Walmart is still the largest employer in the United States, and despite being overtaken this year by Amazon in terms of total revenue, it is still by far the largest US retailer by revenue (AWS making up a considerable portion of Amazon’s revenue). As historian Nelson Lichtenstein once argued, it is the template corporation of neoliberal capitalism, as GM was at mid-century. It is only natural that it would also be viciously anti-union, and for most of its existence, it has meticulously avoided signing a union contract in North America.

That changed on May 7th of this year, when 800 workers at a Walmart Regional Distribution Center in Mississauga, Ontario got a two-year contract with the company. Why would this be the moment when Walmart finally put pen to paper on a union contract? Has the company gotten soft in its old age, without the rabid anti-unionism of Sam Walton to guide it?

The story of the Mississauga win, as well as an attempt to explain it, is the subject of my latest column for Phenomenal World, which you can read here. Some highlights:

• Until this year, Walmart had not put pen to paper on a union contract in Canada or the US. As the largest employer in North America (with a headcount of 1.6 million in the US alone, it is still larger than Amazon) and with annual gross profits last year of $171 billion, Walmart has tremendous resources with which to fight off unions, and has developed a sophisticated playbook for doing so. It is not surprising that the company that labor historian Nelson Lichtenstein has called the “template organization” of neoliberal capitalism should have been, for most of its existence, an impenetrable anti-union fortress.

• The day after Ontario certified the union to represent the Mississauga warehouse workers, Walmart issued an across-the-board wage increase for every single logistics worker in Canada (warehouse workers and drivers both)—except for the eight hundred workers who had just unionized. As Gniposky summed it up, “Here’s the congratulations on getting a union. You are now making less than every single one of your coworkers.”

• “Workers at Walmart in the US and Canada have long made it clear that they want a union,” Nelson Lichtenstein told me. “To actually have a signed contract is a major breakthrough and something genuinely new when it comes to the largest private employer in North America.”

• Chokepoint analysis—based on the idea that organized labor is going to be most successful at points in the production and distribution process where the disruptive impact of labor actions is relatively high—has a mixed reputation in the labor world. In part, this is because it involves the idea that some sites of work are more structurally important to a company’s operation than others, and thus that some workers’ struggles are more strategic or important than others. But when it comes to the contemporary logistical economy, in which the flow of commodities has increased in speed and efficiency but also fragility, and at a moment when labor is limited in the resources it can allocate to any organizing campaign, chokepoint analysis matters more than ever. Organized labor ignores it at its own peril.

Read the full column here.

On Monday, I met a goal of mine since starting this Substack almost three years ago, which was to create distribution network primers for the top twenty largest private-sector employers in the United States that had sizable distribution footprints. Here are those twenty, with the primers linked where applicable.

• 1) Walmart - 1.6 million employees

• 2) Amazon - 1.1 million employees

• 3) Home Depot - 422,500 employees

• 4) Target - 415,000 employees

• 5) Kroger - 403,000 employees

• 6) UnitedHealth - 390,000 employees

• 7) FedEx - 375,000 employees

• 8) UPS - 370,000 employees

• 9) HCA Healthcare - 320,000 employees

• 10) Berkshire Hathaway - 310,240 employees

• 11) CVS - 300,000 employees

• 12) Albertsons - 280,000 employees

• 13) Allied Universal - 280,000 employees

• 14) Lowe’s - 276,000 employees

• 15) TJX Companies - 248,000 employees

• 16) Starbucks - 223,000 employees

• 17) Costco - 223,000 employees

• 18) Walgreens - 200,000 employees

• 19) Dollar General - 194,000 employees

• 20) Walt Disney - 172,000 employees

To my knowledge, this is the best kept list on the internet of the top private-sector employers by domestic workforce size, and it is the only place you can find all of this distribution network information, including where a company’s distribution facilities are, what the differentiation between facility functions looks like, what kind of automation is present in each facility, etc. Every primer comes with a map of the distribution network of the particular company under investigation, and I’ve made the underlying data available from each map here.

Employment Concentration

This post is mostly simply to announce this milestone in the On the Seams project, but while I’m here, it’s worth making a brief comment about the concentration of employment in the US. Here are the top ten American employers by total workforce size in 1955, according to the first Fortune 500 list.

• 1) General Motors - 576,667 employees

• 2) U.S. Steel - 268,142 employees

• 3) General Electric - 210,151 employees

• 4) Chrysler - 167,813 employees

• 5) Exxon Mobil - 155,000 employees

• 6) Amoco - 135,784 employees

• 7) CBS - 117,143 employees

• 8) AT&T Technologies - 98,141 employees

• 9) Goodyear Tire & Rubber - 95,727 employees

• 10) Firestone Tire & Rubber - 90,000 employees

Now, a few caveats before comparing: not everyone responded to Fortune’s request for company information, most notably Ford, which would have likely come in second behind General Motors in terms of employee count. So if I were doing this work in 1955, the list would look different. This is also a total employee count, but since capital was much more domestically trained in that period, let’s just take these to be in the ballpark of the domestic numbers.

In 1955, the US population was 165.9 million, and today it is 345.4 million. According to the numbers above, the top ten companies in 1955 employed 1.9 million people, or 1.1% of the American population. Today, the top ten companies employ 5.7 million people, or 1.7% of the population.

Again, this is back-of-the-napkin math that comes with caveats, but it should come as a surprising finding. In books like David Weil’s The Fissured Workplace and Gerry Davis’s The Vanishing American Corporation, one is led to believe that neoliberalism has involved a dramatic economic decentralization, and yet the big employers today are employing a larger percentage of the population than they did in 1955. If we go by the Biden-era joint-employer ruling from the NLRB, we’d undoubtedly find even more concentration: I’ve previously estimated that there are roughly 275,000 Amazon DSP drivers and 130,000 Amazon Flex drivers, which would bring their total domestic workforce count up above 1.5 million.

This is just to further undergird a point I return to somewhat incessantly, which is that the large employers in the US are still very large employers, comparable to or even larger than those of mid-century, and that if organized labor is going to be revitalized in the twenty-first century, it must now, as it did then, go after the companies that set the employment standard in the economy as a whole. The physical footprint of these companies looks quite different today than it did 80 years ago, but that is a problem to be solved with more strategic organizing, not a reason to give up on strategic targeting.

The TJX Companies are comprised of TJ Maxx, Marshalls, Sierra, HomeGoods, and Homesense. The company groups the first three as their main, off-price retailing segment, and the latter two as their secondary, off-price home goods segment. TJ Maxx was created in 1976, and the company acquired Marshalls in 1995. Importantly for my purposes, the two companies share a distribution network (the “Marmaxx” network). Sierra was acquired in 2012, and given its specialization in off-price outdoor wear, as well as its geographic limitation, it has maintained its own limited distribution network. HomeGoods (opened in 1992)/Homesense (2017) also has their own separate distribution network.

TJX is the undisputed king of the off-price retailing market. TJX’s net sales of $60.4 billion last year is more than that of Ross ($21.1 billion), Burlington ($11.5 billion), and Nordstrom Rack ($5.2 billion), its closest competitors, combined. Off-price retailing takes advantage of the inevitable waste in retail, buying up the big brands’ surplus and giving customers a deal. If TJX has been particularly successful in this market, it’s because they have expanded the appeal of discount shopping to the middle classes, framing it as more of a treasure hunt than something for which any shopper might feel embarrassed. The affordability crisis in America has, of course, helped their case.

TJX has roughly 1200-1300 global buyers who are given a fair amount of autonomy to cut deals with the company’s roughly 21,000 suppliers. According to TJX’s latest 10-K, in the US they have 19 Marmaxx distribution centers, 7 HomeGoods/Homesense distribution centers, and 2 Sierra distribution centers, which more or less accords with what I was able to dig up. Many of the Marmaxx DCs are listed under “NBC Distributors” in the OSHA data - a wholly owned subsidiary of TJX.

As with all of the brief primers on On the Seams, this one comes with a map of TJX’s distribution network.

Marmaxx

There’s not a ton of information out there about the Marmaxx distribution centers. They are large facilities, each employing 1,500 workers on average and occupying 900,000 square feet on average. The largest Marmaxx DC by workforce size in Phoenix employed 2,228 people in 2025. With 2,603 TJ Maxx and Marshalls stores, their 20 DCs (17 1P + 3 3P was what I was able to find) each supply roughly 130 stores.

