GXO Targets Up to 1,000 Apptronik Apollo Humanoid Robots in Conditional Four-Year Plan
A letter of intent ties the Robotics-as-a-Service rollout to 2027 warehouse trials and jointly agreed KPIs

GXO Logistics said on November 12, 2026 that it has signed a letter of intent with Apptronik to expand warehouse trials of the Apollo humanoid robot and to develop a Robotics-as-a-Service framework for commercial deployment. If those trials succeed, the two companies intend to work toward a global rollout of up to 1,000 Apollo robots over the next four years, starting in North America in late 2027. The headline number arrives with a qualifier it does not advertise: three months earlier, GXO Chief Executive Patrick Kelleher told analysts on the company's second-quarter earnings call, "We have not achieved ROI on humanoids yet; I think we are a couple of years away from that."
That gap between stated ambition and operating reality defines the agreement. The letter of intent formalizes a relationship that began as a research and development program in 2024, puts a four-figure number on GXO's humanoid plans for the first time, and names a region and a date for the first commercial deployment. It is not a purchase order. Reaching 1,000 units depends on trials that have not started yet and on performance measures the two companies still have to hit, which makes the figure a ceiling rather than a delivery schedule.
What the Letter of Intent Commits GXO and Apptronik To
Under the agreement, GXO and Apptronik will expand joint trials of Apollo in live operations beginning in 2027. The trials cover goods-to-person and person-to-goods picking, case and tote movement, packing, and sortation. According to GXO, they are meant to validate the safety, security, and technical requirements for scaled deployment and to show how humanoids work alongside existing automation and warehouse staff. Each use case will be assessed and sequenced against jointly agreed productivity, safety, and operational KPIs.
Only if those deployments succeed and the KPIs are met do the companies intend to progress toward the 1,000-robot rollout. The commercial framework is expected to follow a Robotics-as-a-Service structure, in which GXO pays for robots as a service instead of buying the hardware. The announcement does not name the sites involved, the contract value, the per-robot price, or which generation of Apollo would be deployed.
The agreement builds on work already done. The partnership started with a proof-of-concept program announced in June 2024. Earlier in 2026, GXO said, Apollo completed a pilot deployment in the United States focused on autonomous picking of small consumer goods, and the two companies have run data collection and live autonomous evaluations on more than 50 logistics tasks at Apptronik's Robot Park training facility.
Kelleher described humanoids in the announcement as "the latest evolution of intelligent automation." Apptronik co-founder and Chief Executive Jeff Cardenas said, "GXO has the network and the operating discipline to take this technology to real scale."
Why GXO Is Setting a Target Before the ROI Is There
GXO describes itself as the world's largest pure-play contract logistics provider, with more than 150,000 team members across more than 1,000 facilities. On the August 5 earnings call, Kelleher said the company "will deploy 20,000 robots across our network this year." Almost none of that is humanoid. "In production, zero will be humanoids this year," he said when an analyst asked how many of those robots might be humanoids, adding that GXO had run 45 humanoid pilots and that production use was probably two years away.
The economics explain both the caution and the commitment. In a FreightWaves interview published August 12, Kelleher said humanoid operating costs currently run $10 to $15 per hour and should fall below $10 within two years, while the cost of a humanoid unit should be cut in half over the same period. He put total costs at $15 to $20 per hour within two years, a threshold he said is necessary to achieve a return on investment. For comparison, the Bureau of Labor Statistics reports average hourly earnings of $26.02 for production and nonsupervisory employees in U.S. warehousing and storage as of September 2026, a preliminary figure that excludes benefits and other employer costs.
Hourly cost is only half of the calculation. A robot that costs less per hour than a worker but completes tasks more slowly can still be the more expensive option per unit handled. Kelleher has since pointed to hand dexterity and task speed as the largest remaining gaps, according to trade coverage of his remarks at an investor conference in September.
The Robotics-as-a-Service structure is how GXO manages that uncertainty. Paying an operating fee per robot, instead of purchasing hardware outright, leaves capital and hardware risk with the vendor and lets the customer scale deployments up or down based on measured performance. For a contract logistics provider that runs warehouses on behalf of brand clients, a variable operating cost is also easier to build into a service agreement than depreciation on a fleet of machines whose useful life is still unknown. Fortune Business Insights projects the global robots-as-a-service market will grow from $3.31 billion in 2026 to $16.54 billion by 2034, an analyst estimate that covers all robot types, not humanoids alone.
