AutoStore's 2,050 systems: what the installed base proves and leaves out

AutoStore's headline number is large and consistently reported. Its third-quarter 2026 results presentation lists about 2,050 systems, about 96,500 robots and about 1,350 unique customers in 68 countries. That is strong evidence that cube storage is a mature, repeat-purchase product with buyers in almost every warehouse sector.

It is weaker evidence than it looks for anything else. A footnote in the same presentation says the system and robot counts "include installed base and backlog," so some of those systems are ordered but not yet shipped. AutoStore does not publish a definition of a "system," a split between running and on-order systems, or a size distribution. It sells through integrators, four of which produced 74% of its 2025 revenue. The most useful operating evidence comes from customers' own reports, and the clearest one available, from the Nordic retailer Boozt, describes peak capacity, custom software and extra costs that the headline count does not capture.

Analysis based on AutoStore's annual and quarterly reports, results presentations and web pages, integrator case studies and Boozt's annual reports, all retrieved November 9, 2026. AI assistance was used in research and drafting. Servospace did not visit any AutoStore site, interview AutoStore, its partners or customers, or audit operating data. AutoStore's next scheduled disclosure is its fourth-quarter 2026 report on February 11, 2027, according to its financial calendar, and figures here may be superseded by it.

In this analysis

• What AutoStore sells, and who sells it
• One number, three labels
• What the count does establish
• What a system is not: a unit of capacity
• Who actually confirms performance
• What a customer's own reports add
• Counterevidence and limits
• What a buyer should ask for

What AutoStore sells, and who sells it

AutoStore makes cube storage: stacked plastic bins inside an aluminum grid, with robots running on rails along the top that dig out bins and deliver them to workstations called ports, where people or robot arms pick items. It handles each and tote storage, the goods-to-person stage that sits inside the building after the trailer dock and alongside or instead of the case-handling systems covered in our Symbotic and Walmart analysis.

The company does not usually sell to the warehouse operator directly. According to its 2025 annual report, it sells modules to distribution partners, which design, install and service the systems for end users. The report names Swisslog, Bastian Solutions, Element Logic, Dematic, Fortna, Okamura and SoftBank Robotics among them. That structure shapes every number below:

• Revenue is recognized mostly at a point in time, on shipment. The annual report says most revenue is recognized and invoiced when modules ship to partners. Revenue therefore measures components delivered to integrators, not systems running at a customer.
• Revenue is concentrated in a few partners. Four distribution partners each contributed more than 10% of 2025 revenue and together accounted for 74%, up from 65% in 2024, according to the segment note. The note does not name them.
• Orders can be verbal. AutoStore defines order intake as the value of projects where a distribution partner "has received a purchase order or verbal confirmation that a specific installation will be ordered," according to its Q3 2026 report. Backlog is order intake not yet shipped. It stood at $581.6 million on September 30, 2026.

The annual report also flags the partner model itself as a risk: as AutoStore has grown, "there is a risk that this distribution partner model is exposed to a higher degree of stress," and the company monitors the risk of losing partners. The partner count fell from 23 in the first-quarter 2026 presentation to 21 in the third, which the later deck attributes to a "merger between three Partners."

One number, three labels

AutoStore's annual report uses three different descriptions for what is essentially the same year-end figure of about 1,900 systems.

The footnoted version is the most precise and the one to use. It means the figure cannot be read as "systems running." The share in backlog is not disclosed. A rough sense of scale is available: AutoStore reported about 1,950 systems at the end of the first quarter of 2026 and about 2,050 at the end of the third, while robots rose from about 89,000 to about 96,500 over the same six months, per the Q1 2026 presentation and the Q3 presentation. Growth on a count that includes backlog reflects signings as much as installations.

The count also does not move with revenue. Systems sold rose by 250 in 2025, the same increase as in 2024, while revenue fell to $538.6 million from $601.4 million, according to the annual report. In the first nine months of 2026, revenue rose 53% to $549.9 million. A system tally that grows steadily while revenue swings 10% in one direction and 50% in the other is measuring something other than delivered equipment value.

No definition of "system" or "unique customer" appears in the definitions or alternative performance measures sections of the 2025 annual report or in the Q3 report. It is not stated whether an expansion of an existing grid counts as a new system, whether one site with several grids counts once, or whether the small leased Pio units the company sells are included.

What the count does establish

Read with those caveats, the record still supports several conclusions a buyer can rely on.

