Meta has started running robots inside two of its data centers to see whether machines can take over the physical maintenance work that keeps its AI server farms online. According to a report published by TechRepublic on August 31, 2026, the company is piloting hardware from three separate vendors, Watney Robotics, Kinova and ABB, to swap network cables, reboot frozen servers and reseat hardware components. The tests are running at Meta’s Altoona, Iowa campus and at its newer Prometheus site in New Albany, Ohio, and they mark one of the most concrete public examples yet of a hyperscaler testing physical robots against human technicians in live AI infrastructure.
The story lands at a moment when Meta and its rivals are pouring unprecedented sums into AI data centers. Meta raised the low end of its 2026 capital expenditure forecast from $130 billion to $135 billion on October 21, 2026, producing a range of $135 billion to $145 billion, up from roughly $71 billion to $72 billion in 2025, largely to build out the compute needed to train and run its AI models. Against that backdrop, a pilot program testing whether robots can shave labor costs off routine maintenance is no longer a novelty experiment. It is a preview of how hyperscalers plan to run the next generation of data centers at a scale human crews may struggle to service alone.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
What the Robots Are Actually Doing Inside Meta’s Data Centers
The tasks under test are deliberately narrow. Reporting from TechRepublic, Cryptopolitan and TechTimes describes three categories of physical work being handed to machines: swapping and managing network cables, power-cycling servers by cutting power or pressing physical buttons, and reseating hardware components that have come loose or need replacing. None of this is glamorous work, but all of it is repetitive, physically demanding and constant across a data center campus that can house tens of thousands of servers.
Cable management alone consumes a large share of a technician’s shift. Networking racks in AI clusters carry dense bundles of fiber and copper connections that need to be traced, tested and replaced when a link degrades. Power-cycling is simpler in concept but happens constantly, since frozen or unresponsive servers need a hard reset to come back online. Reseating components, such as reinserting a memory module or a network card that has worked loose, requires precision but little judgment, which makes it a natural first target for automation.
Meet the Vendors: Watney Robotics, Kinova and ABB
Meta is not building its own robots from scratch. Instead, it is running what TechRepublic and TechTimes both describe as multi-vendor, task-specific pilots, evaluating different companies against different jobs rather than committing to a single robotic platform.
Watney Robotics, a San Francisco-based startup that closed a $21.1 million seed round on May 30, 2025, according to PitchBook data (a figure corroborated by a Form D filing showing $21,135,478 raised on June 2, 2025, per FormDs.com), has supplied dual-armed robots that Meta has reportedly been testing on cable-swapping tasks at the Altoona campus since June 2025. That early bet looks prescient in hindsight: Watney went on to close an $80 million Series A round in September 2026, according to Tao Media (reportedly announced on September 17, 2026 and co-led by Valor Atreides AI Fund and Hummingbird Ventures), and the company says it has been serving major hyperscaler customers since 2025. Meta has deployed two of Watney’s dual-armed units at Altoona, according to MLQ.ai, making it the longest-running of the three pilots, with WIRED reporting in August 2026 that the robots have been performing cable insertion and removal under human supervision since June 2025. Kinova, based in Quebec, is known for its Gen3 robotic arms, and Meta is using that hardware to cut power to servers and handle power-cycling. ABB, the Zurich-based industrial automation giant, is testing four-wheeled mobile platforms equipped with a lift mechanism and a six-axis robotic arm at the Prometheus campus, focused on reseating hardware components.
