Meta Data Center Robots: Watney, Kinova, ABB Pilots Advance AI Ops

Meta Expands Robot Testing for Data Center Maintenance

Meta began detailed pilots of external robots in its data centers in mid-2025, focusing on Altoona, Iowa and the Prometheus campus in New Albany, Ohio. Workers report dual-armed systems from Watney Robotics handling network cable work since June 2025 at one Altoona building. A Kinova Gen3 arm evaluates power cycling tasks, while simpler remote-activated mechanisms press power buttons on devices like Mac Minis. These efforts address the physical demands of supporting AI training clusters amid rapid infrastructure growth.

The Altoona site, Meta’s largest campus opened in 2014, serves as a testbed where successful methods transfer to other locations. Prometheus in New Albany represents newer AI-optimized construction scaling toward gigawatt capacity. Four-wheeled ABB platforms equipped with scissor-lift risers and six-axis arms perform hardware reseating there. Meta declined specific comment on the pilots but issued a statement emphasizing ongoing hiring needs due to skilled labor shortages across U.S. infrastructure projects.

Vendor-Specific Deployments and Task Breakdown

Watney Robotics supplies dual-armed mobile units for cabling operations under direct human supervision. The San Francisco startup shifted focus toward data center applications after earlier demos in other sectors. Kinova’s Gen3 arm targets precise actions such as disconnecting server power, drawing on the Quebec firm’s established industrial arm designs. ABB contributes larger mobile bases suited for lifting and repositioning components within server racks at the Ohio site.

These systems perform repetitive physical tasks including network cable swaps, equipment inspection, and server resets. One data center employee estimated that mature cable-swapping robots could manage up to 80 percent of workloads in targeted roles. However, current implementations require constant oversight because speeds lag behind trained technicians and battery recharges interrupt continuous operation.

Meta already operates internal wheeled inventory scanners and tugger robots for rack transport across multiple sites including Iowa and Virginia. The new external pilots extend automation into finer manipulation previously reserved for humans. Internal notes indicate difficulty with dense cabling configurations tied to NVIDIA GB300 systems, where legacy equipment layouts favor human dexterity.

Labor Dynamics and Economic Pressures

Data center technicians express concern that successful automation could reduce demand for experienced physical roles. Chat groups at Altoona have discussed potential workforce reductions within a few years. Workers also note shifts toward lower-skilled “smart hands” positions guided by AI instructions rather than independent troubleshooting. Meta counters with America’s Workforce Academy programs offering free training and job guarantees in states with data centers, including Ohio.

Capital expenditures on AI infrastructure remain elevated, with Meta forecasting continued high spending. Automation aims to control labor costs in regions facing technician shortages. Yet Meta’s public position stresses the need for more workers, not fewer, during the largest U.S. infrastructure expansion since World War II. Local officials in Altoona view the facility’s tax arrangements as beneficial despite automation trends.

Technical and Operational Limitations

Robots encounter challenges with visual differentiation of indicator lights, navigation around floor cables, and sustained runtime between charges. Earlier industry attempts sometimes damaged equipment during simple maneuvers, though hardware costs have declined and AI perception models have improved. Dense cable environments and non-standardized server layouts slow progress toward unsupervised operation.

The pilots remain in early stages with no public timeline for broad rollout. Meta has acknowledged internally that certain high-density cabling tasks exceed current robot capabilities without infrastructure redesign. Battery constraints and the requirement for human door operation or remote guidance limit scalability across multi-building campuses.

Industry Context and Future Outlook

Other hyperscalers pursue parallel efforts. Microsoft explores robots for rack replacement, while Google and Amazon have demonstrated systems for inventory and parts recycling. Startups target data center-specific manipulation, yet experts note many pilots lack proven production deployment. Long-term visions include higher-temperature or dark-environment operation enabled by robots, though immediate gains focus on consistency for repetitive actions.

Meta’s approach relies on off-the-shelf industrial platforms rather than in-house humanoid development. The distinction highlights near-term use of proven automation versus longer-horizon generalist systems. Successful cable and reseating robots could influence staffing models across the sector as AI cluster sizes grow.

Implications for Fleet Economics and Production Ramps

Data center operators track mean time between failures and technician utilization rates closely. Automation that reliably handles 60-80 percent of cable and reset tasks would shift economics toward higher uptime and lower variable labor costs per rack. Current supervised operation adds oversight overhead that partially offsets gains until reliability improves.

Meta’s multi-vendor strategy allows rapid iteration across sites. Altoona results feed into Prometheus and future campuses. If ABB and Watney platforms demonstrate consistent performance on GB200/GB300-era hardware, other operators may accelerate similar procurements. Persistent gaps in dense-cabling dexterity and endurance suggest phased adoption rather than wholesale replacement of human teams in the next 18-24 months.