From a video inside from about a decade ago, they appear to be pretty simple operations, with a conveyor belt with tilt trays for small sortables, but a lot of manual picking. But ten years ago TJX’s revenue was half of what it is now, and there DC footprint was not that much smaller, so they must have gotten more efficient in the meantime.

From one angle, as a primarily brick-and-mortar operation, it might not seem that Marmaxx distribution needs to be that efficient, but the company has high inventory turnover and prioritizes the speed with which items get to the rack (they have a slogan, “from door to floor in 24”). As the off-price model requires customers knowing and prizing the "real” value of trendy items, and as TJX is often getting their inventory after other retailers, speed is a more important consideration for them than one might think.

TJX was pretty late to the e-commerce game, with tjmaxx.com starting in 2013 and marshalls.com in 2019. Today it accounts for only 2% of net sales. It appears that most e-commerce order fulfillment is done through a large Marmaxx FC in Memphis, and a secondary FC in Vandalia, OH.

TJX does not appear to operate its own private fleet, contracting out all of its freight. They also have a few third-party warehousing operations, listed in the “Marmaxx Miscellaneous Facilities” in the map. Finally, while I’m not certain they should be included in a primer on TJX’s distribution network, the Marmaxx operation has “ARRCs” (Asset Recovery and Recycling Centers) around the country that process and recycle store materials like cardboard, hangers, display units, etc. These are waste management facilities in one sense, but I imagine part of the hope with them is to do store materials replenishment on the cheap, and thus to avoid having facilities like Walmart RREWs.

HomeGoods

By my count, HomeGoods has 8 distribution centers (from which Homesense inventory is also supplied) - one more than reported in their most recent 10-K, and it’s possible one of their 3P DCs has closed. It had one fulfillment center in Groveport, OH when they briefly ventured into e-commerce, but they shuttered that whole operation in 2023. The average HomeGoods DC has 1,200 people working there and is 865,000 square feet in size. As there are roughly 1,000 HomeGoods and Homesense stores, each DC supplies about 125 stores.

Walgreens is the second-largest pharmacy chain in the US after CVS. After an almost hundred-year run as a public company, they were recently bought for $10 billion by private equity firm Sycamore Partners and immediately broken into five different companies. One company is Boots in the UK, and the other three are their former healthcare subsidiaries, Shields Health Solutions, CareCentrix, and VillageMD. VillageMD, by far the largest of these three by revenue, employed about 20,000 people in 2023. If Shields and CareCentrix employed another 20,000 together, the new standalone Walgreens would still employ ~200,000 people, keeping the company in the top twenty largest employers in the US.

For now, Walgreens is saddled with substantial debt thanks to the acquisition, and so seems unlikely to invest in growth or even adaptation. Even before the acquisition, they were shrinking their distribution footprint, closing two distribution centers employing about 650 workers between them in March 2024. After the acquisition, they closed another one in Houston, leaving 13 total distribution centers, by my count. At the same time, Walgreens has continued the expansion of its micro-fulfillment centers, which “use robotic technology to fill and ship prescriptions to stores,” after a brief pause in 2023. I count 17 micro-fulfillment centers as of the present.

As with all of our brief primers, this one comes with a map of Walgreens’ distribution network, the underlying data for which can be found here.

Distribution Centers

If the video of their distribution center in Anderson, SC below is representative, their distribution centers have been remarkably automated affairs for some time. Designed by Schaeffer Systems, they include tilt tray sorters to deliver cases of goods for sorting into totes, an Automated Storage and Retrieval System (ASRS), and mobile unit picking companions. The video starts with the claim that nearly half of the employees at this DC have a physical or mental disability, which frames the whole description of the facility’s automation. Each DC has an average of 518 employees; the largest DC by head count is in Mt. Vernon, IL, with 1059 employees.

Micro-Fulfillment Centers

The Micro-Fulfillment Centers are small (72,000 square feet on average) and highly-automated operations for filling prescriptions. In 2023, the company claimed to fulfill 60% of their prescriptions through automation, and I assume that meant that 60% originated at an MFC. The justification is that MFCs free pharmacists from having to do prescription fulfillment so that they can perform other services in store, such as diagnostics and testing.

Each MFC services about 200-300 stores, and each one employs between 150-200 workers. In January 2021, Walgreens acquired a majority stake in iA, which has developed the automation behind the MFCs.

My new column for Phenomenal World is out today, this month on the virtues of the Costco model of low SKU count/in-store shopping v. the Amazon model of infinite assortment/home delivery. The basic argument is that the Costco model is socially far preferable and logistically more elegant than that of Amazon, and that there are some important lessons of Costco’s success for Mayor Zohran Mamdani’s plan for public grocery stores.

Some highlights:

• Late to the e-commerce game, minimally invested in their distribution network, and committed as ever to an artificially-limited assortment, Costco is the anti-Amazon. It embodies the precise opposite of everything imagined by the e-commerce futurists—and yet somehow its revenue has grown by an average of more than 10 percent every year for the last five years.

• While the e-commerce share of retail has been steadily growing, it’s still under 17 percent in the United States, and one wonders in a society as anti-social as our own if customers find in-person shopping, degraded a “social” venture as it is, desirable even in its inconvenience.

• To put it crudely, having someone in a Sprinter van deliver a recently-purchased toothbrush to your doorstep is simply not a universalizable action, from either a business or logistical standpoint. It is a modern feat that Amazon is capable of doing this, but that it can be done does not mean that it should, nor even that it can be done writ large.

• This has allowed Costco to achieve an astonishingly low rate of labor turnover: compared to 60 percent turnover in retail generally and 150 percent in Amazon warehouses, the annual workforce turnover rate at Costco is just 6 percent. This is often treated in the business press as a matter of company “culture,” but it has a clear economic underpinning. When you minimize overhead, you can simply pay workers more without squeezing your overall margin.

• Why do you need to figure out how to integrate robotic arms into your fulfillment operation that is dominated by autonomous mobile units when you can cross-dock full pallets? Is it more logistically impressive to solve difficult problems or to eliminate the need to solve them in the first place?

Read the article here.

My latest Phenomenal World column is out today, and it’s a deep dive on the massive changes happening in trucking.

Some highlights:

• “If you got it, a truck brought it.” In this concise manner did Jimmy Hoffa once characterize the economic significance of the trucking industry and the potential power that could result from organizing it.

• Hoffa proudly told Life magazine in 1959 that, if he wanted to, he could “bring a major portion of U.S. transportation—and thus the entire economy—to a halt.”

• A number of seismic developments in trucking have unfolded over recent years, including elevated demand from the staggering data center construction boom, anti-immigrant regulatory changes that have tightened the labor market, and an important Supreme Court decision. Together, these events and other regulatory developments offer an organizing opportunity for the Teamsters.

• From one angle, it’s concerning that flatbed volume, historically tied to the pace of housing construction, is feeding the AI demon instead of adding to American housing stock. But the trucking industry is not concerned with such things, and the fact that this demand surge is industrial rather than consumer-driven gives them a great deal of confidence.

• As a result of these changes to the English language proficiency enforcement and the new policy on non-domiciled CDLs, J.B. Hunt estimates that between 214,000 and 437,000 active CDL holders will be removed from the workforce over the next two or three years. That means that between 5.6 and 11.4 percent of heavy and tractor-trailer drivers will be put out of service during this time.

• A tighter labor market, heightened industrial demand, a new push for regulation that drives out some of the worst low-road operations, a likely shift toward larger firm size, a decimation of the brokerage industry—these are the ingredients of a unique organizing opportunity in trucking, one that unfolds against a broad demoralization amongst truck drivers.

• Major resources would be needed to make the Teamsters the collective masters of the highways once again. But it also matters how their organizers are approaching this work. Shop by shop organizing can be successful, as I showed in my March column on Home Depot’s new last mile. But the National Labor Relations Board election is too fraught and too slow a process to be of much use when scale and speed are of the essence.

Read the full article here.

The Center for Work & Democracy has released a new report, on which I’m a co-author, entitled “Renewing Labor and Winning at Amazon.” You can download the full report here.