These are not identical players, though. Watney has moved fast up the funding ladder – Tracxn’s August 2026 snapshot put its total funding at $21.1 million across a single seed round, following a raise first announced in April 2025 per VCBacked.co and echoed in Wellfound’s job-listing data, but that figure is already outdated: Watney closed an $80 million Series A in September 2026, per Tao Media, taking its total funding past $100 million in September 2026, as Robotico reported. The company also says its robots have logged reliability above 99.99% across hundreds of thousands of operating hours and that its fleet now runs 24/7/365 inside customer facilities, claims Watney made itself in September 2026 rather than figures independently verified by Meta or outside auditors, and as of October 2026 the startup has not publicly identified a newer, numbered generation of its robot beyond the dual-armed unit already deployed at Altoona. ABB, meanwhile, brings decades of experience in industrial and manufacturing automation to the pilot. That contrast suggests Meta is hedging its bets, pairing a fast-scaling but young robotics company with an established industrial giant, and treating data center maintenance as a materials-handling problem it can borrow solutions for, rather than something that requires ground-up humanoid robotics from a single vendor.
| Vendor | Headquarters | Robot Type | Task Assigned | Test Site |
|---|---|---|---|---|
| Watney Robotics | San Francisco, US | Dual-armed robot | Network cable swapping | Altoona, Iowa (since June 2025) |
| Kinova | Quebec, Canada | Gen3 robotic arm | Server power-cycling / cutting power | Meta data centers |
| ABB | Zurich, Switzerland | Four-wheeled mobile platform, six-axis arm, lift | Reseating hardware components | Prometheus, New Albany, Ohio |
Inside Altoona and Prometheus: Where the Machines Are Working
Altoona, Iowa has been one of Meta’s core data center campuses for years and is where the company has run its longest robotics trial, giving Watney’s two dual-arm units more than a year of live testing on cable work since the pilot began in June 2025 – a timeline independently tracked by Tech-Insider and echoed by GadgetReview’s reporting that the Altoona cable-swapping trials have been active since mid-2025. Prometheus, the newer campus in New Albany, Ohio, is tied directly to Meta’s AI expansion and was expected to come online during 2026, according to earlier TechRepublic coverage of the site. Running the ABB pilot at a brand-new campus rather than retrofitting an older one suggests Meta wants to learn whether robotics can be designed into a data center’s physical layout from day one, instead of being bolted onto existing infrastructure built for human technicians.
That distinction matters. Aisles, rack spacing, cable routing and lighting in most existing data centers were designed for people, not machines with wheels, lift mechanisms and limited battery life. If Prometheus becomes a testbed for robot-friendly facility design, it could shape how Meta and other hyperscalers lay out every AI data center built after it.
The 80% Number: How Big Could the Job Cuts Get
The figure driving most of the coverage is a single estimate from a Meta employee interviewed for the original Wired report and repeated by TechRepublic and Cryptopolitan: a fully working cable-swapping robot could eliminate up to 80% of the workload in certain technician roles. That number has not been independently verified and reflects one worker’s estimate rather than a company projection, but it illustrates the scale of what Meta is testing for. Cable work is one of the most time-consuming, physically repetitive parts of a data center technician’s job, so even a partial success at automating it would reshape how many people a facility needs on-site.
It is worth separating the estimate from the current reality. Every report on the pilots, including TechRepublic’s, stresses that the robots being tested today are slower than a trained human technician at every task evaluated so far. The 80% figure describes a hypothetical endpoint if the technology matures, not where Meta’s pilots stand in August 2026.
Why Robots Are Still Slower Than Human Technicians
Data center floors are cramped, cluttered and constantly changing, which turns out to be a harder environment for robots than the assembly lines many of these platforms were originally designed for. TechRepublic and TechTimes both point to the same set of obstacles: aisles blocked by carts or spare parts, limited battery life that forces robots to pause for recharging mid-shift, unreliable visual inspection when cabling is dense or poorly lit, and general navigation difficulty in spaces built for people rather than wheeled machines.
None of these are exotic engineering problems, but together they explain why Meta’s pilots remain human-supervised rather than autonomous, a detail WIRED reconfirmed in its August 2026 reporting on the Watney deployment. A robot that has to stop and recharge partway through a cable swap, or that cannot reliably tell a loose connector from a properly seated one, is not ready to work unattended in a facility where downtime costs real money. That reality sits awkwardly alongside Watney’s own September 2026 claims of better than 99.99% reliability and 24/7/365 uptime across its customer fleet, a gap between vendor marketing and the slower, supervised reality still documented on Meta’s data center floors. For now, the robots are colleagues, not replacements.
