Meta is testing robots capable of plugging in cables, rebooting servers, and inspecting equipment inside its data centers, and current and former employees say the machines could shrink the ranks of technicians supporting the company's AI buildout.

According to a report by Wired, workers who asked not to be identified said the hardware comes from a mix of suppliers: Watney Robotics in San Francisco, Kinova in Quebec, and ABB in Zurich.

The experiments come as hyperscale operators race to expand data center capacity for AI workloads, a buildout that has tightened the market for technicians qualified to maintain racks, cabling, and cooling systems. Automating routine physical maintenance is one way operators are looking to keep pace with that demand while controlling costs.

A Kinova Gen3 arm and a power-pressing finger

In one experiment, a Kinova Gen3 robotic arm is tasked with cutting power to servers, while another machine is being tested to swap networking cables.

At some sites, Meta has already deployed a device resembling a finger, or a pointy stick, that prods the power button on a Mac Mini or similar box to reboot it on the command of a remote human operator.

Kinova and ABB declined to comment, Watney did not respond, and Meta itself declined to comment on the testing.

A cable-swapping robot that works could take over 80% of some people's workloads, one Meta data center employee said. The estimate came from the worker, not from Meta, and the employee noted the machine still cannot match the speed of a human.

"We thought those of us performing the physical tasks were safe for a while, but not anymore," one worker said, adding: "It's coming for us all, unfortunately."

Some employees said Meta is building software that would allow it to hire cheaper, lower-skilled staff and shift some roles to cost-friendly hubs such as Denver. One said the company needs "smart hands" — people who can take instructions from AI without breaking anything.

Meta's capital spending runs to $145 billion

Eric Xu, senior manager for robotics at Meta, said at a conference last year that the company's long-term goal is to place robots in data centers, where they could speed up incident response, monitor the environment, and handle preventative maintenance.

The machines still need supervision and struggle with obstacles, short battery life, visual inspections, and tightly packed cabling. Past industry trials sometimes ended in failure, with robots crushing servers while performing simple tasks.

In May 2026, Meta began laying off about 8,000 employees, roughly 10% of its workforce, and data center staff were among those cut. The reductions were presented in part as freeing up cash for AI spending, with the company forecasting capital expenditure of $125 billion to $145 billion for the year — more than double the prior year.

The testing is part of a broader push into what the industry calls physical AI or embodied intelligence — machines that act in the real world rather than only generating text or images. How quickly such robots move from supervised pilots to routine deployment will depend on whether they can overcome the reliability limits Meta's own experiments have surfaced.

ACE Robotics chairman Wang Xiaogang said embodied AI could have a "ChatGPT moment" by the end of 2027. Cryptopolitan previously reported that Unitree founder Wang Xingxing put a similar leap two to three years out, and as long as five to ten years if progress stalls. He likened it to a robot doing about 80% of the work in an unfamiliar environment.

Lawmakers and economists have also begun weighing taxes on robots to offset automation-driven job losses.