Meta Tests Robots for Data Center Cable Maintenance
Meta is piloting the use of robots in data centers to replace network cables, reboot servers, reset components, and conduct inspections. The hardware comes from Watney Robotics, Kinova, and ABB, but human oversight is still required.
Current and former employees informed the media that since June last year, the Altoona facility in Iowa has been testing the Watney dual-arm robot for some wiring tasks; the Prometheus campus in New Albany, Ohio, uses ABB's wheeled robot with scissors lift and six-axis arm for component resetting; the Kinova Gen3 robotic arm is being evaluated for powering servers on and off. Some data centers have introduced a device resembling a finger or a stick, which is operated remotely to press the power button on devices like Mac Minis. If the cable replacement robots mature, a frontline employee estimates they could replace about 80% of the workload of certain positions, though this figure is a personal estimate rather than a company forecast, and the robots are still slower than humans. Internally, it is acknowledged that the existing robots cannot handle high-density cable bundles like the NVIDIA GB300; visual recognition, battery life, obstacles, and dense cabling remain shortcomings, and cabinets and maintenance processes are designed for human use, making modifications costly. Kinova and ABB declined to comment, Watney did not respond, and Meta refused to comment on the testing itself.
Spokesperson Francis Brennan stated that the U.S. is experiencing the largest infrastructure wave since World War II, with a severe shortage of skilled workers. "What we need is more workers, not fewer," he said, noting that the company is hiring and offering free training for electricians, mechanics, plumbers, and other trades, and is connecting graduates with contractor positions in states where it has facilities, such as Louisiana, Ohio, Indiana, and Texas. Employees are concerned that the demand for skilled technicians will decrease, leaving remaining tasks to low-wage positions operated by generated instructions. On the construction side, there are also heavy pile drivers and other site robots, which are not on the same production line as data center operations. Capital expenditure forecasts have been set at around $125 billion to $145 billion, with both data center expansion and skilled labor shortages existing simultaneously.
The pilot does not change the daily scheduling but alters the assumptions about human-machine ratios for the next facility. The 80% figure is an upper limit imagined after breaking down the skilled cable replacement into supervised repetitive actions, not an official formula for layoffs.
In market mechanisms, this is a hedge of operating costs against capital expenditures, not a robot company IPO event. The buyer is Meta, which aims to control operational costs per unit of computing power, while the sellers are suppliers of industrial arms and inspection platforms. Funding is shifting from data center OPEX to equipment depreciation and remote supervision roles. The beneficiaries are automation manufacturers that can standardize cable replacement and resetting tasks, while the burden falls on technicians who sell their skilled labor by the hour. The company publicly seeks workers while privately tests machines, with two narratives running parallel: the shortage is used to justify training academies, while the pilot is used to explain long-term staffing flexibility. The GB300 cable bundle proves that the physical layer is still designed for human hands, with robots initially handling power buttons and standard network cables, not all fault tickets.
In supplementary remarks, employees stated, "We thought those doing physical labor could still be safe for a while, but that's no longer the case."
Source: Public Information
ABAB AI Insight
Meta's introduction of robots into data centers shifts the OPEX of AI factories from technician labor hours to equipment depreciation. Capital expenditures have reached a level where they cannot hire enough workers, and cable replacement is one of the dirtiest, most repetitive tasks that can easily be automated. Watney, Kinova, and ABB provide external arms, not self-developed humanoids; the pilot locations, Altoona's old building and the new Prometheus campus, aim to replicate validated processes in the next facility. The 80% figure comes from employees breaking down the action of "cable replacement," not from an official model of the entire data center staffing.
The capital path is driven by the infrastructure wave and skilled labor shortages pushing for automation. Free training and hiring commitments address the start of work, while robots handle monitoring. The two are not contradictory: construction needs electricians and fiber optic workers, while operations want to reduce reliance on skilled cable replacement workers. Corning's multi-billion dollar fiber optic solution addresses glass supply but does not solve the issue of misaligned ports at 3 AM. The remote pressing of power buttons has outsourced the lowest skill actions to sticks, while robotic arms merely upgrade the sticks into joints capable of cable replacement. Employees' concerns about lower-paid supervisory roles involve leaving decision-making to software while relegating manual tasks to low-cost time zones.
The analogy is to Amazon's Kiva after the warehouse shelves, not Boston Dynamics' showcase. The data center industry is currently mismatched, with chips, power, and cooling expanding exponentially while labor expands according to apprenticeship cycles. Cabinets are still designed for human dimensions, and the GB300 cable bundle is evidence of this mismatch. Whoever first redesigns cabinet layouts will be the first to make arms usable.
Structural changes involve technological replacements stuck at physical interfaces. The pricing power of repetitive actions shifts from technician hourly rates to robot utilization rates. The mechanism is: standard network cables and power buttons are consumed first, while high-density custom cable bundles keep humans employed; supervisory software turns remaining personnel into execution layers; training academies expand entry numbers while simultaneously lowering the premium on individual skills. Job creation occurs in construction, while the elimination of skilled roles happens in operations—the time lag allows the company to simultaneously claim "we need more people" and test for 80% replacement.
ABAB News · Cognitive Laws
In a year where infrastructure needs workers, operations are already testing machines that require fewer workers.
80% replacement starts with the most standard network cable, not the entire data center.
Cabinets designed for human hands will prevent robotic arms from accessing high-density cable bundles.