Meta tests robots for power cycling and cable swaps inside its data centers

Start with the cheapest thing in Wired's account. At some Meta facilities there is a robot shaped like a finger, or a pointy stick, whose entire job is to press the power button on a Mac Mini or similar box when a human somewhere else asks it to ^[8]. The work is unglamorous: it removes a person walking to a rack to push a button a remote console could not reach.

Proponents pitch robots as relief, taking the repetitive and dangerous tasks off technicians so they can handle harder problems ^[20]. The people doing the work describe it differently. A Meta data center worker told Wired that the cable-swapping bot under test, if it works, could replace up to 80 percent of some people's workloads ^[7][16], and that "we thought those of us performing the physical tasks were safe for a while, but not anymore" ^[17]. The worker's own number does the math: if 80 percent of a repeatable workload goes away, the leftover fifth means the residual hands-on work of five such roles fits inside one ^[21]. That ratio describes a staffing model, not a task-relief program.

The reason this stopped being a demo is unglamorous. Robots were expensive and poorly suited to delicate rack work, and in earlier industry trials they sometimes crushed servers while attempting simple tasks, according to a person briefed on the experimentation ^[13]. Hardware got cheaper and the models driving the arms got more capable ^[14], and several startups are now building machines that assemble servers and replace wires ^[15].

The buyer is easy to name. It is whoever owns the capacity plan for a campus in a thinly populated county where the racks are plentiful and the qualified technicians are not ^[18]. It is also whoever would rather stop conditioning halls for human comfort: Paul Golding, who oversees physical intelligence at Analog Devices, says more customers tell him they want to run data centers with humanoids because the rooms can then run hotter and unlit ^[19].

Brennan's statement and Xu's roadmap can both hold if you separate construction from operations. Meta can hire hard for the people who build and commission capacity ^[2] while the steady-state task list on a live floor gets shorter ^[12]. In Altoona, the workers have already drawn their own conclusion, with group chats saying they will all be gone in a few years anyway because of the robots ^[10]. The activists fighting data center construction have largely left this alone, and human jobs remain the standard justification offered for the tax breaks ^[11].

For your own floor, sort the remote-hands list on two axes: how deterministic the motion is, and what a mistake costs in hardware. High determinism and low damage is where the bids land first, which is why power cycling and cable swaps are the pilots rather than rack builds ^[6][7]. The forcing function is a date: anyone who cited operations headcount in an incentive filing should be able to name the year that number stops holding, and if nobody in the room can name it, the number was a marketing input rather than a commitment.