Meta is testing dual-armed robots that swap network cables inside server racks at its Altoona, Iowa data center. The machines come from Watney Robotics. They work under human supervision. And they move slower than the technicians they might one day replace.
Those details surfaced late last month in a report from WIRED. The story drew on accounts from current and former Meta employees who described an expanding robotics program aimed squarely at the physical work that keeps hyperscale facilities running. One anonymous data center worker offered a blunt assessment. “We thought those of us performing the physical tasks were safe for a while, but not anymore.” The worker added, “It’s coming for us all, unfortunately.”
The trials mark a shift. For years Meta deployed simpler automation inside its facilities. Self-driving tugger robots now move heavy server racks across floors in Iowa and Virginia. Wheeled inventory bots scan equipment and flag failures. Those systems operate with fewer complications. The new generation of arms and grippers must handle dense cabling, latch mechanisms, and hot-swap components without damaging million-dollar hardware.
At Altoona two dual-armed Watney robots have performed cable swaps since June 2025. A Kinova Gen3 arm is under evaluation for power cycling servers, essentially cutting electricity or restarting machines on command. Over at the Prometheus campus in New Albany, Ohio, ABB robots mounted on four-wheeled bases with scissor-lift risers and six-axis arms test the task of reseating hardware inside racks. Each pilot still requires oversight. Battery recharges create downtime. Complex cable bundles tied to the latest NVIDIA GB300 systems remain beyond current robotic dexterity.
Eric Xu, senior manager for robotics at Meta, laid out longer-term ambitions at a conference last year. Robots, he said, could accelerate incident response, watch environmental conditions, and manage preventative upkeep. The comments, first highlighted in the WIRED piece, align with the company’s broader push to control costs as its AI infrastructure bill climbs. Meta expects to spend as much as $145 billion on capital expenditures this year, much of it tied to data centers and chips needed for training and running models such as Llama.
The scale is staggering. Meta operates or has planned more than 20 AI-focused data centers with a combined capacity approaching 16 gigawatts, according to the AI Data Center Index. Its Hyperion project in Louisiana alone targets five gigawatts once fully built out, an area roughly the size of Manhattan. Powering those sites demands enormous renewable and nuclear commitments. It also demands labor. Or it did.
Even construction of supporting solar infrastructure now involves robots. Ten 72-ton autonomous machines from Built Robotics drive nearly 1,000 steel piles a day at a Louisiana solar site tied to Hyperion. Each machine works up to 12 hours, handling 200-pound beams in knee-deep mud where human crews hesitate. Built Robotics CEO Noah Ready-Campbell told Business Insider the robots let crews focus on higher-value work while machines tackle the dirtiest sections. “The robots don’t care,” he said.
Yet the maintenance robots inside the finished buildings spark sharper concern. A successful cable-swapping system could absorb up to 80 percent of certain technicians’ workloads, one Meta employee estimated to WIRED. That figure is not an official projection. It reflects anxiety among staff who once assumed their hands-on skills offered protection from automation. Data centers were supposed to bring stable, well-paid jobs to rural communities. Now some of those positions look vulnerable before they fully materialize.
Meta pushes back on that narrative. A company spokesperson told several outlets, including Decrypt, that America faces a serious shortage of skilled infrastructure workers. “We need more workers, not fewer,” the statement read. The company continues to hire and train despite the robotics experiments. Still, the pilots suggest a future in which fewer humans walk the aisles between racks. Robots could monitor temperatures, swap failed optical transceivers, and reset unresponsive nodes with minimal intervention.
Challenges remain. Server rooms were engineered for people. Tight spaces, reflective surfaces, and delicate connectors frustrate even advanced vision systems. Current robots struggle with visual inspections in some conditions. They require recharging. And they cannot yet manage every cable type without human help. Those limits explain why the trials emphasize supervision and why full deployment sits years away.
Even so, the direction is clear. Hyperscalers confront exploding demand for compute. Labor markets cannot scale at the same speed. Automation offers one answer. Meta is not alone in exploring the option, but its willingness to test multiple vendors across live facilities gives it an early lead in gathering real-world data.
A Futurism article published today captured the irony. Data centers have long been sold as major job creators. The arrival of maintenance robots undercuts that claim. Physical tasks once considered safe from AI now face the same pressures that reshaped software engineering and content creation.
Recent discussions on X echo the tension. One post noted that robotics could replace more jobs than many expect, pointing to Meta’s tests alongside heavy equipment makers training workers for autonomy. Another highlighted the $1.6 billion valuation of Lyte, a company building specialized sensors and models for high-density server manipulation rather than general-purpose humanoids. The market appears to favor targeted solutions over flashy bipeds when the stakes involve keeping racks of GPUs online.
Meta has not disclosed timelines for wider rollout. It has not quantified expected headcount changes. What is evident is the experimentation. Cable-swapping bots in Iowa. Power-cycling arms on test benches. Lift-equipped platforms reseating components in Ohio. Each test feeds engineering decisions that will shape the next wave of facilities.
The irony lands hardest for the technicians. They keep the machines that power the AI models. Now those same models help design the robots that may keep the machines instead. Progress rarely arrives without trade-offs. Meta is calculating that the savings and speed will justify them.
Whether the broader industry follows depends on results still being gathered in climate-controlled rooms far from public view. For now the robots remain slow, supervised, and limited. But they are there. And they are learning.
Meta’s Robot Technicians Take On the Data Center Floor first appeared on Web and IT News.