Meta Tests Data Center Robots as Human Workload Comes Under Pressure

Meta is testing robots for physical data center work, despite building an employment case around its expanding American infrastructure. The machines reportedly swap cables, restart servers, inspect equipment, move racks, and reseat components. One worker estimated that successful cable automation might replace up to 80 percent of certain workloads.

That estimate is not an official Meta forecast. It comes from an employee interviewed by WIRED, whose reporting formed the basis of a widely circulated Google News story. The robots also remain slower and less independent than experienced technicians.

Still, the tests move automation beyond software and into the physical aisles supporting Meta’s AI systems. Microsoft, Google, and Amazon have explored related applications. Meta now faces a difficult question about whether infrastructure growth creates technical jobs or gradually converts them into automated workflows.

This is not simply a story about a robot pressing a button. It concerns who controls the last manual step in an increasingly automated computing system. It also tests whether today’s data centers, designed around human hands, can become machine-serviceable without sacrificing reliability.

Meta Has Moved Data Center Robots Into Real Facilities

Meta is no longer limiting its data center robotics work to controlled demonstrations.

According to robotics testing documented by WIRED, several machines are operating or undergoing trials inside Meta facilities. The work spans sites in Altoona, Iowa, New Albany, Ohio, and other locations.

The company has reportedly tested a Kinova Gen3 robotic arm for power cycling. Power cycling means physically turning equipment off and on to recover it from certain failures.

A separate experiment involves Watney Robotics machines with two arms. These machines attempt cable work under human supervision at Meta’s Altoona campus.

That trial has reportedly been underway since June 2025. The machines cannot yet match the speed of human technicians, but they are working in an operational environment.

Meta is also testing ABB equipment at its Prometheus campus in New Albany. One reported configuration combines a wheeled platform, a scissor-lift mechanism, and a six-axis robotic arm.

That design allows the arm to reach different positions within a rack. Its early assignments reportedly include reseating components, which involves removing or repositioning hardware to restore a reliable connection.

The variety of machines matters. Meta is not betting on one humanoid robot that performs every maintenance task. It is testing specialized configurations against particular operational problems.

Some tasks require mobility. Others require height adjustment, accurate visual inspection, controlled force, or the ability to manipulate flexible cables.

Meta had already introduced simpler automation before these newer trials. In 2023, the company showed a self-driving tugger that transported heavy server racks.

It also developed a wheeled inventory robot that reads barcodes. WIRED reported that these systems now operate at multiple sites, including facilities in Iowa and Virginia.

That progression reveals Meta’s practical development strategy. It first automated movement and inventory, where the environment can be structured around predictable routes.

Now it is testing manipulation. Manipulation is harder because cables bend, connectors differ, equipment blocks camera views, and mistakes can damage expensive hardware.

The Google News headline focused on robots taking over routine work. The more important change is that Meta has created operational testbeds where machines encounter the messiness of active facilities.

Altoona appears central to that process. The Iowa campus has historically served as a place where Meta evaluates infrastructure technologies before considering wider deployment.

That does not mean every machine will reach production. It means the company can now collect performance data inside the environment where automation must eventually prove itself.

The trials also connect physical robotics with Meta’s existing software automation. Meta has long used monitoring systems to detect failures and attempt remote repairs.

Its hardware remediation process can identify unhealthy servers, apply software fixes, and create technician tickets when physical intervention remains necessary. Robots target that final category.

A server reset sounds simple until software cannot complete it remotely. Someone must identify the correct machine, verify its state, reach the control, and avoid affecting neighboring equipment.

Every automated physical task closes part of that gap. The experiments therefore extend Meta’s software control system into the building itself.

Why Meta Is Automating the Last Manual Step

The business case grows stronger as Meta spends more money on infrastructure and runs more specialized hardware.

Meta expects its 2026 capital expenditures, including finance lease principal payments, to reach between $130 billion and $145 billion. That outlook appears in the company’s quarterly results.

The company recorded $31.08 billion in capital expenditures during the second quarter alone. Much of Meta’s investment supports servers, data centers, networking, and the systems required for AI development.

At this scale, maintenance is not a secondary concern. A faulty connection can strand expensive computing capacity until the right technician reaches the right rack.

Meta’s AI clusters also contain more varied hardware than a conventional fleet. The company has described more than 30 maintenance operations involving over 50 components.

It also performs thousands of disruptive AI host tasks each day, according to its capacity maintenance documentation. A disruptive task temporarily removes a host from productive work.

Most of those operations remain software-driven. However, a failed cable, loose component, blocked sensor, or unresponsive power control can still require a person.

