Meta is putting robots to work inside its data centers, testing machines that can swap network cables, power-cycle servers, and reseat hardware components, tasks that have always required a human technician standing in the aisle. The pilots, reported first by Wired and confirmed by current and former Meta workers, mark one of the clearest signs yet that the AI infrastructure boom is starting to reshape physical labor, not just software jobs.
The company is running trials with hardware from three outside vendors, Watney Robotics, Kinova, and ABB, at two of its largest US campuses. One Meta data center worker estimated that a successful cable-swapping robot could eventually eliminate up to 80% of the workload in some technician roles, according to Wired’s reporting. That single number is why this story has spread fast across tech and labor circles this week.
What Meta Is Actually Testing, and Where
Meta’s robotics pilots are not a single project. They are a set of parallel experiments spread across different facilities, each targeting a different physical task that data center technicians handle today. At the company’s Altoona, Iowa campus, one of Meta’s largest, workers have been testing two dual-armed robots from Watney Robotics since June 2025, using them for cabling work such as unplugging and replacing network cables between servers. At the Prometheus campus in New Albany, Ohio, Meta’s newest build, four-wheeled ABB robots mounted with a scissor-lift riser and a six-axis arm are being evaluated for reseating hardware components inside server racks.
Separately, Meta is evaluating whether a Kinova Gen3 robotic arm can handle power-cycling, cutting and restoring electricity to unresponsive servers so they reboot without a person walking over to the rack. None of these systems is running unsupervised. Wired and TechRepublic both describe the robots as slower than skilled human technicians and dependent on people to reposition them, open cage doors, or double-check camera output before a task is marked complete.
The Robots Meta Already Runs at Scale
The Watney, Kinova, and ABB trials build on automation Meta has already deployed. The company has used self-driving tugger robots to move heavy server racks between locations, and wheeled inventory robots that scan barcodes and equipment for stock checks and failure inspections. Those two systems are now running at multiple Meta sites, including facilities in Iowa and Virginia, according to people familiar with the projects cited by Wired. A simpler device, one that looks like a mechanical finger or a pointy stick, can press a physical power button on a Mac Mini or similar box to force a remote reboot when a human operator triggers it.
Even these simpler, already-deployed robots have real limits. The inventory robot’s camera only reads in grayscale, so it can’t distinguish a green status light from a red failure indicator, meaning a person still has to walk over and check. It also struggles with corners and with cables that cross its wheeled path, and because Meta doesn’t have many units, they have to be manually moved between buildings by a technician who opens doors and drives them by remote control.
Meta’s Data Center Robot Pilots at a Glance
| Vendor / System | Task | Location | Status |
| Watney Robotics (dual-arm robot) | Network cable swapping | Altoona, Iowa | Testing since June 2025, supervised |
| Kinova Gen3 (robotic arm) | Server power-cycling | Under evaluation | Pilot stage |
| ABB (wheeled platform + six-axis arm) | Reseating hardware components | Prometheus campus, New Albany, Ohio | Under evaluation |
| In-house tugger robot | Moving heavy server racks | Multiple sites, incl. Iowa, Virginia | Deployed |
| In-house inventory robot | Equipment scanning, failure inspection | Multiple sites | Deployed, limited by grayscale camera |
| Remote power-button device | Forcing server reboots | Select facilities | Deployed |
Why Meta Is Doing This Now
The timing lines up with how much Meta is spending to build AI infrastructure. The company has guided investors toward roughly $115 billion to $145 billion in 2026 capital expenditures, the bulk of it tied to AI data centers, based on 2026 capital-spending estimates compiled across financial press coverage. Robots that can keep a growing footprint of racks and cabling running with a flatter headcount curve are a direct lever on that spending, especially in regions where qualified data center technicians are scarce relative to the pace of new construction.
Wired reported that Eric Xu, a senior robotics manager at Meta, told a conference audience last year that the company’s long-term goal is to put robots in data centers to speed up incident response, monitor conditions, and handle preventative maintenance, adding that broader industry collaboration will be needed to get there. That framing matches what the pilots are testing today: reactive fixes like power-cycling and reseating, not yet the full range of skilled troubleshooting a technician performs.
The 80% Number and What Workers Are Saying
The figure driving most of the reaction to this story is the 80% estimate. It came from a Meta data center worker speaking to Wired, not from a company projection, and it applies to a specific scenario: a fully working cable-swapping robot taking over the bulk of that one task, not an entire job category. Still, the sentiment inside Meta’s data center teams has reportedly turned uneasy. Wired described worker group chats in Altoona taking a demoralized tone, with some employees telling each other they expect to be replaced within a few years regardless of how slowly the robots improve.
Two workers cited by Wired also said Meta is building AI software and other tools they believe will let the company hire less specialized, lower-paid workers and centralize some roles in cheaper labor markets such as Denver, Colorado. That claim reflects worker perception reported by Wired rather than a confirmed Meta policy, and Meta has not confirmed a hiring or wage strategy tied to the robotics pilots.