It is no secret to anyone in the labor movement that labor’s power continues to decline. As a result, working people have been receiving a declining share of the wealth produced by economic activity, and well-paying and unionized jobs are both rare and increasingly precarious. How to reverse the decades-long decline of union density is the billion dollar question for mitigating inequality and revitalizing American democracy today.

If the labor movement is serious about substantially growing its ranks, it must have the large employers of the day in its sites. The largest employers in the country define the labor relations regime that prevails in the economy at large, and the absence of a unionized presence at these companies largely ensures that workers and their interests will be disregarded by executives and managers.

Today, no large employer is more important for the labor movement to tackle than Amazon, which is set to be the first company in history to reach $1 trillion in revenue by 2029. It has such a diversity of segments, from online retail to logistics services to cloud computing to grocery, that arguably no union jurisdiction is untouched by its rise.

In a new report out today from the Center for Work & Democracy, “Renewing Labor and Winning at Amazon,” Michael McQuarrie, Jonathan Rosenblum, and Benjamin Y. Fong lay out the stakes of this fight and the strategic opportunities for this key labor struggle of the twenty-first century. We hope that this report initiates a broad conversation amongst labor leaders, organizers, and scholars about the revitalization of labor and the prospects of bringing unionization to the corporate behemoth of our day.

I might as well start with the most surprising thing I learned in researching this primer: the Teamsters have contracts at a full 40% of Martin Brower distribution centers, Martin Brower being the primary logistical partner of McDonald’s. They furthermore have the west coast on lockdown, with the MB distribution centers in City of Industry, CA (Local 63), Stockton, CA (Local 439), Sumner, WA (Local 117), Portland, OR (Local 117), and Phoenix, AZ (Local 104) all having union contracts. I’ll spell out the political importance of this in a moment, but that’s a wild amount of union presence in the distribution network of a major employer in the US.

McDonald’s doesn’t officially crack the top twenty largest employers, as they only have 105,000 employees in the US according to their latest 10-K. However, this number does not include anyone who works at a McDonald’s franchised restaurant, and that would add anywhere from 800,000 to 1 million people to their US workforce alone, making them the third largest employer behind Amazon.

The McDonald’s distribution network is managed by the North American Logistics Council (NALC), which includes five companies: Armada, Earp Distribution, Martin Brower, Mile Hi Foods, and The Anderson-DuBose Company. Martin Brower is by far the most important player, with all of the other companies serving smaller regional roles. In the map, I’ve included both average annual employee counts for each facility, as well as the union representative, if the site has one.

With roughly 13,000 locations in the US and 35 distribution centers total in the NALC, each DC is responsible for between 350 and 400 locations on average. Apparently french fries comprise 12% of what Martin Brower moves for McDonald’s, and they have had fry-specific automation for a while now:

There doesn’t seem to be real differentiation in the network. Each distribution center receives inventory from suppliers - J.R. Simplot, Lamb Weston, and McCain for fries; Coca-Cola for beverages; Tyson, Cargill, OSI, and JBS for meat - and sends out loads for store replenishment. Reyes Holdings, which owns Martin Brower, is also a major distributor of Coca-Cola products in general.

Now, back to marveling at the degree of unionization: as I sadly learned recently, the fast food council in California, the fruit of what was once a very exciting sectoral bargaining campaign led by SEIU, is now in limbo, without a chair and with no clear direction. The council and a $20 minimum wage in fast food were what the industry was willing to concede in 2023 after a high-profile legislative fight. It seemed like a good compromise at the time, but in retrospect it’s clear that the thing the industry really cared about, and which labor eventually gave up on, was a provision in AB 1228 that would have held ”fast-food franchisors jointly liable for labor violations at their franchisees’ establishments.”

The “fissuring” of workplaces (to use David Weil’s term) has been a core feature of neoliberal capitalism in the United States, and uncoincidentally a central barrier to organizing. From the regulation of the trucking industry to the Delivery Protection Act in NYC to the California fast food campaign, there are various efforts around the country toward reconsolidation, which would make the organizing landscape much more hospitable if successful. But employers understand well the dangers this poses and will fight such efforts tooth and nail (except in trucking, where, for reasons that I’ll get into soon, regulation’s coming no matter what).

Which is why it’s important for labor to use every tool in its bag. If there’s a hammer lying around, best to put it to constructive use.

Now that I’ve got my new Amazon map and primer up, I thought I’d go back and compare last year’s numbers to this year’s.

AAE is “average annual employees” as reported in the OSHA ITA data. Note that the total counts toward the right are not the entirety of the warehousing workforce, as I’ve left out a bunch of facilities for which I either didn’t have good numbers for last year (XL Sortation Centers and Delivery Stations), or which were staffing up in 2024 and so unrepresentative (National IXDs), or where the employment range is pretty all over the place (Customer Returns). Also note that I put up the two primers in May of 2025 and May of 2026, and I’m being lazy and counting those as 2024 and 2025.

As I said previously, I think the Delivery Station count and workforce shrinkage is real, explained in large part by the notable rise in the counts and workforces of Rural-Super Rural Delivery Stations and Sub-Same Day Fulfillment Centers. I think the shrinkage also in the Traditional Non-Sort Fulfillment Center network is also real, for reasons that I’m less clear about. I’ve long predicted continued shrinkage in their ARS Fulfillment Center network, so no surprise there.

On the whole, it’s noticeable that most average facility headcounts seemed to shrink - often not by much, but evidence that Amazon is on the whole doing more with a little bit less. Now of course they’re also investing in a new grocery network, new facilities to deal with Amazon Supply Chain Services volume, the new 1DC and Amazon Now networks, so they are and will be hiring blue-collar workers elsewhere. Which is to say again that

we’re likely to see more of the same from Amazon: intense labor surveillance, discipline, and subcontracting/gigification combined with displacement of labor through robotics improvements that are mostly masked through company growth.

However, having said that, I was much more stringent in my accounting for closed facilities this year than I was last year, so it’s possible/likely that the overall worker counts in each category for 2024 are slightly inflated. My educated guess is that the warehouse workforce size as a whole and in reality might have seen a bit of shrinkage, but probably less than the 6.71% indicated above.

This is the third iteration of my brief primer on Amazon’s distribution network (first and second versions here and here), and it’s now at the point where I have to drop the adjective “brief.” This primer, and the associated map, are updated to the best of my knowledge and ability as of the present—May 2026. These are not facilities that are planned or in the works, but a snapshot of actually existing and functioning Amazon facilities at present.

The map’s a little overwhelming to go through as a whole, so in the following, I’ll use the Los Angeles/Inland Empire region as an example. In addition to providing a little more space than the larger east coast markets, which are so closely packed that it can be difficult to make out what’s going on, this area is also of overriding economic importance given the volume that runs through the Ports of Los Angeles and Long Beach and the vast tracts of warehouses of the Inland Empire (tracked here by Radical Research).

1. Inbound Cross-Docks (Purple)

Note: I’ll keep this section short because I did a whole different brief primer about Amazon’s inbound network here.

The first step in the Amazon supply chain is the inbound receiving network, made up of Inbound Cross-Docks (IXDs). These facilities are naturally close to ports and rail yards. Goods are delivered to Amazon for distribution in various forms, and the whole purpose of the IXDs is to prepare inventory for the Fulfillment Centers (FCs), which are the next step in the chain. The FCs are meant to be highly efficient “just-in-time” environments; inventory is not meant to be there for more than a week. Thus IXDs act as distribution buffers so that FCs are not overwhelmed.

“Cross-docking” is a kind of anti-warehousing arrangement pioneered by Walmart wherein bay doors are on both sides of a facility: container loads are backed into one side, and then transloaded into full truckloads on the other side. Thus, traditionally IXDs were not intended to be like the enormous holding facilities that, for instance, Walmart has built out in its import distribution centers.

There are three basic types of IXD: 1) National IXDs, roughly a million square feet each and most of which were built in 2024, 2) Regional IXDs, roughly 600,000 square feet each and which were the original IXD network before the national IXDs were built, and 3) One DCs, a new kind of IXD just for high-velocity inventory. Despite being smaller, the regional IXDs (2,269 employees on average) tend to employ more people than the national IXDs (628 employees on average) according to the most recent OSHA Injury Tracking Application data, but I think that’s just because the national IXDs are still staffing up. National IXDs are the first layer in the inbound system; regional IXDs are downstream and all replenish FCs directly.