Robots promise to shorten that final response. A machine already inside the facility could act when software escalates a physical repair ticket.

The economic benefit does not depend entirely on replacing workers. Faster intervention can return valuable hardware to service sooner and reduce delays during nights or weekends.

A robot also offers repeatability. It can record the force, position, timing, and outcome of each attempt, creating structured data for later analysis.

That information could improve both robotics software and future hardware design. Meta might discover which connectors, rack layouts, and repair procedures consistently defeat automation.

The pressure becomes especially intense in regions with limited pools of experienced technicians. Large campuses require electricians, network specialists, cooling experts, safety staff, and hardware technicians.

Meta says labor shortages already constrain construction. The company has launched training initiatives intended to expand the available workforce rather than shrink it.

Its LevelUp program with CBRE offers a four-week pathway for prospective fiber technicians. Meta says the initiative will train thousands of people and connect graduates with opportunities through its contractor network.

Meta also says it operates or is building 27 data centers in the United States. Since 2010, those projects have supported over 30,000 construction jobs and more than 5,000 permanent operational positions.

Those figures come from Meta’s workforce announcement, so they should be understood as company-reported numbers. They nevertheless show how heavily Meta links infrastructure expansion with employment.

Robotics complicates that message. Construction demand can rise while operating work becomes more automated.

The distinction between construction and operations is crucial. Building new campuses requires large temporary workforces, while keeping completed facilities running creates a smaller pool of continuing jobs.

Meta can therefore hire more people overall while automating individual duties. Both claims can be true, but their benefits may reach different workers and communities.

This is why the 80 percent workload estimate attracted attention on Google News. It translates a technical pilot into a recognizable labor question.

Yet workload reduction is not identical to job elimination. Technicians perform diagnostics, safety checks, escalation, planning, and unfamiliar repairs that current robots cannot reliably handle.

The immediate pressure falls on repetitive assignments. Cable replacement, inventory checks, basic resets, and routine inspection are easier to define than unpredictable troubleshooting.

Automation often begins by separating those assignments from a broader role. The remaining job then becomes more technical, more supervisory, or less numerous.

Meta’s central challenge is proving that labor scarcity, rather than labor substitution, drives the program. That proof will require deployment data, staffing trends, and transparent descriptions of redesigned roles.

Data Center Robotics Is Becoming a Hyperscaler Contest

Meta’s experiments are part of a wider contest to make physical infrastructure respond like software.

Microsoft Research is developing modular robots that manipulate optical transceivers and cables. Its work also covers fiber cleaning, installation, repair, decommissioning, and hardware reconfiguration.

The company describes a long-term goal of self-maintaining data centers. Under that model, software detects a fault and coordinates a physical response with limited human involvement.

Microsoft’s robotics program emphasizes co-design. The facility, hardware, software, and robot must evolve together instead of treating the machine as an aftermarket technician.

Google has also pursued physical infrastructure robotics. Its hiring has included specialists responsible for deploying mobile robots, guided vehicles, and multi-axis arms across hyperscale facilities.

Amazon has discussed robotics for handling and recycling data center components. Its extensive warehouse automation experience gives it relevant expertise, although server maintenance demands different levels of precision.

Meta’s current position combines an established data center fleet with an expanding robotics organization. It can test multiple commercial platforms without waiting for one general-purpose humanoid system.

That supplier approach reduces technical concentration. Watney, Kinova, and ABB offer different forms, control systems, and levels of maturity.

It also makes integration harder. Meta must connect each machine to facility maps, repair tickets, hardware records, safety policies, and access controls.

The main contest is therefore not Meta against one rival. It is automated maintenance against infrastructure built for people.

Humans can recognize a poorly labeled cable, move an obstruction, open a door, and adjust their grip without a formal instruction. Robots need those variations modeled or detected.

A data center also combines standardized racks with years of accumulated exceptions. Equipment generations differ, cable routes change, and repair histories produce small physical inconsistencies.

These conditions explain why cable manipulation has become a serious research problem. Cables deform, overlap, obscure connectors, and can transmit force to neighboring hardware.

A robot must identify the correct cable and understand its path. It must then control its grip without pulling an adjacent connection loose.

It also needs a recovery procedure when the connector resists. A trained technician can feel resistance and decide whether the angle, latch, or component is wrong.

Current robots can perform such actions under constrained conditions. Reliable deployment requires them to recognize when those conditions no longer apply.

Meta’s experiments suggest specialized robotics will arrive before fully autonomous facilities. A wheeled inspector may handle inventory, while a fixed arm performs controlled power operations.

A supervised dual-arm machine might address selected cable layouts. Technicians would manage exceptions and approve risky actions.

