What’s trending in AI on 12 October 2026: Walmart, the retailer that has spent years selling investors on a robot-powered supply chain, is learning how hard it is to automate a business that already exists. A Wall Street Journal investigation, summarized by Quartz and Supermarket News, describes a programme to automate roughly 200 US supply-chain buildings that has run into cardboard boxes too big for the machines, wheeled robots that break down often enough to disrupt operations, dust that can trigger cascading stoppages and electricity bills at automated grocery warehouses that can top $800,000 a month at peak, against about $250,000 for mostly manual sites. Walmart’s own head of US supply-chain operations described the company as being in “peak complexity.” Yet the same week Walmart opened its fifth next-generation fulfillment center and said it is not slowing down. Below: what went wrong, why Walmart is pressing on anyway, what it teaches any business buying AI or automation, and a scored readiness test you can run on your own project this week.
Key takeaways
- The demo is not the deployment. Walmart’s robots work; the problems come from old buildings, legacy packaging, dust, mixed vendors and the need to keep sites running during the switch.
- Running costs can dwarf the sticker price. Reported power bills of up to $800,000 a month and pricier specialist engineers change the payback maths.
- Reliability compounds. Symbotic says under 10% of its robots were in repair at any time last year and about 5% now; in a tightly coupled system, one failure can stop many.
- Even giants write off bets. Walmart reportedly sold its in-store robotics unit for about half what it paid, after a 20-store rollout proved too inefficient.
- Competition sets the clock. With Amazon ahead, Walmart’s leadership argues waiting for perfect would take too long. Your project needs the same explicit trade-off.
1. What the Wall Street Journal found
Walmart employs more than 140,000 warehouse workers who unload trucks, sort goods and send orders out, and a single worker can walk around 20,000 steps in a shift. Replacing much of that work with machines has been a strategic goal for years. According to the Journal’s reporting, as relayed by Quartz on 9 October, the company is now overseeing the transition across most of its roughly 200 US supply-chain buildings, and the problems have piled up. Some early robot trials failed outright. Others worked but cost too much to scale.
The specific failures are revealing because they are so ordinary. Walmart’s long-standing cardboard cases were too large for the automated systems, so it moved to shorter, taped boxes, which added cost. Small wheeled robots that move merchandise have broken down often enough to disrupt operations. Bulky items, such as giant bags of dog food, still need people. And at Symbotic, Walmart’s main partner for non-perishable goods, dust build-up can cause robot faults that ripple into wider stoppages. Supermarket News adds that Walmart often retrofits automation into warehouses built in the 1980s and 1990s, using different robotic systems from different vendors for different tasks at different sites. Walmart did not respond to Supermarket News’ request for comment.
None of this means the robots are fake or the strategy is wrong. It means the gap between a technology that works in a controlled setting and one that works inside a live, decades-old operation is large, expensive and mostly invisible in vendor demos. We saw a similar gap with humanoid robots in our look at the $18,000 humanoid with a human inside.
2. Why “breakpack” and old buildings are so hard
One of the toughest jobs to automate is what the industry calls breakpack: opening cardboard cases and pulling out individual items so they can be shipped one by one. Every product is a different size, weight and shape, and packaging varies by supplier. After years of experimenting, Quartz reports, Walmart and its robotics partners settled on small wheeled robots with motion and obstacle sensors, now being installed in more than a dozen warehouses.
The deeper challenge is doing all this without closing the buildings. Walmart has to redesign warehouses while they keep shipping goods to thousands of stores. That is the physical-world version of replacing an aircraft’s engines mid-flight, and it is the same problem software teams face when they drop AI agents into live business processes rather than greenfield ones. The table below translates Walmart’s warehouse problems into the questions any automation or AI buyer should ask.
| Walmart problem (as reported) | What it really reveals | The question to ask on your project |
|---|---|---|
| Cardboard cases too big for the machines | Inputs were designed for humans, not systems | What data, formats or documents must change before the tool can work? |
| Wheeled robots break down often enough to disrupt work | Reliability at scale differs from reliability in a pilot | What is the failure rate at full volume, and what happens downstream when it fails? |
| Dust triggers cascading stoppages | Tightly coupled systems amplify small faults | Is there a manual fallback, and can one failure halt everything? |
| Retrofitting 1980s and 1990s buildings | Legacy environments limit what is possible | Are we automating a process that should be redesigned first? |
| Different robots from different vendors at different sites | Integration and support costs multiply | How many vendors, contracts and integrations will we own in three years? |
| Oversized items still handled by people | The long tail of exceptions never goes away | Who handles the 10% to 20% of cases the system cannot? |
| Power bills up to $800,000 a month; pricier engineers | Operating cost was underweighted against capital cost | What is the five-year total cost of ownership, including energy and skills? |
3. The money: reliability, energy, skills and a costly reversal
Reliability is the number that quietly decides whether automation pays. A Symbotic spokesman told the Journal that fewer than 10% of its robots were under repair at any given time last year, and that the figure has since dropped to about 5%. That is real progress, but in a warehouse where robots depend on each other, even a small share out of action can slow an entire line, and the reported dust problem shows how one fault can spread. Symbotic, based in Wilmington, Massachusetts, has worked with Walmart since 2017, and Walmart holds a 12.6% stake in the company.