A few other IXD notes: some IXDs are virtual nodes within larger facilities, i.e., a small inbound area in a Fulfillment Center, for instance. These are marked as “virtual” on the map. There are also a variety of third-party transloaders, mostly Maersk facilities, that Amazon uses near the ports. For instance, about 7% of containers coming in to LA/Long Beach are consigned for XLA4 in Santa Fe Springs, CA, but nothing is unloaded there. I have not included the 3P transloaders on the map.

As you can see here, most of the IXDs around Los Angeles are in the Inland Empire, and there is a preponderance of them there - roughly 1/8 of all Amazon IXDs. This speaks to the enormous importance of the ports of Los Angeles and Long Beach, through which roughly 55% of all Amazon imports flow. The other major ports for Amazon imports are Oakland (14%) and Seattle/Tacoma (12%). This means than 80%+ of Amazon’s imports by TEU volume are coming in on the west coast.

2. Fulfillment Centers

Amazon is perhaps best known in popular and academic accounts for its awe-inspiring FCs. It even gives free tours of these facilities, understanding how associated they are with the Amazon brand.

FCs are generally located in suburban/exurban areas and near major highways and parcel hubs, and they come in a great variety.

2a. Sortable Fulfillment Centers (Blue)

The Sortable Fulfillment Centers handle anything below certain dimensional and weight limits (~20 pounds). In this sense, Amazon does not care what an item is so much as whether it is conveyable and processable within their sortable facilities.

Some of the FCs listed in this layer are traditional sortable facilities, meaning that they are not subject to the robotics wizardry of Amazon Robotics Sortable Fulfillment Centers (ARS FCs), but it is the destiny of all Sortable Fulfillment Centers to become ARS Fulfillment Centers (only a small percentage of Sortable FCs are not ARS FCs), so I have grouped these together. This is because Amazon’s robotics technology is most suitable for sortable item facilities.

The biggest technology win for Amazon has been the widespread adoption of Kivas. The workhorses of ARS Fulfillment Centers are the Hercules Kiva robots, which look like giant Roombas. They move “pods”—giant stacks of inventory weighing up to 2,000 pounds—to and from human stowers and pickers. Hercules robots move around on floors with QR codes every couple feet, which is how they recognize their position and know where to go next. They are generally physically separated from workers by a chain link fence, and only special technicians wearing tech vests that deactivate Hercules robots nearby are allowed behind that fence.

The Hercules has made randomized storage very efficient: in a traditional warehouse, you’d want to keep all the toothbrushes grouped together with other toothbrushes and bathroom items so as to be able to find them easily. At the ARS facilities, new inventory is placed randomly in the stacks carried about by the Kiva robots and bounces around the facility floor. Amazon wants there to be toothbrushes spread throughout the stacks so that when someone orders one, the nearest Kiva robot with a toothbrush in one of its stack’s pods can get to a pick station quickly. It’s a system that would have been barely imaginable just a decade ago.

Amazon has a lot of cool robots. None are as close to important as Hercules. It’s what saves a human being from having to walk a few football fields to sift through items for a particular order. According to Brad Stone in Amazon Unbound, the average pick rate in a Fulfillment Center jumped from 100 items/hour to 300-400 after the introduction of Kiva technology.

Recently, Amazon has invested a great deal in its robotics arm programs in an effort to automate away the baseline picking and stowing role. Instead of having human beings “stowing” items in the mesh-band pods carried around by the Kivas and then “picking” them again after an order comes in, they want robotic arms to be doing that work. They have tried out the “Sequoia” system at SHV1 in Shreveport, LA, but my guess is that they’ll invest more in the “Vulcan” program, as it can work with their normal mesh-band yellow pods:

AR Sortable FCs employ 2,948 people each on average, slightly down from my last calculation of 3,069. As I argued here, that number is the most likely to see shrinkage in the coming years of any of the other employment numbers listed here, as Sortable FCs continue to be the primary target of automation deployments. Traditional Sortable FCs employ 1,439 people each on average, and the “Soft Lines” sortable facilities (those that deal with apparel, bedding, textiles, anything soft) employ 1,609 people each on average. Though during peak season, employment numbers might be 50% above normal.

2b. Non-Sort Fulfillment Centers (Dark Blue)

Much less exciting technologically than the ARS FCs are the Non-Sort Fulfillment Centers that deal with larger items (between 20-60 pounds) that are not conveyable/processable within sortable facilities. It’s possible that automation will come for these facilities one day too, but it hasn’t really happened yet. 873 people on average work at a Non-Sort FC.

2c. Sub-Same Day Fulfillment Centers & Amazon Pharmacy (Brown)

The Sub-Same Day (SSD) Fulfillment Centers are the fastest growing category of Fulfillment Centers. I did a whole article about them in Phenomenal World here.

They are something of a replacement for the Prime Now service, an ultra-fast Amazon service that shuttered in 2021. But whereas Prime Now hubs were very small (25,000 square feet) and focused on a pretty limited number of items that could be delivered within an hour or two, SSDs are bigger (around 200,000 square feet) and, as their name implies, focused on getting your item their very fast (in less than a day) but not as fast as Prime Now hubs. Items fulfilled at SSDs do not go to separate sortation or delivery facilities but are loaded directly into Amazon Flex drivers’ cars.

As you might guess, these FCs are much closer to population centers than other FCs, typically nestled along with Delivery Stations. They employ 390 people each on average.

SSDs have their own distribution centers (listed as “DC”s in the map). These are on average 1,000,000 square feet in size, and each employs 815 people on average. Their whole job is to replenish SSDs. In a few cases, SSD DCs are co-located with Amazon Fulfillment Centers, and in many more cases, Fulfillment Centers will have their own SSD code (V__# - indicated as “Co-located” in the map).

Finally, I’ve included Amazon Pharmacy in this layer because most Pharmacy locations are housed inside of SSDs. Only a seven are stand-alone operations (indicated as “Standalone” in the map), and I believe that’s a legacy of Amazon’s acquisition of PillPack.

2d. XL Network (Red)

Amazon has a dedicated network of facilities, including Fulfillment Centers, Sortation Centers, and Delivery Stations, that deal only with big and bulky items, i.e., things like dishwashers, refrigerators, and other large items that might require installation and removal.

XL facilities are very often co-located, with the fulfillment, sortation, and delivery functions all happening at a single facility. XL Delivery Station codes typically start with an “H”, as regular Delivery Station codes start with a “D.” If XL Delivery Stations are not co-located with XL Fulfillment and Sortation Centers, they are often co-located with regular Delivery Stations.

XL employment levels are much lower than at other Amazon facilities. At the XL Fulfillment Centers, it’s 279 people on average; at an XL Sortation Center, it’s 71; and at an XL Delivery Station, it’s 50.

2e. Grocery Fulfillment Centers (Gray)

Amazon recently closed its Fresh/Go offerings and announced same-day grocery service in 2,300 cities. It does much of the fulfillment for dry grocery from the SSDs, but for perishables, they have a network of Grocery Fulfillment Centers. These are on average 85,000 square feet in size, and each employs 172 people on average.

Amazon would clearly like their grocery segment to grow, and they’ve talked it (and the “everyday essentials category) up recently on quarterly calls. This might seem confusing, as both have thin margins compared to the obscene profits of Amazon Web Services, but both are key to bundling of purchases. If someone’s buying groceries or everyday essentials on Amazon, chances are they’re buying much else on Amazon.

2f. Supplemental Fulfillment Centers (Gray)

These Fulfillment Centers are overflow facilities, often only used during peak season.

Items leave the FCs in the very packages that show up in your lobbies and on your doorsteps. To get there, some are injected into the distributional streams of the US Postal Service, the United Parcel Service, and a few regional carriers, but most (let’s say 85%, though that number is growing every day) are handled by Amazon’s own logistics operation. And from the FC, the next step in that chain is generally either an Air Hub or Sortation Center.