Breadth. The Q3 2026 presentation splits the 2,050 systems by end market: about 670 industrial, 280 apparel and sports, 230 third-party logistics, 220 other retail, 200 healthcare, 170 automotive, 160 grocery, 70 consumer electronics and 50 luxury and personal care. Those figures sum to the headline. No single sector dominates, which suggests the product is not tied to one use case.

Repeat buying. AutoStore says about 55% of sales have gone to existing customers on average from 2021 through the third quarter of 2026. In 2025, it says it added more than 150 new customers and secured nearly 250 new sites, and it names DSV, Medline, Rohlik and Sonepar as multi-site accounts. Customers ordering again is a meaningful signal, though it is the company's own tally and "secured" means signed, not live.

Freedom to operate. A long patent dispute with Ocado, which runs its own grid-based grocery warehouses, ended in a July 2023 settlement. Under the terms both companies announced, all claims were withdrawn globally, each side licensed the other's pre-2020 patents, both can keep selling existing products, Ocado kept exclusive rights to its single-space robot, and AutoStore agreed to pay £200 million over two years. AutoStore's annual report says the liability matured on June 30, 2025, with no remaining obligations. For a buyer, the litigation risk attached to AutoStore's existing product line is resolved.

Expansion beyond totes. AutoStore agreed on November 5, 2026, to buy Logiqs, a Dutch maker of roaming pallet storage robots deployed at about 30 customer sites, for up to €30 million, as we reported. That is an addition to the platform, not part of the 2,050 count.

What a system is not: a unit of capacity

Systems vary enormously in size, and the count weights them all equally. The annual report says the average site has 45 robots, 40,000 bins and 10 ports. Dividing the Q3 2026 figures gives a similar average, about 47 robots per system, though both numbers include backlog.

Individual systems sit far from that average:

Boozt's single fulfillment center, built around three grids, holds roughly 1,200 robots, or about 1.2% of AutoStore's entire reported robot count. How many "systems" that site represents in AutoStore's tally is not disclosed.

AutoStore's published performance figures are also vendor specifications rather than measured fleet results. The annual report cites 99.7% "worldwide uptime," 99.9% picking accuracy and 650 "bins per hour x number of Ports." Its reliability page says the uptime figure comes from "statistics from hundreds of AutoStore systems," not from all of them, and does not define how downtime is measured. The 650-bins figure carries no stated conditions, such as bin mix, grid depth or the number of robots per port. These are useful as upper-range design references. They are not evidence of what a given installation sustains over a shift.

The same caution applies to the "1-3 years" customer payback that AutoStore repeats in its reports and presentations. The documents reviewed give no method, sample or denominator for it. Servospace does not convert any of these figures into a price, cost per system or return on investment.

Who actually confirms performance

The Q3 2026 presentation includes one customer case, which is a useful test of whose word supports a claim. The slide describes Parts Authority, a US automotive parts distributor, with two installations, 190 robots, 202,000 bins and 17 ports. Its quotation comes from a senior account executive at KPI Solutions, the integrator that built the system, not from Parts Authority. The slide lists "6x improvement in throughput, almost four times faster picking, well over 99% system uptime."

Two other documents describe the same customer with narrower claims:

The case page ties the sixfold figure to one shipping process, although its own headline also claims "4x greater throughput" without a qualifier. The presentation drops the shipping-process qualifier. The integrator credits the fourfold throughput gain to a hybrid system in which AutoStore is one of several technologies. In the documents reviewed, the only words attributed by name to a Parts Authority executive, its senior vice president of warehouse operations, are qualitative: "improved picking performance, increased throughput across the building, and a lower cost per line." The case page and slide also give different location counts for the company, more than 400 and over 250 respectively.

None of this suggests the system performs badly. It shows how a number changes as it moves from a customer's operation to an integrator's case study to an investor slide. A buyer should trace a figure back to the narrowest source and the named speaker before using it.

What a customer's own reports add

Customer-published documents are rarer and more informative. Boozt, a Nordic online retailer listed in Stockholm and Copenhagen, has described its AutoStore installation in its annual reports for years. Its disclosures give the clearest public view of what running cube storage at scale involves.

Scale and growth. Boozt's 2022 annual report says it added more than 500,000 bins and about 500 robots that year, reaching about 1.2 million bins and 1,200 robots. In 2019, it described the site as the largest AutoStore installation in the world, a claim it repeated in 2022.