That model resembles automation in other industries. Machines first handle repeated tasks within defined boundaries, while people retain judgment and responsibility for irregular cases.

The bigger change could come from new facility design. If Meta standardizes robot-friendly connectors, wider routes, machine-readable labels, and automatic doors, performance should improve.

Hardware modules could include gripping points and alignment guides. Racks could expose controls where a robot can reach them without navigating dense cable bundles.

Such changes would also deepen vendor commitments. A robot-friendly Meta campus could require suppliers to deliver components that fit defined mechanical and data interfaces.

The most consequential Google News story may therefore arrive later. It would concern a data center designed around robotic service from the beginning, not a machine retrofitted into human space.

The Machines Still Fail Where Human Judgment Matters

The strongest argument against rapid deployment comes from the robots already working inside Meta’s facilities.

WIRED reported that Meta’s inventory robot cannot distinguish green indicator lights from red ones because its camera detects only grayscale. Human checks remain necessary for some failures.

The machine also struggles with corners and cables lying across its path. Moving it between buildings can require a person to open doors and guide it remotely.

Workers reportedly consider the inventory robot too slow. Meta explored drones as a faster option but did not continue with that approach, according to the same reporting.

Battery charging creates another constraint. A robot that spends extended periods unavailable cannot provide dependable incident response without backup machines or automated charging systems.

The Watney robots in Altoona reportedly remain slower than people. They also work under supervision, limiting any near-term labor savings.

Meta has internally recognized that some hardware presents harder problems. The dense cabling required by Nvidia GB300 systems reportedly remains beyond the robots’ current abilities.

That limitation directly challenges the scale argument. New AI systems increase the value of automation, but they can also make the physical work more demanding.

Robot accuracy must be evaluated against the cost of a mistake. Disconnecting the wrong cable or damaging a connector can affect expensive equipment and delay critical workloads.

A successful demonstration does not establish operational reliability. Meta needs evidence across thousands of attempts, multiple hardware generations, and unusual failure conditions.

The company has not publicly disclosed success rates, intervention frequencies, repair times, or damage events from these trials. It also declined to address WIRED’s detailed description of the testing.

Helen Oleynikova, CEO of Exclaim Robotics, offered a blunt assessment in the report. She said the industry has produced many pilots and demonstrations but no provable working solution.

That view does not establish that Meta’s program will fail. It defines the verification gap between a promising prototype and dependable infrastructure.

Data centers impose strict safety and availability requirements. A machine must coordinate with power controls, access systems, monitoring software, and technicians working nearby.

Cybersecurity matters as well. A robot authorized to reset servers or manipulate network connections becomes part of the facility’s control surface.

Operators must authenticate commands and prevent unauthorized physical actions. They also need reliable logs showing what the machine touched and why.

Software failure presents another problem. A robot can lose its map, misclassify equipment, or stop during a repair.

Human supervisors need safe procedures for recovering the machine without creating another outage. Emergency stops and restricted operating zones must work every time.

The labor narrative also contains uncertainty. The 80 percent figure represents one employee’s estimate, not measured displacement or a formal staffing plan.

Repeating it without that context would turn a warning into a prediction. The trials have not independently established that level of automation.

Google News distribution can flatten such distinctions. A provocative workload estimate travels faster than details about supervision, charging, grayscale cameras, or human-assisted doorways.

The responsible conclusion is narrower. Meta is actively pursuing automation for tasks that occupy meaningful portions of some technicians’ schedules.

Current machines remain dependent on people and controlled environments. Deployment speed will depend on whether Meta redesigns facilities and procedures around their limitations.

That transition could take years. It could also advance unevenly, with inventory and transport automating faster than cable work or complex repairs.

Workers should focus on the task mix rather than one percentage. The important questions concern which duties disappear, which responsibilities expand, and how many technicians each facility ultimately retains.

Meta’s Employment Promise Now Faces a Credibility Test

Meta must reconcile two messages that point toward different futures for data center labor.

The first message emphasizes job creation. Meta says America needs more skilled workers to build and operate the infrastructure behind AI.

Spokesperson Francis Brennan told WIRED that the country is experiencing its largest infrastructure boom since World War II. He argued that the shortage requires more workers, not fewer.

Meta’s training programs support that position. The company is preparing people for fiber, electrical, mechanical, and plumbing work while guaranteeing some graduates jobs through partners.

The second message comes through engineering. Meta’s robotics manager Eric Xu has described data center robots as a long-term goal for incident response, monitoring, and preventive maintenance.

These positions are not automatically contradictory. Robots might cover repetitive work while trained people handle construction, troubleshooting, safety, and complex equipment.