3. Air Hubs (Yellow)

Amazon’s main air hubs include their superhub in Erlanger, KY (KCVG - 3,824 average annual employees in 2025), and their big regional hubs in San Bernardino, CA (KSBD - 1,383 employees), Hazlet, TX (KAFW - 1,775 employees), or Lakeland, FL (KLAL - 1,264 employees). They also have a sub-branch of their Kentucky superhub in Wilmington, OH (KILN - 1,410 employees), and one other 1P air hub in Baltimore, MD (KBWI - 1,102 employees). They recently closed a 1P operation in Riverside, CA (KRIV). The rest of Amazon’s air hubs are third-party air gateways, and to my knowledge no direct-hire Amazon employees work at any other airports listed on the map.

Without this air game, that random item that you order once and never again, which might only be stocked in a Fulfillment Center on the other side of the country, is not going to get to you in two days. So Amazon Air is key to the company’s ability to deliver on the promise of speed for their Prime customers. But sending packages on a plane is far and away the most expensive way to go, and after the pandemic splurge, Amazon realized that it needed to rein in air shipment costs. Thus their regionalization plan, which could be thought of simply as a “Put things on planes as little as possible” plan.

4. Sortation Centers (Green)

In 2014, Marc Wulfraat said that outbound transportation was Amazon’s “most significant weakness.” In 2026, according to ShipMatrix, Amazon surpassed the United States Postal Service in terms of parcel volume. I’m not sure anyone has appreciated what a mind boggling fact that is: from zero outbound capability to the largest carrier by volume in 12 years.

Key to this outbound operation has been the buildout of Amazon’s “middle-mile” Sortation Centers, which sort packages and palletize them by zip code. Multiple FCs might be sending packages inbound, and the aim here is to get packages organized for local delivery either by Amazon Delivery Stations or by USPS/UPS. Sortation Centers are 414,00 square feet on average, and each employs 832 people on average. However, the workforce at SCs is predominantly part-time, and the shifts are short (4-5 hours) and non-overlapping. This presents a formidable organizing barrier, and unsurprisingly the Teamsters have little penetration at Sortation Centers.

Sortation Centers function something like the big UPS and FedEx package hubs and are the keystones to Amazon’s play as a third-party parcel company. From what I hear, they will also play a key role in the rollout of Amazon Supply Chain Services more generally.

The majority of Amazon packages are going to be routed through a Sortation Center before arriving at a Delivery Station, but where Amazon can pick off particular lanes to go straight from Fulfillment Center to Delivery Station, they will. Minimal package touches is still the logistical ideal.

5. Delivery Stations (Orange)

The “last-mile” Delivery Stations are where parcels are loaded onto vans to be delivered within a roughly 45-mile service radius. They are located close to urban areas and are 237,000 square feet on average. They employ 246 workers each on average, and a couple hundred drivers might work out of each station - but not for the company (more below).

Delivery Stations are not the automated wonders that ARS Fulfillment Centers are, but many are now outfitted with Auto Divert to Aisle (ADTA) technology, replacing many of the people who used to work picking along a conveyor belt. As reported by Luis Feliz Leon, the ADTA system reduced sortation staffing by about 20% at Delivery Stations where it is used.

Last-mile delivery is by far the most expensive part of any package’s journey, accounting for 50% or more of shipping costs by some estimates. Amazon’s “innovation” to solve this problem is not some robotics wizardry but an aggressive subcontracting scheme, where they get different Delivery Service Providers (DSPs) to compete over contracts covering a certain volume for a Delivery Station. So despite wearing Amazon vests and driving Amazon-branded vans, no worker delivering a package to your doorstep or lobby is a W2 employee of Amazon.

In addition to DSP drivers, Amazon also has packages delivered through its Flex app, which you can think of as Uber for packages. People will sign up for 2-3.5 hour blocks to deliver a certain number of packages from their own vehicles and get paid per block. Most Flex drivers pick up from Sub-Same Days, but they’ll pick up from Delivery Stations too. Often the vans will clear out of Delivery Stations by lunch time, and then Flex drivers will start to trickle in. Amazon has other last-mile programs - sidewalk roamers, drones, e-bikes, walkers - but none are bound to replace paying a subcontracted or gig driver any time soon.

6. Rural-Super Rural Delivery Stations (Teal)

For most of its existence, Amazon has built mathematically around population density, and so it just didn’t have facilities in many parts of the country. That all changed with their Rural-Super Rural (RSR) buildout beginning around 2023. Rural customers definitely buy things, and so it’s a new market for Amazon, but my hunch is that the main impetus behind the RSR buildout has to do with the post office: both wanting to wean off reliance upon USPS for rural deliveries but also to be in a position to benefit from postal privatization should it occur.

From what I know, RSRs are just rural delivery stations, though many have separate Sortation Centers co-located (noted in the Sortation Center layer). They are 100,000 square feet on average in size, and each employs 114 people on average.

That map screenshot of LA and the Inland Empire won’t work to depict the RSR’s because there are precisely zero in that area. Here they are in the western US:

7. Miscellaneous Facilities (Gray)

7a. Customer Returns

There is a great amount of variation in Amazon’s Customer Returns centers, so averages don’t really work. Its largest returns center by employee count is LAS2 in North Las Vegas, employing 3,505 people. Many of the other standalone CR centers also employ in the thousands, but a good number of these are built into Sortation Centers, and when that’s the case, employment is in the 300-400 range.

7b. Make on Demand

Make on Demand production facilities print books on demand. They’re an abomination. We will be judged poorly by history for treating books this way.

7c. Merch by Amazon

Merch by Amazon production facilities make custom t-shirts/hoodies on demand. They’re fine. Merch can be treated this way.

7d. Kariba/Storage and Distribution Centers

As I understand the “Kariba” or Storage and Distribution Centers, they’re a middle layer between the IXDs and the FCs - an upstream storage site to make sure the FCs aren’t overwhelmed. They’re pretty big - 916,000 square feet on average - and each employs 702 people on average.

8. Other Facilities Not on the Map

8a. Amazon Now

These are brand new and quite small (~5,000-10,000 square feet) facilities to do ultra-fast, 1- and 3-hour delivery for the highest-volume SKUs. It seems like they’ve re-created the now defunct Prime Now, with threats from Walmart and others in the ultra-fast space, but they’re even smaller than Prime Now hubs because Sub-Same Days can handle much of the fast, as opposed to ultra-fast, volume. They will be on future iterations of the map once they’re built out, but these facilities are still coming online.

8b. Vendor Flex

Vendor Flex, as I understand it, works in one of two ways: either it is an Amazon fulfillment operation set up in a small corner of a third-party warehouse in order to do dedicated volume for them (like they have at a Quaker Oats distribution center in Carlisle, PA), or it is a small corner of an Amazon distribution node dedicated to one vendor’s fulfillment. Here’s a label for a package that originated at a DHL Supply Chain warehouse in Palmetto, GA before being routed through three different Sortation Centers on its way to being dumped to USPS.

8c. Corporate

Amazon has offices everywhere, but their corporate presence is especially strong in Seattle/Bellevue/Redmond, WA (HQ1), Arlington, VA (HQ2), Boston, MA (Robotics HQ), Nashville (Logistics HQ), and the Bay Area, CA.

8d. Amazon Web Services Data Centers

Many other people are keeping track of the data center buildout with a much closer eye than me, but I might put up a primer/map if I can synthesize something useful. So more soon maybe!

So first, the deflationary view of this announcement: third parties have been utilizing Amazon’s logistics network for a while now, so this isn’t some revolutionary new venture for the company. That bit about “Parcel Shipping” in the graphic is just Amazon Shipping, which was announced in 2023. Here’s an example of an Amazon Shipping label from my package label collection project (which you can contribute to here), indicating picking up at Delivery Station DLX5 in Los Angeles and delivered via USPS.

They also announced Supply Chain by Amazon in 2023, offering “end-to-end supply chain management” for all of its sellers. But Supply Chain by Amazon, as opposed to ASCS, was really just an extension of Fulfillment by Amazon (FBA), offering FBA clients the option of having Amazon manage their inventory all the way upstream. This new announcement is bigger than that in that third parties can theoretically pick and choose any part of Amazon’s logistics network that they want to use, even without doing fulfillment through Amazon. So that could mean, in a full reversal of Amazon’s original strength in fulfillment, that someone could use Amazon as a freight broker to get their stuff across the ocean, have it routed to their own warehouses stateside, completely bypass Amazon Fulfillment Centers, but then use Amazon Shipping to deal with last-mile delivery. Wild stuff.