Capacity, stated as a peak. Boozt describes handling capacity "exceeding 15,000 items per hour during peaks" in its 2022 report and "exceeding 20,000 items per hour during peak periods" in its 2024 and 2025 reports. That is a peak capacity, not a sustained average, and it covers the whole fulfillment center, including people at the ports. An AutoStore case page, relaying Boozt, cites up to 140 items per hour per operator for single-touch pick and pack and a best day of 190,000 items. Those are vendor-published figures without a stated period or average.

Custom software. Boozt's 2019 annual report says it implemented its own robot management system and "replaced the standard system, which was not built to handle the capacity we are at today," and that in 2020 it replaced the standard software in the robots with its own. At that scale and at that time, the standard control software was not what the customer chose to run. That is Boozt's 2019 and 2020 disclosure, not a verdict on AutoStore's current control software.

Multi-grid friction. Boozt's 2024 report says the site is organized around three AutoStore cubes and that more than 50% of 2024 orders contained products from more than one cube, which required manual consolidation. Boozt installed transfer cells to move bins between cubes by conveyor. Building and testing them required more manual hours during the second and third quarters before bringing "material productivity improvements" from the fourth. Its 2025 report says the full-year benefit "generated significant efficiency and cost savings." Expanding by adding grids can create work between them.

Costs outside the grid. Boozt's 2025 report says its 2026 capital spending includes about SEK 40 million for a "sprinkler roof on Autostores required by the insurance partner." Fire protection and insurer requirements are a site-level cost that sits outside AutoStore's module price.

Timing shifts. The 2024 report described an investment of about SEK 500 million in capacity expansion over 2025 to 2027. The 2025 report refers to "a planned expansion of the Company's fulfilment capacity from 2027." The documents do not say whether that expansion will use AutoStore.

Boozt is one customer, at an unusually large and customized site, so it is not representative of the average 45-robot system. It is valuable because the claims come from the operator, in documents with legal reporting obligations, and include the friction alongside the benefits.

Counterevidence and limits

Several items weaken or bound the installed-base story without contradicting it.

• No running count. AutoStore does not disclose how many of its systems are live, how many are in backlog or how many have been decommissioned or replaced. A company-wide count that includes backlog cannot show churn.
• Inconsistent labels. The same annual report calls the figure systems "sold," systems "running" and installed base "and backlog." The boilerplate in results releases drops the caveat entirely.
• Partner dependence. With four partners producing 74% of 2025 revenue, AutoStore's view of end customers runs largely through intermediaries. Its "unique customers" figure depends on what partners report.
• Uptime sample. The 99.7% uptime figure comes from "hundreds" of systems, a subset of more than 2,000, with no stated method.
• Narrow customer evidence. The customer-published evidence found for this analysis is concentrated in one company. AutoStore's annual report lists well-known names such as PUMA, Siemens, UPS, DHL, CVS Pharmacy and Best Buy as customers. This analysis did not find site-level operating data published by those companies, and their presence on a vendor list does not establish the size or status of any system.
• Not a fire record. Searches for this analysis found grid fires at Ocado's own warehouses in 2019 and 2021. Ocado uses its own technology, so those incidents are not evidence about AutoStore systems and are not counted here.

What a buyer should ask for

The installed base answers "is this a proven product category with many buyers?" with a clear yes. It does not answer "will this system do my work at my volumes?" These are the questions that the public record leaves open, and the evidence that would close them:

AutoStore's next scheduled checkpoint is its fourth-quarter 2026 report on February 11, 2027, which should update the system, robot and customer counts. The partner concentration figure comes from the annual report; the 2025 edition was published April 23, 2026. This guide will be reviewed after that report, after Boozt's 2026 annual report, or when another customer publishes site-level operating data. Otherwise, it will be rechecked within 90 days. For related coverage, see warehouse robot news.

How the sources were used

AutoStore's annual and quarterly reports establish what the company disclosed as a listed issuer, including revenue recognition, partner concentration and definitions. Its presentations and web pages are treated as company statements. Integrator case studies are attributed to the integrator. Boozt's annual reports establish what that customer said about its own operation. Ratios such as robots per system are simple divisions of company-reported figures that include backlog and are labeled as approximate. Nothing here is converted into a unit price, cost per system or return on investment. Corrections can be sent through Servospace's contact page.