GXO's plan for up to 1,000 Apptronik Apollo humanoids over four years: what a robots-as-a-service letter of intent actually commits to, and what it does not
The largest pure-play contract logistics company has signed the biggest humanoid framework yet, but it is stepped, KPI-gated and paid per robot-hour; three months ago its CEO said humanoids had not yet returned their cost, which makes this a test the whole industry will be graded on.

GXO Logistics and Apptronik announced on November 12 a letter of intent under which Austin-based Apptronik will supply up to 1,000 Apollo humanoid robots to GXO facilities worldwide over four years. Expanded trials begin in 2027 and the first commercial deployments are planned for North America in late 2027. The commercial model is robots-as-a-service: GXO pays for robot capacity rather than buying a fleet, and the rollout proceeds in steps, with each use case validated against jointly agreed performance, safety and operational KPIs before it scales. The two companies have worked together since 2024 and have run data collection and live autonomous evaluations on more than 50 logistics tasks at Apptronik's 90,000 square foot Robot Park, after an earlier US pilot in which Apollo picked small consumer goods autonomously.
What the number means
One thousand is a ceiling, not an order. GXO runs more than 1,000 facilities and 200 million square feet, employs over 150,000 people and is putting roughly 20,000 conventional robots to work across its network this year, so even a full rollout would be one humanoid per site against twenty wheeled or arm robots. The company has run more than 45 humanoid pilots and was Agility Robotics' first commercial customer in June 2024, has since signed a second humanoid RaaS deal with Reflex Robotics, and is expanding Agility's Digit to two more warehouses in Statesville, North Carolina and Lebanon, Tennessee from early 2027. Apptronik's Apollo is being pointed at a wider task set than Digit's tote handling: goods-to-person and person-to-goods picking, case and tote movement, packing and sortation. CEO Patrick Kelleher said on the August earnings call that GXO had not achieved ROI on humanoids yet and was a couple of years away, naming hand dexterity and speed as the hurdles. Nothing in the November announcement moves that date; it formalises the path to it.
Why the structure matters more than the headline
Humanoid announcements have mostly been counted in units promised. This one is instructive because of what GXO did not commit to. There is no purchase, so the capital risk of a platform that is still improving sits with Apptronik, which raised $935 million this year and can carry it. There is no fixed delivery schedule, only a stepped expansion gated on KPIs, which lets GXO stop at any use case that fails. And the payment is for service, which means Apptronik is paid for robot-hours that do useful work, the metric investors have struggled to get from any humanoid vendor. J.P. Morgan's forecast last week of 300,000 to 400,000 US humanoids a year by 2030 named hands, actuators, the brain and manufacturing capacity as the bottlenecks; a 1,000-unit RaaS framework with a demanding operator is the kind of contract that produces the data to tell whether those bottlenecks are clearing.
The counter-argument
A letter of intent with a four-year horizon and a late-2027 first commercial deployment can be read as a press release about 2028. GXO gets to say it is the humanoid leader in logistics; Apptronik gets a marquee name ahead of its next raise; neither has disclosed pricing, uptime commitments or the KPIs. Sceptics will note that GXO's conventional robot count grows by tens of thousands a year while its humanoid count is still measured in pilots, and that OpenAI folded its standalone robotics team last week on the view that general models will subsume much of the robot-specific stack. The reply is that GXO is the operator with the most humanoid pilot data in the world, and it chose to go bigger, not smaller.
What to watch
Whether GXO discloses per-task throughput or cost-per-pick when expanded trials start in 2027, which would be the first operator-reported productivity figure for a humanoid at scale. Whether Agility, Reflex and Apptronik end up competing for the same GXO sites, which would turn RaaS into a price-and-performance contest. And the February earnings call, where Kelleher will be asked again whether the ROI timeline has moved.
The story so far
- Nov 15, 2026 Mirendil's $1 billion Series A at a reported $5 billion, eleven months after founding: what investors are buying when th…
- Nov 14, 2026 OpenAI disbands its standalone robotics team for the second time, betting GPT-6 Astra's 74.8% simulation score makes a s…
- Nov 14, 2026 ChatGPT Wallet: network tokens, one biometric approval per purchase and no OpenAI fee at launch. What merchants and secu…
- Nov 13, 2026 J.P. Morgan: US humanoid production could reach 300,000–400,000 units by 2030 and 1.5 million by 2035, with hands and ac…
Key terms in this story
- Robots-as-a-service (RaaS) A commercial model in which a customer pays for robot capacity, usually per robot per month or per hour of work, instead of buying the hardware.
- Evaluation (eval) A test set with known-good answers used to measure a model or system on your own tasks.
More in Robotics
All →
Genesis AI raises $500 million at $3.5 billion for a robot trained on glove data from 100 workers, and GlobalFoundries becomes Atlas's first non-automotive tester. Two routes into the factory, priced very differently
One company is being valued on how it collects hand data before it has customers; the other is earning a humanoid pilot after 90,000 inspections by a …

Open-weight robot brains on both sides of the Pacific: Huawei's CloudRobo-WM1.2 on Ascend, Runway's Praxis-1 at 46.2% on RoboLab-120, and Agile Robots' $8.5 billion without a shipment figure
In one week the two most capable open world-action models for robots arrived from a video-generation company in New York and a chip company in Shenzhe…

OpenAI disbands its standalone robotics team for the second time, betting GPT-6 Astra's 74.8% simulation score makes a separate robot brain unnecessary
The researchers move under Mark Chen, the hardware staff under Greg Brockman, and the empty booth at CoRL tells robot makers what OpenAI thinks the mo…
Get the brief in your inbox
One email every morning: the stories that mattered in AI, each with a one-line "why it matters" for your role. Free, no ads, unsubscribe any time.
Published daily by AI Briefing. We never share your address.