However, Meta has not explained how those roles will change. Its public workforce messaging emphasizes opportunity, while its robotics tests remain largely undisclosed.

That gap matters in communities considering data center projects. Local officials often evaluate employment alongside tax incentives, land use, electricity demand, and water consumption.

A facility can represent a major capital investment without producing a proportionate number of permanent jobs. Greater automation widens that difference.

The Altoona example illustrates the issue. Meta opened the Iowa facility in 2014, and the campus now employs hundreds of people.

The city accepted tax concessions partly because the project promised economic activity and recognizable technology employment. Robotics changes the assumptions behind future negotiations.

Communities will want to know whether a planned campus needs hundreds of permanent workers, dozens of specialized operators, or a smaller remote team.

They may also distinguish between construction jobs and long-term operations. A temporary building surge does not guarantee durable technical employment after completion.

Meta’s recent headcount decisions add another layer. The company reported $1.18 billion in severance expenses connected with a May 2026 headcount reduction.

That reduction did not specifically target data center technicians in the published results. Still, it weakens any assumption that rising infrastructure spending protects employment across the company.

The better test is measurable staffing. Meta can disclose how technician numbers change after robots enter production and whether affected workers move into higher-skilled roles.

Training outcomes also deserve scrutiny. Enrollment numbers matter less than completion, placement, retention, and wage progression.

Workers need access to the roles automation creates, including fleet supervision, robot maintenance, safety validation, and infrastructure integration.

Those positions may demand different credentials than traditional server work. A company cannot call the transition worker-friendly if existing technicians lack practical routes into them.

The experiments also affect vendors. Server, rack, and connector manufacturers may face pressure to make their equipment easier for robots to service.

Contractors could see routine maintenance agreements shrink while demand grows for robotics integration. Facility management companies may need teams that combine mechanical and software expertise.

Knowledge workers should notice this pattern too. Data center automation mirrors the way AI changes office roles by targeting tasks before replacing entire occupations.

A job can survive while its autonomy, required skills, workload, and headcount all change. That makes task-level evidence more useful than broad claims about jobs created or eliminated.

For readers following the story through Google News, the labor conflict is the durable angle. A robot arm is visible, but the operating model behind it will determine the real impact.

Three Signals Will Show Whether Meta’s Robots Are Ready

The next phase should be judged through deployment evidence, job design, and infrastructure changes rather than another polished demonstration.

The first signal is wider operational deployment. Meta has tested several machines, but it has not announced a fleet rollout across its data centers.

A meaningful expansion would include named facilities, defined tasks, and measurable performance. Repair completion rates and human intervention rates would reveal whether the technology saves time.

Cable-swapping performance deserves special attention. It combines high potential value with difficult manipulation and serious consequences when something goes wrong.

If Meta moves that work beyond supervised trials, the case for automated maintenance becomes stronger. Continued confinement to a few testbeds would support the skeptics.

The second signal is a change in staffing and job descriptions. Meta and its contractors will reveal priorities through hiring, training, and internal role design.

More openings for robotics technicians or centralized operators would indicate that work is shifting rather than simply disappearing. Falling site-level technician demand would suggest direct substitution.

The most informative measure will be technicians per unit of computing capacity. Meta can grow its workforce while sharply reducing the labor required for each facility.

Training programs should also publish placement and retention results. Those outcomes would test whether Meta’s employment commitment reaches workers after automation changes the task mix.

The third signal is robot-friendly infrastructure. New rack specifications, connectors, service APIs, machine-readable labels, or facility layouts would represent a deeper commitment.

Retrofitted robots must work around environments designed for people. Purpose-built infrastructure can remove doors, cable obstacles, visual ambiguity, and awkward access points.

Such design changes would strengthen the case that Meta expects automation to become standard. Their absence would suggest the company still views robots as experimental tools.

Competitor actions will provide another useful reference within these signals. Microsoft already describes self-maintaining data centers as a research objective.

If multiple hyperscalers adopt common mechanical interfaces, suppliers will have stronger incentives to redesign equipment. Shared standards could accelerate adoption beyond any single company.

What should readers do with the current claim? Treat it as evidence of direction, not proof of completed job replacement.

Meta has placed specialized robots inside real data centers and assigned them economically relevant work. That is more significant than a laboratory demonstration.

The machines still require supervision, charging, environmental assistance, and carefully limited tasks. Meta has not published enough operating data to establish their reliability or labor impact.

Watch the next infrastructure announcement with three questions. Did the robots move beyond test sites? Did technician roles change? Was the facility designed for machine service?

Those answers will matter more than another Google News headline. They will show whether Meta is adding tools for workers or rebuilding the data center around fewer human hands.