However, some of the big names announced with ASCS were undoubtedly just Supply Chain by Amazon customers who are now ASCS customers. Land’s End was said to be “using a unified inventory pool within Amazon’s network to fulfill orders across multiple sales channels,” which either just means they’re using FBA or Vendor Flex, a program where Amazon has third-parties set up their own micro-fulfillment operations within Amazon warehouses. In other words, much of what was announced on Monday already existed. I’m guessing that Amazon nailed down a big contract or two and then decided to rebrand a few plays within a unified program.

In addition, the assets listed in their press release are impressive in the abstract - 80,000 trailers, 24,000 containers, 100 aircraft - but less impressive when you know those things are already being used to move Amazon’s own stuff around. There is a story circulating out there that Amazon Web Services simply grew from “excess capacity” in Amazon’s computing infrastructure, and the thought here is that ASCS is similarly growing out of excess capacity in the company’s distribution network. The AWS bit is a myth (as debunked well here), and it’s just as fanciful to think that Amazon’s going to serve an American Eagle or a Procter and Gamble with the extra space in their existing fleet.

For much of the year, Amazon might have a bit of extra capacity in its network, though that bit is likely used up during peak, when they have all their people working overtime. So what happens when that “excess capacity” goes away in the most important months of the year for retailers? When companies like 3M or Procter and Gamble sign a contract, they aren’t looking to have their stuff shipped reliably for nine months of the year. So the idea that Amazon could build something as ambitious as ASCS with the excess capacity in its fleet doesn’t make sense; if they want it to succeed, they have to build new infrastructure to handle ASCS volume.

HOWEVER, with the deflationary perspective now laid out, it’s worth also emphasizing that Amazon is definitely in the process of building this new infrastructure. They are adding to their sortation network to accommodate new cross-docked volume (for retailers who want to import their goods with Amazon but have it sent to their own warehouses once state-side), and Inland Empire residents are becoming accustomed to seeing new “Amazon Business” trucks driving about. As Freightwaves reported, this could also be an opportunity for Amazon to grow its intermodal volume significantly. Intermodal is more or less useless for Amazon’s outbound operation, where speed is the name of the game. But with lots of containers and slower timeframes for B2B distribution, intermodal makes a lot more sense, and they wouldn’t need to add stress to their trucking network.

In sum, there are plenty of reasons to question whether this announcement is truly as “disruptive” as the business press is claiming, but ASCS is absolutely a threat to Amazon’s parcel competitors, and Amazon is 100% investing in building out the physical capacity to make good on its promise. This all undergirds the case I’ve been making that Amazon is the existential threat to UPS, USPS, and their unionized workforces, and that the labor movement must go all-in to organize this multi-headed corporate behemoth.

I’ve updated my Amazon distribution network map in preparation to write a new primer. I sort of dropped the ball last time I did this data update work and only completed a part one of the primer, but I promise to finish the whole thing this time. This is part of my “brief primers” series, but the Amazon one is getting so long now that “brief” is no longer an adequate descriptor. In any event, it should be out in the coming weeks.

The last time I updated the map was about a year ago, so I thought it made sense to compare the new data to last year’s, to see how Amazon’s distribution network has evolved. Here are the facility counts in the map:

I’ve tried to contextualize some of the differences here in a way that can only be fully explained once that primer comes out, but some of the noticeable differences can be explained away as less significant than they might appear. As an example, there are now technically 98 Inbound Cross-Docks in the new map (as compared to 81 last year), but 21 of those are virtual nodes in larger facilities. So I don’t think the IXD network really expanded in terms of square footage.

But there are some meaningful changes here, most notably that the number of Delivery Stations (482→413 facilities) and Non-Sort Fulfillment Centers (105→88) shrunk, while the Rural-Super Rural (123→178) and Sub-Same Day Center (79→94) networks expanded. I don’t have a good explanation for the shrinkage of Non-Sort FCs. These are not, for the most part, very automated operations, but maybe they’ve figured out some efficiency there.

I devoted a whole column recently to Sub-Same Days, and there’s a chunk of my most recent column devoted to Rural-Super Rural. There are very different reasons for the growth of these segments: one to be able to do faster delivery in urban areas for high-volume SKUs, and the other to acquire rural coverage to wean off USPS. It’s not quite that simple, as I go into in the columns, but that’s how I’d describe it in a sentence. I assume the shrinkage in the Delivery Station network is mostly an effect of the RSR and SSD buildout. Both of these are alternate points from which packages are delivered; if they’re growing, Amazon doesn’t need quite as many Delivery Stations.

Again, primer coming soon!

Happy May Day!

My new column for Phenomenal World is out today, this time about the social disaster that postal privatization would wreak. I think most people understand that this proposal - long a dream of conservatives and neoliberals, and raised as a possibility in both the first and current Trump administration - is a bad one, but in the article I try to get very concrete about why and how.

Some highlights:

• “no private company has built logistical infrastructure even approximating the breadth and coverage of USPS. It’s not that private entities won’t deliver postal service; it’s that they quite literally can’t.”

• “To gird their bottom lines, investment bankers are prepared to sacrifice one of the last public agencies that physically binds the United States together as a nation.”

• “No other e-commerce company has built out the logistics network that Amazon has, and if postal service were privatized, Amazon would truly have its competitive moat.”

• City Carrier Assistant Jake (not a real name) spelling out the socially integrative function that USPS provides: “Information about your SNAP benefits is distributed by mail. Many residents of public housing as well as tenants of private landlords pay rent by mail. Many people, especially the elderly and those in public housing, are paid by mail. Tax returns, and especially refunds, are often distributed by mail. Many medications are delivered by mail. Our immigration system distributes notices and can accept submissions by mail. A paper trail in physical rather than digital form is easier to use in, e.g., housing court; sections of New York housing law actually require certain notices to be sent by certified mail.”

• President of the American Postal Workers Union Jonathan Smith: “It is still a miracle that you can get a letter from New York to California for less than a buck. Challenge anyone to try and do that through any other means. It’s just a remarkable operation. The ubiquity of it is something really to be cherished. In moments of crisis, in the early days of COVID, or after hurricanes or earthquakes, you read news stories, and people will often say, “I saw the letter carrier going down the street, and I knew we were going to be okay.” The Postal Service is the first one back into a community after a disaster like that. You can count on the mail moving and getting to you wherever you are. It really is quite a marvel.”

Read the full article here.

My monthly Phenomenal World column came out this week, this time devoted to Home Depot’s last-mile buildout and the opportunity for organized labor that it presents. In February, Temco drivers in Lithonia, GA became the first fully-owned subsidiary of Home Depot to unionize, and I situate this win within the company’s broader strategy and distribution network upgrades. In short, Home Depot has onboarded thousands of directly-employed drivers in recent years in their effort to win the speedy delivery game for bulky materials, and in the process, they’ve created a sizable toehold for the Teamsters.

My favorite part of the article was getting to interview Matt Rupley and Jaree Beatles, a Teamsters organizer and Temco worker-leader respectively, about the win in Lithonia. As Jaree concluded our interview, “If drivers come together, and they really stick together, the companies will have to start treating everybody fairly instead of treating us like a number.” More here.

Also I was on Jacobin’s Confronting Capitalism podcast with Vivek Chibber and Melissa Naschek recently here, and on Bungacast with Alex Hochuli and George Hoare here. Thanks to both crews for having me on to talk labor and logistics, and in particular the thorny problem of Amazon.

Three stories came out recently about the United States Postal Service:

• March 16th: ShipMatrix announces that Amazon is now the largest parcel carrier by volume period, finally overtaking USPS.

• March 17th: The Postmaster General testified before Congress that USPS “will be out of cash and unable to deliver mail, and pay vendors and employees, in less than 12 months unless Congress lifts the carrier’s statutory debt limit of $15 billion and eliminate mandates that handcuff its ability to make money and burden it with uncontrollable expenses.”

• March 17th: Amazon announces that it plans to cut ⅔ of the volume it sends through USPS.

Technically, these are three unrelated stories about the dire straits of USPS, but it’s hard to feel like their rapid succession is coincidental.

With the buildout of its rural network, Amazon has been positioning itself to be able to take over USPS service should the Trump administration proceed with postal privatization. When I updated my Amazon distribution network primer last summer, I found 123 Rural-Super Rural (RSR) facilities, and I’m pretty sure they’re over 200 now. RSRs and Sub-Same Days are the two fastest growing categories of Amazon’s distribution nodes.