Meta’s Official Response
Meta declined to comment directly on the specifics of the pilots described by Wired. A company spokesperson, Francis Brennan, said in a statement reported by Wired that the US is in the middle of its biggest infrastructure boom since World War II and that a shortage of skilled workers means Meta needs more people, not fewer. Meta has also launched a training program this year aimed at teaching thousands of people annually, regardless of background, electrical, mechanical, and plumbing skills at no cost, and it has guaranteed employment in states including Louisiana, Ohio, Indiana, and Texas for those who complete it, partnering with building trade unions to fund apprenticeships.
How Microsoft, Google, and Amazon Compare
Meta is not the only hyperscaler experimenting with physical automation, and its approach differs from its three closest rivals in what it is trying to solve. Microsoft set up a dedicated data center automation and robotics team in October 2023 and has since detailed the CRAFT (Creative Robotics and Future Technologies) project through Microsoft Research, which explores dexterous robots that can handle optical transceivers, clean fiber connections, and install cabling as part of a broader push toward what the company calls self-maintaining data centers. Microsoft’s Ashley Llorens, a managing director at the company’s research accelerator, told Wired that going after economically valuable physical tasks, like replacing a faulty server rack, is absolutely something Microsoft is pursuing.
Amazon’s approach so far has centered on a different part of the hardware lifecycle. AWS partnered with robotics startup Molg on AI-powered robots that visually inspect used data center electronics, assess whether components still function, and carefully disassemble them for reuse or recycling, an application ABB itself detailed in a joint announcement on robotic microfactories. That’s a lifecycle and sustainability play rather than a direct swap for on-site maintenance technicians.
Google’s public robotics activity has leaned toward the industrial side. The company has tested robots for handling hard drives and equipment inside its own data centers, and Google DeepMind announced a collaboration with Agile Robots to pair its Gemini Robotics foundation models with Agile’s industrial hardware, a deal covered by CNBC that points toward AI-controlled robotic reasoning aimed at manufacturing and adjacent high-value settings, an approach that could plausibly extend into Google’s own facilities over time.
Hyperscaler Capex and Robotics Strategy, 2026
| Company | 2026 Capex (reported estimates) | Primary Robotics Focus | Automation Stage |
| Meta | ~$115B–$145B | Watney, Kinova, ABB maintenance pilots | Supervised trials, not production-wide |
| Microsoft | ~$150B annualized run-rate | CRAFT research, dexterous cabling robots | Research and early prototyping |
| Amazon / AWS | ~$200B (up from $131.8B in 2025) | Molg partnership for e-waste/recycling | Deployed for lifecycle processing |
| ~$175B–$185B (up from $91.4B in 2025) | Hard-drive handling; DeepMind-Agile Robots tie-up | Early deployment plus industrial R&D |
The pattern across all four companies is consistent: massive capital spending on AI infrastructure paired with early, cautious steps into physical robotics, none of it yet running without human oversight at scale.
From Script-Based Automation to Physical Robots
Data center automation used to mean something almost entirely digital. Configuration management tools, autoscaling policies, and remote monitoring dashboards let a small team manage thousands of servers without touching a single rack. What Meta, Microsoft, Google, and Amazon are testing now pushes automation into the physical layer for the first time at meaningful scale, the actual swapping of cables, the actual power-cycling of a dead node, the actual reseating of a card that’s thrown an error.
That shift has been slow to arrive because the economics didn’t work until recently. Robots capable of delicate, precise movement in tight server aisles were expensive, and early industry trials sometimes ended with a robot crushing a server while attempting a simple task, according to a person briefed on the experimentation cited by Wired. Cheaper hardware and far more capable AI models for perception and control have changed that math over the past two to three years, which is part of why a wave of startups, including Watney and companies like Exclaim Robotics, have moved into this space just since 2023 and 2024.
A simple way to think about how these pilots slot into existing operations is a task hand-off between software monitoring and a physical actor:
monitoring_system detects: server_unresponsive(rack_id)
-> creates maintenance_ticket(task="power_cycle", priority="high")
-> human_technician OR robot_arm assigned
-> if robot: supervised_execution, camera_verification
-> if fail: escalate_to_human_technician
-> ticket closed, uptime loggedThat illustrative flow is not a description of Meta’s actual internal tooling, but it captures the pattern reporters have described: software still triggers the work order, and a robot is only one of two possible executors, with a human as the fallback whenever the robot can’t finish the job.
Why Robots Still Lose to Human Technicians
The gap between a demo and a dependable data center worker is still wide. Helen Oleynikova, CEO of Exclaim Robotics, a startup building prototypes to replace broken parts and reset cables in data centers, told Wired that the field has seen a lot of pilots and demos but no provable working solution yet. That’s a notable admission from someone whose business depends on this technology succeeding.
Battery Life and Downtime
Workers at Meta have complained about how much time the existing robots spend recharging, cutting into the hours they’re actually available for tasks. Because Meta doesn’t have many units at each site, moving a robot from one building to another still requires a person to open doors and steer it by remote control, which undercuts some of the labor savings the robots are meant to deliver.
Equipment Designed for Human Hands
Meta has also acknowledged internally that its robots aren’t yet equipped for the dense, intricate cabling required to run high-end AI compute like an Nvidia GB300 rack, according to Wired. A former Meta employee put it plainly to Wired: everything in a data center has been designed for human hands to make repairs fast, and redesigning that for robots will take time. That single line probably explains the pilot-stage caution better than any capex figure does.