Read more: AGIBOT puts six humanoid robots through a store stress test
How Apollo's Force-Control Design Differs From Conventional Warehouse Robots
When GXO first announced its work with Apptronik in 2024, it described Apollo as a robot that stands 5 feet 8 inches tall, can carry 55 pounds, and runs on swappable batteries. The company said Apollo is powered by linear actuators that mimic the mechanics of human muscles, and that its force control architecture and flexible safety zone perimeter allow it to work around and directly with people.
That design choice matters for warehouse work. A conventional industrial robot is typically position-controlled: it is commanded to move a joint to a target angle and hold it there. That approach works well on a manufacturing line where every part arrives in the same place, and less well in a warehouse where item weights vary, totes arrive slightly out of position, and a co-worker can step into the workspace. A force-controlled robot regulates the forces it applies instead of driving rigidly toward a programmed endpoint, so it can register resistance and yield. For a logistics operator handling mixed products for many clients at once, that compliance is the difference between a machine that tolerates variance and one that damages goods or faults when something is not where it expected.
The current version of the platform is Apollo 2, which Apptronik unveiled on June 30, 2026 in bipedal and wheeled-base configurations. Apptronik said the wheeled base is designed to meet existing safety standards for industrial mobile robots, while the bipedal configuration is intended for maximum adaptability in complex environments.
Apptronik does not present Apollo 2 as its commercial product. The company calls it the workhorse behind its data collection effort and says what it learns on Apollo 2 is powering the development of Apollo 3, its next-generation commercial robot. GXO's announcement does not say which generation would go into its warehouses in late 2027, and that open question bears directly on what the first commercial units will be able to do.
The Google DeepMind Data Loop Behind Apollo
Apptronik's most distinctive asset may be the data infrastructure that runs alongside the hardware. The company entered a strategic partnership with the Google DeepMind robotics team in December 2024. Its newly expanded Robot Park in Austin, Texas, covers nearly 90,000 square feet and anchors what Apptronik describes as a growing network of Robot Park sites at customer and partner locations, including Mercedes-Benz and GXO.
Robot Park works as a data factory. Fleets of Apollo 2 robots perform logistics, manufacturing, and retail tasks through a mix of teleoperation and autonomous execution, and the resulting data is used to train and refine Gemini Robotics, Google DeepMind's foundation models for robots. Better models are then meant to improve the robots, which generates more useful data. Every Apollo unit working in a GXO facility would feed that loop.
The most recent public measure of where the loop stands came on July 30, 2026, when Google DeepMind introduced Gemini Robotics 2 and published task-level success rates. According to a summary of the company-reported results, a single model checkpoint controlled three configurations: Apollo 2 with SharpaWave hands, Apollo 2 with Inspire hands, and a Franka Duo dual-arm system with a Robotiq gripper. Whole-body picking on Apollo 2 with Inspire hands succeeded 76.3 percent of the time from a shelf, 68.4 percent from a table, and 45.7 percent from the floor. Multi-finger tasks on the SharpaWave hands ranged from 92 percent for unscrewing a light bulb down to 32 percent for a dustpan task, with screwing a bulb in, sealing a zip-top bag, and tying a trash bag all below 50 percent.
Google DeepMind itself said multi-finger dexterous manipulation remains challenging and acknowledged that movement speed needs to improve. The figures are company-reported, and no independent test results have been published. They are also directly relevant to GXO: picking and packing, two of the use cases named in the letter of intent, depend on exactly the kind of dexterity where the reported results are weakest.
Gemini Robotics is not Apollo's only route to learned behavior. Apptronik announced in March 2024 that Apollo would integrate with Project GR00T, NVIDIA's foundation model for humanoid robot learning.
GXO Is Running More Than One Humanoid Platform
The Apptronik announcement is not GXO's only humanoid commitment, or its most advanced one. Agility Robotics' Digit has been working at GXO's Flowery Branch, Georgia, facility under a Robots-as-a-Service agreement signed in June 2024, and has moved more than 100,000 totes there. On October 26, 2026, the two companies announced an expansion to GXO sites in Statesville, North Carolina, and Lebanon, Tennessee, with deployments expected to begin in early 2027 on Digit 4 and to migrate to Digit 5 when it becomes commercially available.
