GXO Has Run 45 Humanoid Robot Pilots in Its Warehouses. None Have Paid for Themselves Yet

Patrick Kelleher, CEO of GXO Logistics, the world's largest contract logistics operator, told investors on an August 2026 earnings call exactly where his company stands on humanoid robots: “We have not achieved ROI on humanoids yet. I think we are a couple of years away from that.” Asked how many of the roughly 20,000 robots GXO plans to have running in its warehouses by the end of 2026 would be humanoids, his answer was blunter still: “In production, zero will be humanoids this year.”

That gap between what GXO is testing and what it is willing to deploy at scale is the clearest data point yet on where humanoid robotics actually stands in the one industry everyone assumed would adopt it first. Warehouses are structured, repetitive, and already full of robots — conveyor systems, autonomous mobile robots (AMRs), goods-to-person shuttles. If a general-purpose humanoid can't earn its keep there, the case for it anywhere else gets a lot harder.

45 Pilots, Zero in Production

GXO has run 45 separate humanoid pilots so far, a figure Kelleher gave investors directly, with a 46th cited by the company at a September industry conference. That's a serious testing program by any standard — not a single proof-of-concept demo, but dozens of real trials across different tasks and facilities. And the company still isn't deploying any of them into production this year. The two numbers aren't contradictory; they describe exactly what a rigorous pilot program looks like when the economics aren't there yet. GXO keeps testing because the long-term case might work. It keeps humanoids out of production because the short-term case doesn't.

Kelleher was specific about why: hand dexterity and task speed, in his words, are the hardest unsolved problems. Those are not minor engineering details. A warehouse picker needs to handle irregular bags, loose cables, shrink-wrapped pallets, and mixed-SKU totes at a pace that beats or matches a human worker doing the same job for $50,000 to $60,000 a year in loaded cost. Bipedal locomotion and natural language interaction — the parts that make humanoid robot demo videos go viral — are largely solved. Reliable, fast, general-purpose manipulation of unpredictable objects is not, and it's the only part that actually determines warehouse ROI.

The Forecast-vs-Operator Gap

This is where the humanoid robotics narrative and the humanoid robotics deployment data diverge sharply. Goldman Sachs has modeled the market shipping roughly 100,000 humanoid units as pilots scale into production over the next few years. Vendor and analyst material frames payback periods under two years, with one illustrative model pricing a unit at $20,000 plus $3,000 a year in service against a $50,000-a-year warehouse worker — a breakeven inside 18 months on paper. IDTechEx's 2026 analysis is more measured: it finds a payback pathway is emerging under favorable conditions, but explicitly warns that a shorter payback period on a spreadsheet is not the same thing as guaranteed profitability in a real facility with real variability.

GXO's actual experience sits closer to the IDTechEx caveat than the vendor pitch deck. A separate industry estimate puts total humanoid robots in commercial settings worldwide, as of January 2026, at roughly 3,000 units — about 60% still in pilots, 25% in limited production, and only 15% in anything resembling scaled deployment. For a technology that's been covered as imminent since 2023, that's a small, early-stage installed base, concentrated almost entirely in testing rather than earning a return.

Where the Real Automation ROI Already Exists

None of this means GXO has stopped automating. The company plans close to 20,000 robots in operation by the end of 2026 — just not humanoid ones. Those are the boring, already-proven categories: AMRs that move totes between zones, goods-to-person shuttle systems, automated sorting. Kelleher drew the distinction explicitly, calling conventional warehouse robotics technology that already generates ROI and can be deployed at scale today, in contrast to the humanoid pilots still being evaluated. That's the part of the story that gets less attention than humanoid robot demo reels, but it's where the actual productivity gains are showing up on GXO's books this year.

The Hedge: Agility Digit Arrives Anyway

Despite all of the above, GXO is still moving forward with a real humanoid deployment — just a deliberately staged one. The company plans to put Agility Robotics' Digit humanoid into production at facilities in Statesville, North Carolina, and Lebanon, Tennessee, starting in early 2027. The initial fleet will run the current Digit 4 hardware and migrate to Digit 5 once that model is commercially available. Reading the ROI comments and the deployment plan together, the picture is a company hedging two different bets at once: holding humanoids out of this year's production numbers while still buying an options position on the technology maturing by the time it actually needs more labor-saving capacity.

What This Means If You're Evaluating Humanoid Robots

For anyone outside GXO sizing up humanoid robots for their own operation, the lesson in these numbers is specific, not generic skepticism:

Separate the automation budget from the humanoid R&D budget. Treat a humanoid pilot as a research investment with a multi-year payoff horizon, not a line item competing against proven AMR ROI this fiscal year. Conflating the two will make any humanoid pilot look like a bad investment next to a goods-to-person shuttle, because right now, it is one.

Benchmark against your existing automation stack first, not against manual labor. The honest comparison for a humanoid robot isn't “a human worker” — it's “the AMR or shuttle system you could buy today for the same budget.” GXO's own fleet math shows which one wins in 2026.

Ask vendors for task-specific throughput data, not uptime stats. Dexterity and cycle time on your actual SKU mix are the bottleneck, per the operator with the largest testing program in the industry. A robot that stays upright and functional 99% of the time but handles irregular items slowly hasn't solved the problem that determines your ROI.

Expect the 2027–2028 window, not 2026, for production-scale deployment. Kelleher's own timeline — “a couple of years away” — and GXO's own staged 2027 rollout for Digit are the most concrete data points available from an operator running the largest known pilot program. Plan budgets and headcount decisions around that horizon rather than marketing timelines from robot manufacturers.