Amazon Now Runs More Than 1 Million Robots: From Kiva to Vulcan, How AI Foundation Models Are Reshaping Warehouses and Human Work

In October 2026, Ars Technica published a long look at Amazon's robots. The opening fact is blunt: no organization in the world operates more robots than Amazon. It already has more than 1 million robots working in warehouses around the clock, helping make tens of millions of same-day and one-day deliveries happen daily, and it plans to open new factories to build more capable ones. At the same time, Amazon is the second-largest corporate employer in the world, with more than 1.5 million full-time and part-time workers according to its 2025 annual report. The article asks how those two numbers coexist.

Start with the fleet. Most of it is still wheeled mobile robots. Since acquiring Kiva Systems in 2012, the company, renamed Amazon Robotics, mostly builds its robots in-house in Massachusetts, and this year it announced factories in Austin, Texas, and Greenwood, Indiana. The pool-table-sized Hercules and piano-sized Titan carry tall fabric inventory pods to workstations. The newest, Proteus, has sensors that let it move among people more freely. Swarms in narrow lanes can create "robotic traffic jams," so Amazon coordinates them with a suite of AI foundation models called DeepFleet that predicts congestion. The company claims its Sequoia system, which coordinates swarms, can identify and store items up to 75 percent faster.

The robotic arms show AI at work best. Robin began testing in 2021 and debuted in Lakeland, Florida, in 2022, using eight suction cups to sort cluttered piles of packages. In a 2023 paper, Amazon found that replacing hand-crafted strategies with machine learning raised pick success to 93.73 percent from 92.28 percent for the best traditional method. That sounds small, but it meant about 16,500 fewer failed picks across more than a million picks. Cardinal places packages up to 50 pounds into carts like a game of Tetris. Sparrow faces hundreds of millions of products of every shape and, says Dresser, was "the first place where we truly deployed foundation models," avoiding hand-coding every item. The newest arm, Vulcan, is the first to combine vision with force feedback for a sense of touch, picking and stowing inside fabric pods. It began operating in Spokane and Hamburg in 2025, and Amazon expects to scale it in 2027. Robotic arms still number in the "single-digit thousands," per Scott Dresser, Amazon Robotics vice president, and CFO Brian Olsavsky said on a July earnings call that they would more than double this year.

The effect on people is where evidence diverges. Amazon says robots take over the toughest tasks, and since 2019 its global recordable incident rate has improved 43 percent and its lost-time incident rate 70 percent. But the Strategic Organizing Center, a coalition of unions, published reports in 2021 and 2022 showing higher injury rates at robotic warehouses than at non-robotic ones, and a 2024 US Senate HELP committee investigation linked robotic productivity to more pressure and injuries. Amazon replies that those reports rely on older data from before its latest robots. The most nuanced study, published in ILR Review in April 2025, found 40 percent fewer severe injuries at robotic warehouses than at legacy ones, but 77 percent more less-severe injuries tied to repetitive motion. Co-author Brian Greenwood of George Mason University said one mechanism is the pace of work: pick rates are "off the charts higher," with the robot dictating the pace.

On jobs, Amazon says robots create new roles. It says fulfillment centers with advanced robotics typically need 30 percent more people in reliability, maintenance and engineering, new robot factories should each bring more than 300 jobs, and robotic maintenance technicians are up over 30 percent, in a "much higher-skilled, higher-compensated job." It also says its footprint has grown, creating hundreds of thousands of jobs. Yet in October 2025 The New York Times reported internal strategy documents suggesting the robotics push could let Amazon avoid hiring more than 160,000 people in the US by 2027 and 600,000 by 2033; Amazon replied then that the documents reflected the views of one group.

Our reading is that the article's value is that it gives no single verdict. It lays out the evidence: robots do remove the most dangerous tasks and do speed up people. The company's injury numbers and the research point in different directions because the scope and the data years differ, not necessarily because anyone is lying. Keep in mind that many figures come from Amazon's own statements or from older 2016 to 2019 data. Watch three things: whether Vulcan scales in 2027, whether injury data keeps improving after robotic arms more than double, and whether the first Amazon Labor Union's Safety Bill of Rights, whose image reads "We are not robots," enters bargaining.