Even with this phenomenal growth, I still think Amazon’s case to benefit from postal privatization is not as strong as, say, UPS’s. But Amazon is no doubt politically in a much better place than UPS to make their case, and if I had to guess, they’re assessing the current state of the world and thinking that they shouldn’t let a good crisis go to waste.

Operationally, Amazon’s biggest problem right now is that they are still in the infancy of figuring out reverse logistics. Which is to say, they are very good at outbound distribution, but less so at injecting outside parcels into that one-way flow. But with Amazon Shipping, they’re certainly figuring it out. Here’s a label I collected through my Amazon package label project of a package that was picked up by a driver and first processed at the DLX5 Delivery Station in LA, routed through three Sortation Centers (ONT5 in San Bernardino, CA, MTN6 in Sparrows Point, MD, and EWR5 in Avenel, NJ), before being dumped to USPS. No Delivery Station involved in the last-mile, but one involved in the first!

But it seems unlikely that they would scale this system up quickly enough to take over USPS service. My sense is what they would do is scrap USPS’s own distribution network but retain post office branches and inject all volume from there into the Sortation Center network, mostly bypassing their Delivery Stations for inbound parcels.

But this is just for parcels. They’d have no idea what to do with the 109 billion pieces of first-class and marketing mail that went through USPS’s system last year. And while they would most likely function at an operating loss not to piss the public off too quickly, it would only be a matter of time before Amazon went back on USPS’s Universal Service Obligation to “provide prompt, reliable, affordable, and efficient postal services to all Americans, regardless of where they live.”

Looking at USPS’s last-mile coverage is instructive here. In the state of Wyoming, USPS currently has 77 last-mile delivery facilities. UPS has 9, FedEx has 8, and Amazon has 3. How are any of these companies going to deliver mail six days a week? Even operating at a loss, postal privatization would likely involve immediate reduction of daily to weekly delivery service or worse.

In brief, then, postal privatization would be a true disaster, a fundamental interruption and degradation of a public service that Americans take for granted, and a political disaster for any administration under which it was executed. Most devastatingly, it would also result in a decimation of the postal workforce, a unionized stronghold in the country (about half a million union members in total). This is the future that Amazon is trying to hasten.

At the end of last year, I tried to do a comprehensive overview of Amazon’s warehousing automation and concluded that we shouldn’t expect fully “dark” warehouses and an associated employment apocalypse from the company, but rather that

we’re likely to see more of the same from Amazon: intense labor surveillance, discipline, and subcontracting/gigification combined with displacement of labor through robotics improvements that are mostly masked through company growth.

Shortly after that, I questioned the narrative that Amazon’s corporate layoffs were really about AI, and since then my skepticism of the “AI/the robots are coming” narrative has only hardened, less because I doubt the transformative nature of the technologies being deployed and more because I don’t think the operational context of AI deployment is being considered in doomer-ish accounts of such things.

Recently, Amazon held an internal meeting to address a

“trend of incidents” in recent months, characterised by a “high blast radius” and “Gen-AI assisted changes” among other factors, according to a briefing note for the meeting seen by the FT. Under “contributing factors” the note included “novel GenAI usage for which best practices and safeguards are not yet fully established”.

This meeting comes on the heels of laying off more than 10% of its corporate workforce in recent months, and new imperatives for all staff to incorporate AI use into their regular work flows. Amazon employees I’ve talked to say that the company is dead serious about having every employee use AI, and documentation of AI use is required in things like promotion packets.

From an individual perspective, this makes total sense. Every software engineer I know thinks Claude Code is a genuine problem for new computer science grads, and that the industry will never look the same. But when they say this, they are thinking about their own productivity. At a company like Amazon, however, it is not simply that individuals need to be productive but that many teams working together to maintain and improve a mind-bogglingly complex system need to be productive, and that is a social and organizational problem, not a technological one.

To be clear, this social and organizational problem can and will be solved, and when it is workers at these big tech companies will indeed be more productive than they were before. But at present, Amazon’s clearly out in front of its skis, and a hiring course correction seems inevitable. (Also: I’m still waiting on a FOIA request at USCIS, but I’m guessing, as with the layoffs of 2022-23, that Amazon onboarded a ton of new H-1B’s, right before the layoff announcements hit in November 2025.)

Independent researcher Vsevolod Shabad has a great short commentary on what’s happening at Amazon, where he worries that the layoffs in combination with rapid AI adoption have created an “organizational amnesia” that it will take a long time to overcome.

Professional judgment is not an innate trait; it is forged through the cognitive friction of entry-level analysis that Amazon is now automating away.

By treating junior coding as a mere inefficiency to be optimised, the company is hollowing out the institutional memory required to recognise a “high blast radius” before it crystallises.

To my mind, there are two options here: either Amazon continues down its current path and risks major incidents that endanger its entire operation, or else it quietly recognizes an excessive enthusiasm for AI adoption and staffs up again. Jassy’s team would really have to be disconnected from the reality on the ground to do the former (very possible, of course!). Meanwhile, the latter, despite all the doom and gloom in the white-collar world, presents quite an organizing opportunity: an existing workforce of skilled workers disillusioned with the company for its callous irresponsibility and newly prized for their institutional knowledge. That seems like a good situation in which to organize!

Either way, the present moment is a “prime” one for Amazon corporate workers to take the lead in defining the parameters of responsible AI deployment, one of the many reasons why white-collar AI doomerism is so pernicious.

This is a newsletter about labor & logistics, and in its service, I have done my best to meld my mind with Claude’s Excel extension, but still a remainder of my prior subjectivity remains! In that vein, I wanted to let everyone here know about a new book that Paul Prescod and I have just released: a co-edited volume called Rustin’s Challenge, which collects together some previously unpublished or underappreciated writings of the civil rights organizer and socialist strategist Bayard Rustin, and pairs them with commentaries from contemporary thinkers on the Left.

Book Order Page

You can order the book here, OR if you’re in or around New York City, you can attend our launch event at Verso Books in Manhattan on Thursday, March 26th at 7pm.

Book Launch RSVP

If you want to get a sense of the book, my editorial intro is here, and Dustin Guastella’s remarks at our Philly launch event are here.

Thanks for subscribing to On the Seams. Now back to the regular programming.

In Jacobin yesterday I laid out the case for Amazon being the key target that labor should have its sights set on, and for now being the moment when institutional labor should loosen the purse strings on an all-out effort to organize the company.

Some highlights:

• In two or three years, unless there are major operational hiccups, Amazon is set to be the first company in history to record $1 trillion of revenue annually. It is today simultaneously the second largest retailer by revenue, the largest private parcel company by volume, and the largest cloud computing company by revenue.

• The fight to organize Amazon is an existential one not just for the Teamsters but for the labor movement as a whole. I don’t mean that as a platitude; I mean that it’s very difficult to imagine labor exiting its current death spiral unless unions as a whole commit to tackling this massive and metastasizing problem. Either Amazon is organized, or it will be the weight on labor’s chest as it struggles for its last breath.

• A real organizing effort at Amazon is not something that can be scraped together from spare parts. It requires a cohesive vision, unified execution, diligent and capable organizers, and money — lots and lots of money. If we are not honest about this fact, then we are playing to lose.

• Labor is about to spend about half a billion dollars on a midterm election that a) it probably doesn’t need to influence, given the political winds, and b) that it bears little influence to impact anyway, given the severely diminished weight of its political expenditures. How else could half a billion dollars be spent?

Read the full article here.

Organizing efforts to date have targeted Delivery Stations, both warehouse associates and DSP drivers, for understandable reasons: the last-mile of a package's journey is by far its most expensive leg, and it's the one most resistant to automation efforts. Amazon has been trotting out drone delivery pilots, humanoid robot delivery assistants, and other last-mile tech fixes that gets the business press salivating, but I don't think we are anywhere near the implementation of anything that would substantially change up their last-mile strategy, i.e., paying human beings as little as possible to deliver their packages. The large warehouse automation efforts are a different story.

