24
The Data centers They are the new 21st century factories. And like any factory, they need energy. A lot of energy. The main technology companies are building and operating Great data centers that allow offering services (video or Streaming video gamesfor example), but also where the different models of artificial intelligence. The problem is that They need more and morewhich translates into a growing energy demand.
And although there are those who bet on the nuclear power and for the reactivation of fossil fuels, a study considers that the future is in the Solar energy outside the energy network. The problem is that, although the solution sounds great, it is not being applied.
Hyperscalists. This is an important term. Technology companies that operate cloud computing infrastructures on a global scale are known as ‘hyperscalist’. Its data centers are crucial for the development of digital services, but also for the ‘big data’ and the advance of AI, and the term “hyperscalist” responds to those data centers can be scalar quickly and on demand.
Climbing … how? Well, depending on the needs of that company and the fan that wants to cover, that scalability translates into more storage, a faster processing or a greater bandwidth on the network.
Demand. The main players in this are Google, Microsoft, Meta or Amazon and although they have the capacity to expand their data centers, they are running with a huge problem: the amount of resources they consume. In large server centers, Water consumption It has always been a problem that companies have solved in different ways to be more responsible with the environmentbut the arrival of AI has been a revolution.
Train and maintain these models consume a A large amount of energy resources And, apart from the water to dissipate the heat of the servers, a great energy capacity is needed. So much so, although there are companies Building more sustainable data centers At the structural level, energy demand is so brutal that They require coal and natural gas To meet demand. And some like Google either Goal they will use nuclear energy to feed your needs
Energy out of the network. Contextualized the problem and seeing that these energy needs play against decarbonizationinvestigators of companies such as Stripe, Paces or Scale Microgrids have got to work to determine the best solution to feed those data centers in a sustainable way. His conclusion has been presented in a study in which they estimate that the total energy demand of the AI for 2030 will range between 30 and 300 GW.
In the case of the centers where AI training is carried out, that demand will be between 15 and 150 GW. It is a huge fan, but the solution they pose and that they consider optimal is the creation of micro -redes outside the network, fed by solar energy. According to them, systems with 44% solar energy are already competitive in costs compared to those based only on gas, and those that reach 90% renewable can be even more profitable than nuclear projects such as Three Mile Island of Microsoft.
Green spots are green plots for a 90/10 stage
Build where the sun glued. The advantage of this system is that its construction is fast because you do not have to reactivate a nuclear power plant. You are not tied to what the energy market demands, Geopolitical conflicts They do not leave you without supply, it is clean energy, buy solar panels is getting cheaper And, above all, it is easily scalable. If more energy is needed, it is as simple as adding more panels, but the most important thing in this equation is that they can build these centers in optimal places.
Unlike servers centers, which do need to be close to the end user to offer a better service, the data centers in which the AI training is carried out have geographical flexibility. This implies that they can install them in areas with optimal solar radiation and in places where the land is cheap.
Optimal areas
Optimal areas. In the study they have identified plots in the United States with a potential for up to 1,200 GW of solar energy outside the network, with gas support and an optimal area with large esplanades and radiation during a large percentage of the year. So, CaliforniaNevada, Arizona, New Mexico and the east of Texas They would be ideal places to house those data fueled data centers outside the network by 90%, with the remaining 10% backed by gas.
Beyond this, the study points out that most of the appropriate land is private, so it can be purchased to build these facilities and that, in addition, many are within lands that would allow subsidies. If you start today, the construction bond would be between 12 and 24 months and everything seems positive, but it is not being done.
If it is so good … why isn’t it? According to researchers, there are three issues that come into play. Two are closely related and have to do that this of AI training is a very recent phenomenon. The designers of the data centers have historically been skeptical when it comes to getting off the network because what they wanted was to enhance, above all, the reliability. They can’t stay a second without energy, go.
Related to the historical tendency is inertia: it has never been done before, although current technology would allow to operate only with renewables (as some countries already do). And the third reason is the cost, $ 23 per MWh, specifically. The panels are increasingly affordablebut it is more expensive than not buying those panels -evidently. However, the researchers point out that this extra cost would be dampened by the Cost of emissions and compensation that would be avoided in the short term.
Therefore, these solar micro -lands outside the network seem a quick way to feed large -scale data centers, but although technology is mature, it seems to spend time until we see large data centers fed at 90% by their own renewable energy. They estimate that, in short, technology is ready, but technological … not so much.
Images | Xataka, MD MNS
GIPHY App Key not set. Please check settings