Market Impact and What Investors Are Watching
None of the four hyperscalers has broken out a stock-price reaction tied specifically to data center robotics, and that’s expected this early in the pilot stage. What investors are tracking instead is capex efficiency, whether the hundreds of billions of dollars going into AI infrastructure each year can be run at a lower marginal operating cost per unit of compute. Robots that cut technician headcount growth relative to rack growth are one lever on that ratio, alongside chip pricing, power costs, and cooling efficiency.
Tax Breaks and Local Politics
There’s a local political angle too. Altoona, Iowa, where Meta opened its largest campus in 2014, still has several years left on a property tax break that the city has treated as a worthwhile trade for local jobs and prestige. Altoona Mayor Dean O’Connor told Wired he hadn’t given much thought to robots displacing jobs at the campus, framing automation as something the city will deal with as it comes. As communities that host data centers reconsider tax incentives and push for new rules on the industry, robots quietly taking over physical maintenance work could become a bigger part of that debate than it has been so far.
Five Predictions for Data Center Robotics
- Meta, Microsoft, Google, and Amazon will each publish at least one additional named robotics vendor partnership before the end of 2026, as pilots move from single campuses to two or three sites each.
- Cable-swapping and power-cycling will reach limited production use before hardware reseating does, since reseating requires finer force control around delicate connectors.
- Expect at least one high-profile incident, a robot damaging a server or rack, to surface in trade press within the next 12 months as pilots scale beyond tightly supervised test cells.
- Data center technician hiring will keep growing in absolute terms through 2026 and 2027 even as robotics pilots expand, because new campus construction is outpacing any near-term automation gains.
- Robotics vendors serving this niche, including Watney, Kinova’s industrial division, ABB, and newer entrants like Exclaim Robotics, will see a wave of new funding rounds as hyperscalers signal willingness to pilot outside hardware rather than build everything in-house.
What Happens to Data Center Jobs From Here
The most honest answer, based on what’s been reported so far, is that nobody knows yet, including the people inside Meta running the pilots. The robots work, in a limited sense, but they’re slower than the people they’d replace, they need supervision, and they still can’t handle the messiest, most time-sensitive repairs. What’s changed is that the question is now being asked seriously at four of the largest technology companies in the world, at the same moment those companies are spending more on physical infrastructure than at any point in their history.
For now, Meta’s position is that it needs more skilled workers, not fewer, and its training and hiring-guarantee programs in states like Ohio and Texas back that up on paper. Whether that holds if a cable-swapping robot actually gets fast enough to match a human technician is the question this story leaves open, and it’s one the whole data center industry will be watching Meta answer first.
Frequently Asked Questions
What robots is Meta testing in its data centers?
Meta is piloting a dual-arm robot from Watney Robotics for cable swapping, a Kinova Gen3 robotic arm for server power-cycling, and ABB robots mounted on wheeled platforms for reseating hardware components. It also already runs in-house tugger robots for moving server racks and wheeled inventory robots for equipment scanning.
Where are Meta’s robot pilots taking place?
The Watney cabling robots are being tested at Meta’s Altoona, Iowa campus. The ABB robots are being evaluated at the Prometheus data center campus in New Albany, Ohio.
Could Meta’s robots really replace 80% of technician workloads?
That 80% figure came from one Meta data center worker’s estimate of what a successful cable-swapping robot could achieve for that specific task, as reported by Wired. It is not an official Meta projection and does not describe the company’s entire technician workforce.
Has Meta confirmed it will cut data center jobs because of robots?
No. Meta declined to comment on the specifics of the pilots and has said publicly, through spokesperson Francis Brennan, that it needs more skilled workers, not fewer, citing a broader US infrastructure labor shortage.
Are Google, Microsoft, and Amazon doing the same thing?
All three have their own data center robotics efforts, but with different priorities. Microsoft has a dedicated robotics research team and its CRAFT project on self-maintaining data centers. Amazon’s AWS focuses on robots for recycling and reusing used electronics through a partnership with Molg. Google has tested robots for hard-drive and equipment handling and has a robotics foundation model partnership with Agile Robots through DeepMind.
Why are robots slower than human technicians right now?
Reported limitations include battery recharge downtime, difficulty maneuvering around cables and corners, cameras that can’t reliably distinguish status-light colors, and hardware that was originally designed for human hands rather than robotic grippers. Meta has also said its robots aren’t yet equipped for the dense cabling required by high-end AI servers.
How much is Meta spending on AI data centers in 2026?
Meta has guided toward roughly $115 billion to $145 billion in 2026 capital expenditures, according to capital-spending estimates reported across financial media, with the bulk of that spending tied to AI infrastructure and data centers.
Is this the first time Meta has used robots in its data centers?
No. Meta previously deployed self-driving tugger robots for moving server racks and wheeled inventory robots for equipment tracking, both of which are already running at multiple sites. The Watney, Kinova, and ABB pilots go further by targeting actual hands-on maintenance tasks rather than logistics and inspection.
