Meta is testing a group of robots to perform operational and maintenance tasks inside data centers, including connecting and disconnecting cables, restarting servers, and reinstalling some components. According to Wired, which cited current and former employees familiar with the project, testing is still underway. Through it, the company may aim to operate its expanding facilities with fewer workers, at a time when the cost of building and operating the infrastructure needed for artificial intelligence is rising.
During these trials, Meta is using hardware from several companies, including Watney Robotics, Kinova, and ABB. Kinova and ABB declined to comment, while Watney did not respond to a request for comment. Meta also declined to comment directly on the tests, but said through spokesperson Francis Brennan that it is investing heavily in training and hiring employees to build and operate data centers, pointing to a severe shortage of technical labor in the United States.
What Is Meta Actually Testing?
The company is evaluating the Kinova Gen3 arm to perform a server “power cycle,” meaning cutting its power and then restarting it. It is also testing another robot to switch network cables. One Meta data center employee estimated that the technology’s success could enable robots to perform up to 80% of the workload of some employees, a figure attributed to that person’s estimate and not an announced company result.
At other facilities, Meta has deployed a simple robot resembling a finger or a pointed rod that, based on a remote human command, can press the power button on Mac Mini computers and other devices. Autonomous tugging robots move heavy server racks, while wheeled robots read bar codes, track inventory, and inspect some faults.
Altoona and New Albany as Testing Grounds
Meta’s complex in Altoona, Iowa, is used to test new technologies before moving them to other facilities. Since June of last year, the company has been testing two dual-arm robots from Watney to perform some connection tasks under human supervision, but they remain slower than workers. At the new Prometheus complex in New Albany, Ohio, four-wheeled ABB robots have entered service to reinstall components, with the possibility of expanding their tasks and reducing human supervision in the future.
Available information indicates that transport and inventory robots are already operating in several data centers, including facilities in Iowa and Virginia, but they face clear limitations. The inventory robot’s cameras cannot distinguish between green and red indicator lights because they capture images in grayscale. The robots also have difficulty turning and navigating around cables, and require human intervention to open doors and control them between buildings. The recurring need to recharge their batteries creates additional downtime.
Why Does This Development Matter?
The testing reveals that automation in data centers is moving from transporting and tracking equipment to directly handling servers and cables. This could affect the nature of maintenance jobs, particularly repetitive or dangerous tasks, while also raising questions about job quality and the skill levels required, especially in communities that justify tax incentives for data centers by citing job creation.
Meta says it needs more workers and this year launched a long-term program to train thousands of people annually in electrical, mechanical, and plumbing fields. It has promised employment in Louisiana, Ohio, Indiana, and Texas for participants who complete the training, in addition to working with construction-sector unions. However, statements by Meta’s head of robotics, Eric Xu, described deploying robots in data centers as a long-term goal, with expectations of improving response speed for faults, monitoring, and preventive maintenance.
The technical gap remains large. The company acknowledges internally that robots are currently unable to handle some of the intensive connection work required to operate Nvidia GB300 systems. Developers in the sector also believe there are still no proven and stable commercial solutions for replacing damaged parts and organizing cables. Therefore, the confirmed development at present is the expansion of testing—not the large-scale replacement of maintenance teams by robots.