Meta’s Data-Center Robots Could Automate Tasks—But Not Replace Technicians Yet

Meta is testing physical robots inside data centers in Altoona, Iowa, and New Albany, Ohio, to handle some of the repetitive work normally performed by technicians. The systems can plug in or replace cables, power-cycle servers, reseat hardware, and inspect equipment, according to reporting based on current and former workers. The vendors reportedly include Watney Robotics, Kinova, and ABB. 1718

The trials show where automation may enter data-center operations first: standardized, repeatable actions. They do not yet show that Meta can run large facilities without skilled human workers.

What Meta’s robots are being tested to do

The reported tasks include:

• Connecting and swapping networking cables
• Restarting servers or cutting and restoring power
• Reseating components
• Pressing controls and inspecting equipment
• Performing other routine maintenance actions

One reported experiment evaluates a Kinova Gen3 robotic arm for server power-cycling, while another system is intended to handle network-cable replacement. 18

These are attractive targets for automation because they are repetitive and occur inside a controlled environment. But data centers are also dense, safety-sensitive workplaces where a robot must identify the correct component, reach it accurately, deal with obstacles, and respond appropriately when a routine procedure does not go as expected.

What does the 80% estimate mean?

A Meta worker estimated that a successful cable-swapping robot could eventually take over as much as 80% of some technicians’ workloads. 18

That figure should be read narrowly. It is an employee’s estimate about the share of work associated with certain repetitive tasks—not a measured result, a promise from Meta, or evidence that 80% of data-center technicians could be eliminated.

A robot might perform many cable changes while still requiring people to prepare jobs, supervise operations, diagnose failures, handle exceptions, maintain the machines, and complete work the robot cannot perform. In other words, automating a large portion of a workload is not automatically equivalent to automating an entire role.

Why human technicians are still needed

The reported limitations point toward a technician-and-robot model rather than a fully autonomous data center. The machines remain slower than people in some work, require supervision, and can struggle with dense cabling, visual inspection, and navigation.

The distinction matters because data-center work is not limited to pressing buttons or moving a cable from one port to another. Technicians also have to interpret unfamiliar problems, work around changing infrastructure, and make judgments when equipment, layouts, or procedures do not match the expected pattern.

Meta has also indicated that it continues to need data-center workers. That position is consistent with the current state of the trials: robots may reduce the amount of repetitive maintenance labor required, while people remain responsible for deployment, repairs, oversight, and difficult interventions.

How the robotics trials connect to Project OT

The robot program is part of a broader automation strategy at Meta. Under the reported Project OT, short for “Organization Transformation,” executives explored making the company more “AI native” by assigning routine work to AI agents and having smaller human teams supervise them. Internal planning scenarios considered reducing the size of some teams by as much as 60%. 12

The physical-robot experiments apply a similar idea to infrastructure. Software agents would automate digital workflows; robots would automate selected physical workflows in the facilities that support Meta’s AI systems.

But Project OT did not proceed at its most ambitious scale. Meta carried out roughly 8,000 layoffs, or about 10% of its workforce, while the larger follow-up reduction wave was canceled. Reporting attributed the retreat to employee resistance, unreliable or disruptive agent behavior, and weaker-than-expected productivity gains. 13

One internal comparison reportedly showed a 220% increase in code changes but only a 36% increase in new user-facing features. The gap illustrates why more automated activity does not necessarily produce a proportional increase in useful output. 1

Why Meta has a financial incentive to automate data centers

Meta is building out the infrastructure needed for its AI ambitions at enormous cost. The company reported $60.8 billion in second-quarter 2026 revenue and $31.1 billion in quarterly capital expenditures, while forecasting $130 billion to $145 billion in full-year capital spending.

That spending creates a clear incentive to reduce the labor and operating costs associated with a rapidly expanding data-center footprint. Robots could eventually help Meta perform standardized maintenance with fewer people or allow technicians to supervise more equipment.

The financial incentive, however, is not proof that the technology is ready to replace data-center specialists wholesale. The current evidence supports a more limited conclusion: Meta is testing whether robots can absorb defined, repetitive tasks while human workers continue to manage the broader operation.

The likely near-term outcome

Meta’s experiments are best understood as task automation, not complete job replacement. Cable handling, server resets, and other routine actions may be suitable for machines if the systems become reliable, fast, and economical enough to deploy widely.

For now, the reported need for supervision, the robots’ physical limitations, and Meta’s continued hiring of data-center specialists point to a hybrid workforce. The central question is not whether robots can perform any technician tasks. They can. The unresolved question is how much of the surrounding judgment, exception handling, and operational responsibility can be automated without reducing reliability.

That is also the lesson from Project OT: producing more automated output is not enough. Automation has to deliver dependable, valuable work before it can safely substitute for the people responsible for the result.