That expansion came 17 days before the Apollo letter of intent, and the two are not in competition for the same slot. Digit has a track record at GXO in a narrow, repetitive task: moving totes. The Apollo trials are aimed at a broader and harder set of workflows, including picking individual items and packing. GXO has also signed a Robots-as-a-Service agreement with a third humanoid developer, Reflex Robotics, Forbes reported.
Both of GXO's main humanoid suppliers are well capitalized. Agility completed its merger with special-purpose acquisition company Churchill Capital Corp XI on November 4, and said its shares were expected to begin trading on Nasdaq under the ticker AGLT the following day. Apptronik closed a $520 million Series A extension in February 2026, bringing its Series A total to more than $935 million, with existing investors B Capital, Google, Mercedes-Benz, and PEAK6 joined by AT&T Ventures, John Deere, and the Qatar Investment Authority. Apptronik did not disclose a valuation; TechCrunch reported a post-money figure of about $5.3 billion.
Taken together, the pattern looks like a portfolio approach. GXO is paying for humanoids as a service from several vendors, matching each platform to the tasks it can already perform, and avoiding an exclusive bet while its own chief executive says the return on investment is still unproven.
Where Humanoid Logistics Deployments Stand in Late 2026
GXO's target lands in a market where announced ambitions still far exceed verified production deployments, and where most commercial agreements involve a handful of machines.
Toyota Motor Manufacturing Canada signed a Robots-as-a-Service agreement with Agility in February 2026 covering seven Digit robots at its Woodstock, Ontario, plant, where they load and unload totes from an automated tugger. Figure AI announced a commercial agreement with Catalyst Brands, the operator of JCPenney, Aéropostale, and Brooks Brothers, in May 2026, starting at Catalyst's distribution center in Reno, Nevada. In Germany, drugstore chain Rossmann is running a year-long pilot of a single UBTECH Walker S2 at its logistics center in Burgwedel.
Boston Dynamics is on a different path. Production versions of its electric Atlas are being tested at Hyundai facilities, and the wheeled Atlas the company showed in November is a research testbed for stress-testing motors and control software, not a product.
Read more: Why Boston Dynamics put its Atlas humanoid on wheels
Analysts disagree on how quickly this changes. Gartner predicted in January 2026 that through 2028, fewer than 100 companies will move humanoid proofs of concept beyond experimentation and fewer than 20 will run humanoids in live production for supply chain and manufacturing. BofA Global Research is more bullish on volume, forecasting 90,000 humanoid shipments in 2026 and 1.2 million in 2030. Against either view, 1,000 robots over four years is a meaningful commitment for a single logistics operator and a small fraction of a network with more than 1,000 facilities.
What the ROI Gap Means for the 2027 Timeline
The conditions in the letter of intent tie the scale number to execution that has not begun. Expanded trials start in 2027, and the first commercial deployment is targeted for late that year, roughly 12 months from now. In that window, Apollo has to show it can perform several distinct workflows reliably enough, and quickly enough, to clear KPIs that GXO and Apptronik have agreed on but not published.
The dexterity results from Gemini Robotics 2 suggest how much ground there is to cover. Tote and case movement sit closer to what humanoids already do in production settings. Picking and packing varied consumer goods ask for the fine manipulation that the published results show is still inconsistent, and for the task speed that both Kelleher and Google DeepMind have identified as a limitation.
Robot performance is not the only variable. GXO says the trials must also validate safety, security, and technical requirements and demonstrate how humanoids work alongside existing automation and employees. In practice that means integrating a new class of machine with warehouse management software built to direct conveyors, mobile robots, and people; establishing safety procedures for shared work zones; and building the maintenance and support capacity to keep a fleet running across multiple sites. Each of those adds time and cost between a signed letter of intent and 1,000 robots in production.
For the wider humanoid sector, the agreement sets a concrete reference point. A large logistics operator has attached a number, a region, and a date to its humanoid plans before the return on investment is proven, and has chosen a subscription model designed to contain the risk of being wrong. Whether late 2027 produces a working commercial deployment, and whether the quarters after it show the robot count rising toward that ceiling, will say more about the state of humanoid logistics than the announcement itself.



