Artificial intelligence seeks to reduce its frantic energy needs

AFP

La Jornada Newspaper, Wednesday, July 16, 2025, p. 6

New York. Thanks to new cooling techniques, more powerful chips, and advances in programming, the artificial intelligence (AI) industry is trying to limit its energy consumption in a context of frenetic growth.

AI infrastructure relies on data centers, which, according to the International Energy Agency (IEA), will account for 3 percent of global electricity needs by 2030, double the current proportion.

US President Donald Trump announced $92 billion in infrastructure development and energy production agreements aimed at meeting the growing demand for artificial intelligence.

Trump attended the Pennsylvania Energy and Innovation Summit at Carnegie Mellon University, and much of his message focused on beating China in the global AI race.

Today's commitments ensure that the future will be created (...) right here in Pennsylvania and right here in Pittsburgh, and I have to say, right here in the United States , Trump said at the event.

Consulting firm McKinsey is talking about a race to build enough centers to cope with the massive acceleration in AI use , while warning that lean times are coming.

There are several ways to solve the problem , said Mosharaf Chowdhury, an associate professor at the University of Michigan.

One is to create more energy sources , an avenue that AI heavyweights are also pursuing; another is to reduce electricity demand to an equivalent capacity, he explained.

Water cooling

For the professor, smart solutions can be found at all levels of the AI chain, from physical equipment to algorithms.

According to Gareth Williams of the consulting firm Arup, the energy required to maintain a data center today represents 10 percent of what the servers themselves consume, compared to 100 percent 20 years ago.

This reduction can be attributed to the widespread use of liquid or water cooling instead of conventional ventilation, which circulates fluids directly through the interior of the servers.

All the big players are looking to use water cooling , Williams said, because we're at a point where you don't have an option not to do it .

Nvidia's new chips have increased the power consumption of a range of servers by more than 100 times compared to 20 years ago.

As a result, the liquid can reach much higher temperatures than before, Williams said, but paradoxically, this makes it easier to cool when in contact with the outside air, given the temperature difference.

In early July, Amazon introduced a new liquid cooling system called IRHX, which can be installed in a data center without having to be integrated into the initial architecture.

Another advancement is that data centers are now equipped with sensors that AI can use to monitor temperatures not on a center-wide scale, but in very small areas , and optimize water and electricity consumption in advance, according to McKinsey's Pankaj Sachdeva.

Mosharaf Chowdhury's lab has developed algorithms to accurately assess how much electricity each chip needs to operate, with potential savings of 20 to 30 percent.

Likewise, progress has been made in the microprocessors themselves.

With each generation of chips and GPU (graphics processing unit) design at the semiconductor level, you're starting to get more power-efficient , Sachdeva recalled.

The team led by Yi Ding, a professor at Purdue University in Indiana, demonstrated that it is possible to extend the life of the most powerful AI chips, GPUs or graphics cards, without sacrificing performance , he told AFP.

In January, DeepSeek unveiled its R1 generative AI model, which performs on par with the major US players, although it is built with less powerful GPUs.

The Chinese company's engineers achieved this by programming the graphics cards with greater precision. They also virtually skipped a training stage in the model, previously considered essential.

However, despite these technological advances, there is no way to reduce energy consumption because of something called the Jevons paradox , Ding noted.

Carolina Gómez Mena

La Jornada Newspaper, Wednesday, July 16, 2025, p. 6

3D Lidar technology, a remote sensing system that uses laser light, is capable of optimizing the operation of airports, said Diego García de Paredes, Regional Sales Director of the Outsigh platform, which has tested this spatial artificial intelligence tool to measure the flow of people at airports in France, the United States, Spain, Rome, Cyprus, and Singapore.

At the Third Generation of Passenger Monitoring: Terminal Optimization conference, the entrepreneur asserted that without violating users' privacy, it is possible to measure the time they spend in lines for check-in, baggage claim, and access to stores in the terminals or taking taxis in those spaces, and thus take measures to, for example, reduce these waits or eliminate bottlenecks .

All of this can not only improve the service provided, but also generate savings for airports in various areas, such as knowing how much cleaning staff is needed based on passenger flow, among others.

The executive asserted that with the use of 3D Lidar technology, it is possible to accurately and anonymously track the flow of people from the curb to the boarding gate, optimizing terminal operations, reducing wait times, and improving the passenger experience.

García Paredes explained that they created Outsigh six years ago “with the vision that Lidar technology could go far beyond self-driving cars or iPhones (…) we have won many competitions and have been recognized by the aviation market, winning some of the innovation projects.”

He noted that they've already conducted a pilot at Guadalajara Airport, and they're interested, and in a few weeks we'll be rolling out the first functional deployment at an airport in Latin America . He declined to mention which one it would be.

Lidar emits pulses of light to scan what is around it, and it has been used for different sectors, but we have seen the possibility of scanning airports in three dimensions, in real time to provide spatial information .

This technology can also contribute to improving airport security and boosting sales for stores and restaurants in general, since if travelers spend less time in lines, they'll have more time to visit those areas. All of this data can be used at the commercial, human resources, and maintenance levels to determine whether cleaning is currently manual or, in the future, automated by robots , he predicted. In short, many departments will use it.

▲ Mexico reaffirmed its position as an international leader in educational technology after winning its second championship at the FIRST Robotics World Championship, held in Nagoya, Japan. The LamBot 3478 team, made up of students from the Prepa Tec Campus San Luis Potosí, led a strategic alliance that allowed them to win the title in one of the most demanding student competitions in the world. Photo taken from Facebook

La Jornada Newspaper, Wednesday, July 16, 2025, p. 6