Delivery Stations are also in or close to urban areas, and you tend to get more organizing interest there. In addition, there are already organized delivery driver workforces at USPS and UPS, and so it's possible to leverage that density. Combine all this with the NLRB's joint employer decisions, essentially saying that Amazon so structures the work of DSP employees that they can be considered their “joint employer,” and focusing on Delivery Stations makes a lot of sense.

Unfortunately for the last-mile strategy in the United States, there is a large, precarious gig workforce of Flex drivers that also deliver packages. Flex drivers work by the block: an average one might be for 3.5 hours, during which time a driver might be expected to deliver about 40 packages from their own vehicle. Recently they’ve started doing express and rapid deliveries, and this has resulted in the proliferation of one-hour blocks, during which time a driver will deliver only 1-2 packages. Flex drivers sign up for these blocks on the Flex app; think Uber for packages.

Flex is a problem for organizing Amazon Delivery Stations for a number of reasons: if you're strategic and capable enough to get to the point of disrupting Amazon Logistics, they'll substantially increase the number of Flex blocks, in addition to just dumping more packages into USPS and UPS streams. If Amazon can quickly up their last-mile capacity outside of the DSP network by offering more and better compensated Flex blocks, it's pretty hard to materially impact the company through Delivery Station disruption.

For its importance to the company, however, it's very difficult to get a sense of just how big the Flex workforce is. Amazon claims that 2.9 million people have downloaded the app, but that certainly doesn't tell you how many people are actually driving for Flex.

From what I can tell, Flex drivers will pick up from one of five nodes in Amazon's distribution network: Delivery Stations, Sub-Same Day Fulfillment Centers (SSDs), Grocery Fulfillment Centers (GFCs), Whole Foods, and Rural-Super Rural (RSR) facilities. To my knowledge, Flex drivers exclusively handle deliveries at SSDs, GFCs, and Whole Foods. Both DSPs and Flex drivers serve Delivery Stations and RSRs.

I couldn’t come up with a better method for estimating the size of the Flex workforce than going to some of these facilities, counting cars, and then generalizing based on what we know of Amazon’s distribution network. As an example, there are 79 Sub-Same Day (SSD) Fulfillment Centers in the United States. They're on average 189,000 square feet in size, and employ on average about 400 warehouse associates. Flex blocks at SSDs typically start in the 3am range, and the last ones start around 8pm, meaning Flex Drivers arrive at the facility throughout the day for about 17 hours, but clustered around CPTs, or “critical pull times.”

If I visited a representative SSD (but stayed off property) and got an average hourly inbound Flex car count, I could multiply that by 17 hours to get an average daily count, then by 7 to get an average weekly count. Flex drivers are capped at 40 hours a week, so at 3.5 hours, that’s a max of 11 blocks/driver. So if I then divided the weekly car count by 11, I’d get the rough number of full-time Flex drivers operating out of that SSD. Flex drivers tend not to work that many blocks (from Reddit, it seems like somewhere in the 20-25 hour range is about average), but I figured it made sense to estimate full-time (or 11 block) Flex drivers, not knowing how exactly to account for gig work variability and wanting a unit comparable to the DSP workforce. Multiply that by 79, and you get some sense of how big the Flex workforce operating out of SSDs is. Now repeat for Delivery Stations, Grocery Fulfillment Centers, Whole Foods, and RSRs - making sure always that the Flex driver delivery entrance/pick-up is differentiated from warehouse associate parking lots and any other spots in the facility where non-Flex driver cars enter.

Here’s what I found.

I observed two SSDs, each for about 7 hours. At one that was a bit above the average employee count, there were 100 inbound Flex cars per hour. At one that was an equal amount below the average employee count, there were 69. Averaging the two, that’s 84.5 cars/hour inbound at an average SSD. Doing the calculations above, you get: 84.5 Flex cars/hour x 17 hours/day x 7 days/week = 10055.5 cars/week. Divided by 11 blocks = 914.14 full-time Flex drivers per average SSD. And times 79 makes about 72,217 full-time Flex drivers servicing SSDs. (*Important Note: I did these observations before Amazon adopted express and rapid delivery options, so all of the cars that came on those days were for normal blocks.)

Delivery Stations are trickier, as there appears to be a bit of variety in how Flex operates in that world. For the most part, Delivery Stations will be all DSP pick-ups in the mornings, there will be a bit of a lull around lunchtime, and then Flex drivers will start to trickle in in the afternoon. At one Delivery Station that I observed, which was built recently and so has no OSHA ITA headcount data associated with it, there were 25.4 outbound Flex drivers/hour from 1pm-8pm. Before 1pm, it was all DSP vans, and after 8pm everything slowed to a trickle. 25.4 cars/hour x 7 hours/day x 7 days/week = 1244.6 cars/week. Divided by 11 blocks = 113.1 full-time Flex drivers per average Delivery Station (provided we’re taking that Delivery Station as representative). And times 479 Delivery Stations = 57,175 full-time Flex drivers servicing DSs.

However, after doing this observation, I realized there was an easier way to calculate Flex service provision at both Delivery Stations and RSR facilities. An anonymous source reported to me that the standalone SSDs handle about 2/3 of the same-day volume (the volume that Flex drivers handle), while Delivery Stations and RSR’s handle the other third. If that’s the case, then we can just halve the SSD estimate, and there would be 36,109 Flex drivers servicing both Delivery Stations and RSR facilities. That’s significantly less than the estimate I got from car counting, but I’m pretty sure the Delivery Station I observed had a particularly beefy operation and is not representative (also, in retrospect I realized that my particular vantage point at the Delivery Station did not allow me to distinguish Flex drivers from warehouse associates leaving). In any event, I’ll take my longer sampling of SSDs to be more more accurate and run with the Delivery Station/RSR estimate based upon that.

So now without counting GFCs/Whole Foods, we’ve got an estimated 108,326 full-time Flex drivers. Just food delivery left to count!

Both GFC and Whole Foods blocks are shorter, usually 2 hours or less rather than the 3+ that are typical at SSDs. This would make sense, as there are delivery window restrictions for perishables. I observed a GFC for three hours, and saw 12 Flex cars on average outbound. This was at a GFC that was a bit smaller than the typical GFC, employing 80% of the average employee count at GFCs (again, all according to OSHA data). So let’s say then that the average GFC is seeing 15 cars/hour. GFC drop off window start times usually go from 7am-8pm, so let’s say there are Flex drivers arriving continuously between 6am and 8pm. 15 cars/hour x 14 hours/day x 7 days/week = 1,470 cars/week. A full-time driver could do 20 shifts/week, so divided by 20 =73.5 drivers. Times 82 GFCs and that’s 6,027 Flex drivers working out of GFCs.

For the moment, I’m going to be lazy because I have sat on the street outside of facilities for a physically uncomfortable amount of time now, so I’ll say that each Whole Foods is seeing about half the volume as a GFC (without a dedicated fulfillment area, I’m not sure how they could do more than that). So 36.75 drivers x 550 Whole Foods = 20,213 Flex drivers working out of Whole Foods. Let’s just use that as a placeholder for now. And we arrive at the grand total of 134,556 Flex drivers in the United States (72,217 SSD, 36,109 DS/RSR, 6,027 GFC, 20,213 Whole Foods).

That’s again an estimate of full-time equivalent Flex drivers. If we take most Flex drivers to be driving an average of 25 hours a week rather than the limit of 40, then we might estimate that roughly 215,306 people are actually driving for Flex currently.

Whew! I’m going back to staring at spreadsheets now. But should I gain the courage to go back to the world of car counting, or should some noble insider have better information on which to base some of these estimates, I will update here.

As I covered in my recent Amazon annual report recap, CEO Andy Jassy was talking up the growth in sales of “everyday essentials” and grocery, two low-margin categories that seem quite unsexy compared to the AWS numbers. Amazon cares about these categories because they are gateways to further purchases: if you’re buying everyday essentials or groceries on Amazon, you’re probably buying much else besides. SSDs are crucial for the delivery of both.

They also uniquely operate with Flex delivery drivers, i.e., gig workers who deliver packages from their own vehicles. As I show in the article, this is not simply about degrading labor standards but also about the continuous flow model that SSD’s embody, which is better serviced through small-batch delivery than the larger batches that go out with Amazon DSP delivery vans. It just so happens that the operational logic of the most sophisticated same-day operation in human history fits almost perfectly with the gigification of last-mile delivery.

Read more here.