The Two-Stage Automation Squeeze on Data Center Labor
TechTimes framed Meta’s approach as a “two-stage automation squeeze” in its August 31, 2026 analysis of the pilots. The first stage is already deployed and largely uncontroversial: self-driving tugger robots that move heavy server racks, wheeled inventory robots that scan and log equipment, and even a simple mechanical “finger” device that presses a server’s physical power button on remote command. These tools already let fewer technicians manage more hardware without directly threatening skilled roles.
The second stage is what the Watney, Kinova and ABB pilots represent: automation aimed at the more skilled, judgment-heavy tasks that have so far resisted machines. Cable diagnosis, precise component reseating and power management all require a degree of troubleshooting that simple tugger or scanning robots do not. If Meta succeeds in pushing automation into this second stage, the job cuts would land on more experienced, better-paid technicians rather than the most repetitive entry-level tasks that were automated first.
How This Fits Into Meta’s AI Infrastructure Bet
Meta’s robot pilots cannot be separated from the sheer scale of money the company is putting into AI infrastructure. Meta has raised the low end of its 2026 capital expenditure guidance from $130 billion to $135 billion, for a range of $135 billion to $145 billion, according to reporting compiled by AOL and other outlets tracking hyperscaler earnings calls, up sharply from around $71 billion to $72 billion in 2025. Much of that spending is going directly into AI-focused campuses like Prometheus, and into the compute clusters described in Meta’s own recent AI infrastructure announcements.
When a company is spending well over $100 billion a year on data centers, even modest per-technician efficiency gains compound into real money once applied across dozens of campuses. That is the strategic logic behind testing robots for tasks that might look mundane in isolation: cable swaps and power cycles happen thousands of times a day across Meta’s global footprint, so shaving minutes or headcount off each one adds up fast at that scale.
Market Impact: The Hyperscaler Capex Race
Meta is not alone in this spending surge, and the scale of the broader hyperscaler buildout helps explain why automation of any kind, physical or software, has become a competitive necessity rather than a nice-to-have. Fortune estimated that Big Tech’s combined AI infrastructure spending could approach $630 billion in 2026, a figure the outlet noted rivals the entire GDP of Sweden, according to its February 2026 analysis. A separate Fortune report published in April 2026 pushed that estimate higher still, suggesting total 2026 AI infrastructure spending could surpass $700 billion.
Amazon has been the most explicit among the hyperscalers about where robotics fits into that spending. In its own capex guidance, reported by TechCrunch in February 2026, Amazon projected roughly $200 billion in 2026 capital expenditure explicitly covering “AI, chips, robotics, and low earth orbit satellites,” naming robotics as one of its named strategic spending categories rather than a side project. That echoes the direction Meta appears to be moving in with its own data center pilots, even though Amazon has not disclosed specific robot models or vendors comparable to Watney, Kinova and ABB in public reporting so far.
| Company | 2026 Capex (Projected) | 2025 Capex | Robotics Named as Spending Category |
|---|---|---|---|
| Amazon | ~$200 billion | ~$125-131 billion | Yes, explicitly |
| $175-185 billion | ~$91 billion | Not specified in reporting | |
| Meta | $135-145 billion | ~$71-72 billion | Implied via robot pilots |
| Microsoft | ~$110-120 billion | ~$90 billion | Not specified in reporting |
The pattern across all four companies is the same: capex has roughly doubled or more year over year, driven almost entirely by AI compute demand. Network World’s reporting on hyperscaler earnings calls, drawing on comments from Amazon executives in April 2026, described backlogs of AI infrastructure demand that continue to outstrip supply, which is part of why the pressure to automate every possible layer of data center operations, including physical maintenance, keeps intensifying.
