Meta is testing robots inside its AI data centers, and the hard question is not whether the machines look impressive. It is how much of the human work behind the AI boom Meta wants to stop paying people to do.
Walk into Meta's data center campus in Altoona, Iowa, and the future of AI infrastructure may look less like a glowing supercomputer than a robot arm trying to deal with a cable. According to WIRED, Meta has been testing machines from Watney Robotics, Kinova, and ABB to handle the physical jobs that keep server farms running: swapping network cables, power-cycling servers, moving racks, checking inventory, and reseating parts.
That is not a small experiment. It is the dull, practical work that sits underneath every big AI promise. Someone has to walk the aisle when a server fails. Someone has to find the right rack and pull the cable. Then someone has to put a hand on the hardware. Meta is now asking how much of that job can be handed to a machine.
WIRED reported on August 28 that Meta has been trying a pair of dual-armed Watney robots for cabling work at one Altoona building since last June. The robots are supervised by humans and still cannot work as fast as people. Still, they are inside a real data center, not sitting in a trade-show booth.
At Meta's newer Prometheus campus in New Albany, Ohio, ABB machines on four-wheeled platforms, with a scissor-lift-style riser and a six-axis arm, are being used to reseat parts inside racks. Meta has also evaluated a Kinova Gen3 robotic arm for power cycling, the simple but tedious act of cutting electricity to a server and bringing it back. Separately, WIRED described a much simpler finger-like robot that can press a power button on a Mac Mini or similar device when a human sends the command remotely.
Meta Closes a Second Loophole That Let Ray-Ban Glasses Record in Secret Meta is closing a second loophole that let Ray-Ban smart glasses secretly record video after users covered the privacy LED mid-capture, following a June fix for drilled-out lights. Regulators in France and Germany argue the real issue isn't the light at all, but the lack of any lawful basis for filming non-consenting bystanders. - how to prevent secret recording with smart glasses - ray ban glasses privacy light blocking loophole fix
That little finger matters.
It tells you where this starts. Meta does not need a humanoid robot walking around like a person to change the labor math. It needs a machine that can do one boring task reliably, then another, then another. The first version can be ugly and limited. If it works at 2 am, when a rack needs attention and a technician would otherwise be sent down the aisle, Meta will keep looking.
The Job Risk Is Already Visible #
The sharpest number in WIRED's report came from inside the workforce. One Meta data center worker estimated that a successful cable-swapping robot could replace up to 80 percent of some people's workloads. That is not a Meta forecast. It is not a vendor promise. It is a worker looking at the task list and doing the math.
You should take that distinction seriously. Companies often talk about automation as help for workers, and sometimes it is. Robots can take the repetitive jobs and the dangerous ones. That frees skilled technicians from wasting hours on basic resets. But when an employee says one machine could take most of a role's workload, the pro-worker framing starts to look thin.
Meta told WIRED it is investing heavily in training and hiring workers to build and operate its data centers. The company has also launched a program to train thousands of people a year in electrical, mechanical, and plumbing work, with employment commitments in states including Louisiana, Ohio, Indiana, and Texas for people who finish the training. Keep that fact in the story because it is real.
But do not let it soften the point too much. Training more workers and automating parts of their future jobs can both be true at the same time. Big companies are perfectly capable of needing more people during a buildout and fewer people per unit of capacity later.
Robots Still Have To Survive The Server Aisle #
Meta is nowhere near a fully robotic data center floor. WIRED reported that the machines still need human supervision, run into battery limits, struggle with obstacles, and have trouble with the dense cabling found in actual racks. The company's inventory robot reportedly cannot tell green and red indicator lights apart because its camera sees in grayscale. That is a very ordinary failure, which is exactly why it is useful.
China Forces Meta to Unwind Its 2 Billion Dollar Acquisition of Manus AI Beijing's NDRC ordered Meta to unwind its $2 billion acquisition of AI agent startup Manus, ruling that a Singapore relocation didn't shield the China-born company from Chinese jurisdiction. Manus confirmed on August 11 it's independent again, with affected user data set for deletion starting August 23. - China forces Meta to unwind acquisition - Meta's Manus AI deal unwound by
Robots fail in the details. A clean demo does not have the same mess as a live server row full of cables, corners, doors, batteries, alerts, and hardware built for human hands. One former Meta employee told WIRED that things have been designed for human hands forever to make repairs quick. Redesigning that world takes time.
Frankly, that is the story. Meta is not testing robots because the technology is finished. It is testing them because AI infrastructure has become so large that even small savings in repetitive physical work can matter across a giant campus. Meta has said Prometheus will deliver 1 gigawatt of capacity once complete, spanning several data center buildings in a single larger region. At that scale, a cable is no longer just a cable. It is a cost center repeated thousands of times.
Other big cloud and AI operators are looking in the same direction. WIRED noted that Microsoft, Google, and Amazon have also discussed or unveiled robotics work tied to data center operations. That does not make the outcome guaranteed. It does show that the physical layer of AI is now under the same pressure as software work: measure it and automate it, then cut the human steps out where you can.
The robots are slow today. Some of them need a person nearby. Some need a person just to get them through a door. Still, Meta is running the tests at Altoona and New Albany with three separate robotics vendors because it wants a practical answer, not a slogan. How much of the work behind the AI boom still needs a human hand?
Also read: Big Tech Booked $160 Billion in Paper Gains From AI Bets Last Quarter • Nvidia Is Closing In On A $12.9 Billion Deal To Buy Hugging Face • Marvell says its data center chip business will hit $18 billion by 2028