Amazon gave its most detailed on-record account of its robotics program on 12 October 2026, in an Ars Technica report built around interviews with Amazon Robotics vice president Scott Dresser and senior applied science manager Kapil Katyal. The news is not the size of the fleet. Amazon now manufactures robot arms itself, in the United States, and is willing to put numbers on the rollout.
What Amazon put on the record in October 2026
Robotic arms remain a small part of the fleet: Dresser told Ars they number in the single-digit thousands, and chief financial officer Brian Olsavsky said on a July 2026 earnings call that Amazon expects to more than double that count this year. Vulcan, the company’s first arm with force feedback, began operating in 2025 in Spokane, Washington and Hamburg, Germany, and Amazon says it will scale up its use in 2027. Cardinal, another arm, has begun wider deployment that Dresser said will continue over several years.
The production is moving in-house. Amazon announced a multi-billion-dollar manufacturing facility in Austin, Texas on 19 August 2026 and a plant of more than $100 million in Greenwood, Indiana on 24 September 2026. Amazon says Austin will create 300 to 500 jobs and Greenwood 300 jobs averaging nearly $100,000, with Indiana production expected to start by 2028. Those counts come from Amazon’s own announcements; no independent count was found.
Amazon also says its next-generation fulfillment center in Shreveport, Louisiana, launched in late 2024, requires 30% more employees in reliability, maintenance and engineering roles. Dresser gave Ars a separate figure, a rise of over 30% in robotic maintenance technicians across the network. The two measure different things, and neither has independent verification.
The million-robot figure is from June 2025
The number circulating again this month is older than it looks. Amazon announced its 1,000,000th robot on 30 June 2025, in a post written by Dresser that also introduced DeepFleet, a generative model Amazon claims improves fleet travel time by 10%. Before that, the company’s own figure had stood at “more than 750,000 robots”.
Scale is not in dispute. Amazon runs more than 1 million robots across more than 300 facilities, alongside more than 1.5 million full-time and part-time human employees, the headcount Ars Technica cites from its 2025 annual report. The arms are the part that is changing.
What the peer-reviewed study found
The only independent academic evidence in this story is a peer-reviewed paper published in ILR Review on 22 April 2025: “Lucy and the Chocolate Factory: Warehouse Robotics and Worker Safety,” by Gordon Burtch, Brad N. Greenwood and K. Ravindran.
It analyzes weekly data from more than 150 Amazon warehouses between 2016 and 2019. Its abstract reports a 40% decrease in severe injuries associated with robotics and a 77% increase in non-severe injuries. The paper says the rise in non-severe injuries is “at least partially” attributable to the increased pace of work at robotic facilities. Co-author Brad N. Greenwood told Ars Technica that pick rates at robotic fulfillment centers are “off the charts higher”, and that he has not seen much evidence the dynamics are changing.
Amazon disputes the relevance of that data. The company describes reports of this kind as flawed and based on older data from before its most modern robots, including the arms, and points to network-wide improvement since 2019: a global recordable incident rate down 43% and a global lost time incident rate down 70% over six years. No Amazon statement addressing this paper by name was found.
The study’s limits sit beside its result. It is observational, covers a single company, uses internal data obtained by journalists, and stops in 2019. It supports an association rather than proof of cause, and the arm generation now being built in Texas and Indiana falls outside the window entirely.
Four injury measures that count different things
Sources use different definitions of injury severity, and percentages do not transfer between them.
The ILR Review paper separates injuries into severe and non-severe cases and finds fewer of the first and more of the second at robotic sites. Its full text is paywalled, so the thresholds behind those categories could not be checked for this article.
Reveal, the Center for Investigative Reporting, used a broader measure in its September 2020 investigation. It found that at the most common Amazon fulfillment centers, the rate of serious injuries from 2016 to 2019 was more than 50% higher at warehouses with robots than at ones without. Reveal defined serious injuries as work-related injuries requiring days away from work or a job restriction.
The Senate HELP committee’s December 2024 report is a third comparison, and it measures Amazon against the warehousing industry. The committee found Amazon warehouses recorded more than 30% more injuries than the industry average in 2023, and that over the previous seven years Amazon workers were nearly twice as likely to be injured as other warehouse workers. Those figures say nothing about automation level.
Amazon reports a fourth set of numbers. It publishes network-wide recordable incident rates (RIR) and lost time incident rates (LTIR), neither broken out by whether a site uses robots. RIR covers injuries needing treatment beyond first aid; LTIR covers cases involving days away from work, a narrower category than Reveal’s serious injuries.
The Reveal finding and the paper’s severity split draw on overlapping 2016-2019 data and measure different quantities. Presenting either one as a refutation of the other would misdescribe both. The paper’s severity definitions would settle part of the gap, and they are not readable.
The projections Amazon calls one group’s view
Amazon’s plans for human headcount rest on internal documents rather than audited figures. The New York Times reported on 21 October 2025 that the company’s robotics team expected automation to flatten its hiring curve, allowing Amazon to avoid hiring more than 600,000 workers over a decade. Morgan Stanley worked a 160,000-plus reduction in projected US warehouse hires by 2027 into its own model, treating the figures as credible enough to rebuild its savings estimates on them.
Amazon said the documents were incomplete and reflected one group’s view. Tye Brady, the company’s chief technologist, called the premise of a shrinking workforce “speculation” at Amazon’s October 2025 Delivering the Future event.
These are projections of hiring Amazon expected to avoid rather than plans to cut existing jobs, and they are not evidence of 600,000 layoffs. Reuters reported a separate round of up to 30,000 Amazon layoffs in October 2025.
Why the post-2019 question has no published answer
A reader’s first question is whether Amazon’s robotic warehouses are safer than its older ones today. Nothing published answers it.
The peer-reviewed analysis stops in 2019. Amazon’s own injury reporting is network-wide, without a robotic and legacy split. Facility-level filings exist: the Strategic Organizing Center, a labor-union coalition, republishes OSHA Form 300 and 300A data for 2021 to 2025 covering more than 185,000 injuries since 2021, but those filings carry no automation field. Deciding whether a facility is robotic or legacy therefore depends on third-party classification, and SOC states that the datasets it uses are not official and have not been verified.
The automation that took Amazon’s fleet past a million robots is missing from the comparison. Closing the gap would take facility-level injury data for 2020 onward with an automation field attached, or a fresh peer-reviewed analysis covering the arms now being manufactured in Texas and Indiana. Amazon has published no such facility-level data, and no regulator or researcher has produced an analysis the public can check.