Meta Puts Robots to the Test in Its Data Centers Instead of Technicians

Meta is testing robots that can plug in cables, reboot servers and perform other tasks inside its data centers. That's according to current and former employees familiar with the projects. The goal is to service a rapidly growing fleet of computing capacity with fewer people and to keep labor costs down while spending on AI infrastructure soars.

According to the sources, Meta is using equipment from various vendors — Watney Robotics, Kinova and ABB. In one experiment, a Kinova Gen3 robotic arm is being trialed as a device for cutting power to servers, while a separate robot is being trained to swap network cables. One of the workers estimates that, if successful, such a bot could take on up to 80% of certain duties. “We thought that those of us doing physical work were safe for a while longer. But not anymore,” he says.

There are also some very simple devices. At some sites, Meta has quietly rolled out a mechanism resembling a finger or a pointed stick: on a remote command from a person, it presses the power button on a Mac Mini or another device to reboot it.

Officially, People Are Needed

Publicly, the company insists on the opposite. A Meta spokesperson said the firm is investing in training and hiring: “America is going through its biggest infrastructure boom since World War II, and there's a shortage of skilled workers. We need more people, not fewer.” Meta has launched a free program to train thousands of specialists a year in electrical work, mechanics and plumbing, and in a number of states it even guarantees jobs to graduates.

At a conference, however, Meta robotics manager Eric Xu flatly called placing robots in data centers a “long-term goal” — to speed up incident response, monitoring and preventive maintenance.

Limitations and Sentiments

Robots used to be poorly suited for delicate work: they were expensive and, in early trials, would sometimes crush servers. Now the hardware has gotten cheaper, and AI models have gotten noticeably smarter. But there are plenty of problems. An inventory robot already at work in Iowa and Virginia can only distinguish indicator lights in shades of gray — it can't tell a red bulb from a green one, so a human is still needed. The machine gets stuck on corners and cables, and it has to be moved between buildings by hand. The robots lack autonomy because of recharging, and they can't yet handle the complex wiring of Nvidia GB300 supercomputers.

In Altoona, Iowa, where Meta's largest campus is located and where the new tech is being shaken out, the mood in the workers' chats is grim: “in a couple of years robots will replace us all.” Some employees fear something else, too — that the company is developing software that will let it hire less-skilled and lower-paid workers. “They no longer need people who can think for themselves,” one of the technicians says. “They need ‘smart hands’ that can just follow the AI's instructions and not break anything.”

Source: Ars Technica