Amazon operates more than one million robots in its warehouses around the clock, as part of a network that relies on mobile robots, robotic arms, and artificial intelligence models to fulfill tens of millions of same-day or next-day delivery orders. The company plans to expand this fleet, while it still employs more than 1.5 million full-time and part-time workers, according to its 2025 annual report.
This expansion does not mean replacing humans entirely; rather, it changes the nature of work inside fulfillment centers. Robots handle moving shelves, sorting packages, picking products, and stowing them, while workers continue to perform tasks that automated systems are still unable to handle at scale.
From Kiva to a Fleet Exceeding One Million
The major transformation began after Amazon acquired Kiva Systems in 2012, one of the first companies to develop mobile robots that bring shelves to workers instead of requiring them to walk between aisles. The company was renamed Amazon Robotics, which develops most of the systems at facilities in Massachusetts.
Amazon also announced the creation of additional factories in Austin, Texas, and Greenwood, Indiana. Mobile robots make up the majority of the current fleet, including Hercules and Titan, while the Sequoia system can coordinate swarms of robots to identify and stow products at speeds the company says are up to 75 percent faster than previous methods. Proteus, meanwhile, uses sensors that allow it to move more freely near humans.
To address traffic bottlenecks among robots, the company uses a group of foundational artificial intelligence models called DeepFleet to predict congestion and coordinate movement.
Robotic Arms Are Becoming More Capable
Amazon expects the number of its robotic arms to exceed twice its current level during 2026, while the number publicly reported today remains in the low thousands. This fleet includes Robin, which sorts packages from mixed piles, and Cardinal, which places packages weighing up to 50 pounds into carts more densely.
Robin relied on a machine-learning model initially trained on more than 394,000 picking operations. In a trial exceeding one million operations, the best machine-learning method achieved a success rate of 93.73 percent, compared with 92.28 percent for a traditional algorithm, resulting in approximately 16,500 fewer failed operations. Another trial involving two million operations also showed a 20 percent reduction in the failure rate.
Sparrow, meanwhile, was designed to handle hundreds of millions of different products, using foundational models to recognize items and determine how to grasp them without requiring every product to be programmed manually. Vulcan adds force-feedback sensors, or something resembling a sense of touch, to help the arm pick products from tight spaces or stow them inside fabric shelving. Vulcan began operating at facilities in Spokane, Washington, and Hamburg, Germany, in 2025, and the company expects to expand its use in 2027.
Why Does Automation Not Necessarily Mean Easier Work?
Amazon says robots handle the most strenuous and dangerous tasks, such as reaching the upper rows of shelves or handling heavy loads. However, reports from labor organizations and an investigation by the U.S. Senate Committee on Health, Education, Labor, and Pensions linked automation to increased pressure on workers to maintain speed.
A study published in ILR Review in April 2025 presents a more complex picture: robotic warehouses recorded a 40 percent decrease in severe injuries compared with traditional facilities, but they recorded a 77 percent increase in less severe injuries associated with strain caused by repetitive movements. Brian Greenwood, a professor at George Mason University and one of the study's authors, points out that a robot may set the pace of work, pushing humans to perform repetitive movements more quickly.
Amazon responds that its global recordable incident rate has improved by 43 percent since 2019, and that the rate of injuries resulting in lost work time has improved by 70 percent. However, it argues that the critical reports rely in part on data predating the latest waves of automation.
What Is Changing in Practice?
Amazon's experience shows that technological value is not measured only by the number of robots, but by their ability to operate amid enormous product variety and deal with unstructured environments. A small improvement in the picking success rate also becomes highly significant when repeated millions of times.
At the same time, the figures show that transferring dangerous tasks to machines does not resolve the safety issue on its own. The potential benefit to workers may be accompanied by higher production rates and a faster pace, making the design of performance metrics, task allocation, and measurement of recent injuries open questions that an increase in the number of robots cannot answer by itself.
Amazon says that facilities with advanced robotics typically need 30 percent more workers in reliability, maintenance, and engineering roles, and that it has increased the number of its robot maintenance technicians by more than 30 percent. However, strategy documents cited by The New York Times in October 2025 indicated the possibility of reducing estimated U.S. hiring needs by more than 160,000 jobs by 2027 and 600,000 by 2033; Amazon said at the time that the documents reflected the view of one group within the company.