Competitive Comparison: Meta vs Other Hyperscalers on Data Center Robotics
What sets Meta’s disclosure apart is its specificity. TechRepublic and TechTimes were able to name three vendors, describe the exact tasks under test and identify the two campuses involved. That level of detail is unusual. Amazon’s robotics ambitions, by contrast, are described mostly through budget line items rather than named hardware or pilot sites, at least in public reporting so far. Google and Microsoft have not disclosed comparable data center robotics programs in the sources reviewed for this story, even though both are part of the same $630 billion-plus capex wave.
That gap could reflect genuine differences in how far along each company is, or it could simply mean Meta’s program leaked into public view first through worker interviews while its rivals have kept similar efforts quieter. Given that AWS explicitly lists robotics as a capex category and that Amazon already operates one of the largest robotic fleets in the world inside its fulfillment warehouses, it would be surprising if the company were not running parallel experiments in its own AI data center buildout, even without public confirmation yet.
Historical Context: From Warehouse Robots to Server Room Robots
Robots doing repetitive physical labor inside logistics and industrial facilities is not new. Amazon has used robotic systems in its fulfillment centers for over a decade, and ABB itself has decades of history automating factory floors long before data centers became a priority. What is new is applying that same category of hardware to the server room, an environment historically treated as too delicate, too dense and too irregular for anything other than trained human hands.
The shift tracks a broader trend in AI hardware more generally, where specialized systems increasingly take over tasks once considered too complex to automate. Companies such as Cerebras have pushed specialized silicon into AI inference workloads, while chipmakers keep racing to squeeze more performance per watt out of every server rack Meta and its peers deploy. Robotics is simply the physical-world extension of that same pressure: build infrastructure that scales without a proportional increase in human headcount.
Worker Reaction and the Labor Market Fallout
Cryptopolitan’s coverage of the story leaned into the anxiety among Meta’s own staff. Current and former employees told the outlet they fear the robots will eventually reduce the number of technicians needed to keep AI data centers running, even though Meta has not announced any specific layoff numbers tied to the pilots. That uncertainty is itself part of the story: workers are watching a live experiment that could directly affect their job security, without a clear timeline for when, or whether, the robots will be judged good enough to scale up.
This kind of tension is not unique to Meta. As AI-related automation debates spread across the tech industry, from software engineering to physical infrastructure, they increasingly intersect with policy questions about how fast companies should be allowed to move and what obligations they have to affected workers. That broader debate is playing out in Washington too, where the patchwork of state and federal AI rules has struggled to keep pace with how quickly companies are deploying automation, a dynamic explored in recent coverage of the US AI policy landscape.
What This Means for Meta’s AI Infrastructure Race
Meta’s willingness to test robots for the unglamorous work of running data centers says something about how it views the AI buildout: as a logistics and operations challenge at least as much as a machine-learning one. The company has spent heavily on GPUs, custom silicon and compute clusters, but keeping thousands of racks running around the clock requires an army of physical maintenance work that scales with every new campus. Automating even a fraction of that work directly affects how fast Meta can expand capacity without proportionally expanding its on-site workforce.
It also puts pressure on Meta’s AI infrastructure rivals, several of whom have separately been buying up hardware at record pace, including reported Mac purchases tied to AI training workloads elsewhere in the industry. If robotics pilots prove that physical maintenance costs can be meaningfully reduced, expect Amazon, Google and Microsoft to either accelerate disclosure of their own programs or quietly copy the approach without publicizing it, given how sensitive the labor optics already are.
Predictions: Where Data Center Robotics Goes From Here
- More disclosures within 12 months. Given that Amazon already names robotics as a capex category, expect at least one other hyperscaler to confirm a comparable data center robotics pilot by mid-2027.
- Robots stay supervised through 2027. The battery life, navigation and visual inspection problems documented in current reporting are unlikely to be solved quickly enough for fully autonomous operation within the next 18 months.
- Cable-swapping robots reach cost parity first. Because cabling is the most repetitive, best-documented task in the pilot, it is the most likely candidate to reach practical, scaled deployment before power-cycling or component reseating do.
- Vendor funding and contracts grow. Watney Robotics, Kinova and ABB are likely to see increased hyperscaler interest and potential funding or contract expansion as more AI data center operators evaluate similar automation.
- Labor scrutiny intensifies. As worker concerns reported by Cryptopolitan spread beyond Meta, expect louder public and policy debate over automation’s effect on data center employment, echoing fights already underway over AI’s effect on other job categories.
The Bigger Picture: Automation as Infrastructure Strategy
Meta’s robot pilots are a small, narrow experiment on paper: three vendors, two campuses, a handful of tasks. But they sit inside a much larger reordering of how hyperscalers think about running AI infrastructure at scale. With combined industry capex approaching $630 billion to $700 billion in 2026 alone, according to Fortune’s tracking of hyperscaler spending, every layer of the data center stack, from chips to cooling to the technicians walking the aisles, is under pressure to become more efficient. Robotics is simply the most visible, most human-facing part of that push.
Whether the 80% workload reduction estimate ever becomes reality remains genuinely uncertain. The robots being tested today are slower than the people they might one day replace, and every report on the pilots stresses that human supervision remains essential. But the direction of travel is clear enough: Meta, and likely its rivals close behind it, are treating physical labor inside the data center as the next frontier for automation, not a permanent exception to it.
Frequently Asked Questions
What robots is Meta testing in its data centers?
Meta is piloting hardware from three vendors, a count WIRED confirmed in its August 2026 reporting: dual-armed robots from Watney Robotics, which raised an $80 million Series A round in September 2026 according to Tao Media, for cable swapping; Kinova Gen3 robotic arms for server power-cycling; and ABB mobile platforms with six-axis arms for reseating hardware components.
Where are Meta’s data center robot pilots taking place?
The pilots are running at Meta’s Altoona, Iowa campus, where two of Watney’s dual-armed cable-swapping robots have been performing cable insertion and removal since June 2025 – a detail corroborated by both WIRED and a Japanese-language report from Zaikei – and at the newer Prometheus campus in New Albany, Ohio, where ABB’s mobile platforms are being trialed. Watney itself says it has served major hyperscaler customers since 2025, a claim consistent with its role in the Altoona pilot.
Will robots replace Meta’s data center technicians?
Not immediately. Every report on the pilots describes the robots as currently slower than skilled human technicians and operating under human supervision. A Meta worker estimated that a mature cable-swapping robot could eventually cut up to 80% of the workload in some roles, but that figure is an estimate, not a confirmed target.
How much is Meta spending on AI data centers in 2026?
Meta raised the low end of its 2026 capital expenditure forecast from $130 billion to $135 billion on October 21, 2026, for a range of $135 billion to $145 billion, up from roughly $71 billion to $72 billion in 2025, largely driven by AI infrastructure and data center construction.
Are other tech companies testing data center robots too?
Amazon has named robotics as an explicit category within its roughly $200 billion 2026 capital expenditure plan, though it has not disclosed specific data center robot models or pilot sites comparable to Meta’s. Google and Microsoft have not publicly confirmed similar programs in current reporting.
What tasks are the data center robots performing?
The robots are being tested on network cable swapping, power-cycling servers by cutting power or pressing physical buttons, and reseating hardware components such as memory modules or network cards.
Why are the robots still slower than human technicians?
Reporting on the pilots points to several unresolved issues: cluttered aisles that are hard to navigate, limited battery life that requires recharging, and unreliable visual inspection when dealing with dense or poorly lit cabling.
What automation was already in place before these robot pilots?
Meta already uses self-driving tugger robots to move server racks, wheeled robots for inventory scanning, and a simple mechanical device that presses server power buttons on remote command. The Watney, Kinova and ABB pilots represent a more advanced second stage of automation targeting skilled maintenance tasks.