The second number is operating cost. Automated grocery warehouses can draw so much power that monthly bills exceed $800,000 at peak, roughly three times a mostly manual site, according to people familiar with the matter cited by the Journal. The machines also need specialist engineers who earn far more than conventional warehouse staff. Energy is becoming a hidden line item across AI generally, as we covered in our piece on AI’s energy crunch and power-bill politics.
The third is the cost of bets that do not work. Walmart has invested nearly $1 billion since 2016 in automating online-order fulfillment inside its stores. Quartz reports that it bought robotics firm Alert Innovation for $400 million and later sold it to Symbotic for $200 million, a loss. Supermarket News says the earlier in-store system reached 20 stores before proving “too inefficient.” A new version is now being tested in a Dallas store, though reports differ on timing: Quartz says Symbotic does not expect a workable model until 2028, while Supermarket News reports Symbotic telling investors its new SymMicro system should be running in about six months. Either way, the in-store bet is still unproven a decade in. For more on why pilots so often fail to pay off, see our guide to AI cost overruns.
4. Why Walmart is pressing on anyway
The same week the Journal’s story ran, Walmart opened its fifth next-generation fulfillment center, a 900,000-square-foot, highly automated site in Stockton, California, that it says will create more than 1,000 jobs. It has also announced a sixth, a planned 1.5-million-square-foot center in Carnesville, Georgia, joining sites in Indiana, Texas, Pennsylvania and Illinois. Walmart says the new design cuts order fulfillment from 12 steps to five.
Competition explains the urgency. Amazon leads in warehouse automation and has overtaken Walmart as the largest US company by annual revenue, Quartz notes. Walmart expects automation spending to peak this year and next, and its supply-chain chief told the Journal that leadership and the board see automation as a long-term priority, arguing that waiting for perfection would take too long. That is a legitimate strategic choice. The lesson is not “don’t automate”; it is that a company with Walmart’s scale and budget still has to absorb years of friction, and smaller businesses need to plan for their own version of it.
5. What this means for businesses buying AI and automation
It is tempting to file this under “logistics” and move on, but Walmart’s warehouse story is the physical mirror of what many companies are discovering with AI agents and software automation. The tools are capable. The friction lives in the environment they are dropped into: messy inputs, legacy systems, exception cases, hidden running costs and the need to keep the business operating during the change.
Three parallels stand out. First, inputs matter more than the engine. Walmart had to change its boxes; most AI projects have to change their data, documents and process definitions before the model can help. Second, the bottleneck moves rather than disappears. This morning’s research round-up on AI agents that overstate their own results showed software output rising while finished work stayed flat, because review became the choke point, much like Walmart’s exceptions still landing on people. Third, individual gains do not automatically become system gains, the theme of our analysis of why a 10x individual boost shrinks to 1.8x at team level.
There is also a resilience and security angle that operations leaders should not ignore. Highly automated sites concentrate risk: when tightly coupled systems depend on a few vendors, networks and software platforms, a single fault, a supplier outage or a cyber incident can halt throughput that a manual site would have absorbed. Plan manual fallbacks, test them, and treat robotics and automation platforms as critical systems with the same patching, access control and vendor-risk review you would give any core application. Our overview of agentic AI versus traditional automation covers how those controls translate to software agents.
6. The automation readiness scorecard
Score your next automation or AI project against these ten questions, drawn directly from the problems Walmart hit. Give each 0 (no answer), 1 (partial) or 2 (documented and tested). A total of 16 to 20 means you are ready to scale; 10 to 15 means run a longer pilot and fix the gaps; below 10 means you are buying a demo, not a deployment.
- Inputs. Have we listed every format, document or physical input the system needs, and who will change those that do not fit?
- Exceptions. Do we know what share of cases the system cannot handle, and who handles them, at what cost?
- Reliability at volume. Do we have failure-rate data from production-scale use, not just a pilot?
- Coupling. If one component fails, does the whole process stop? Is there a tested manual fallback?
- Running costs. Does the business case include five years of energy, compute, licences, maintenance and specialist staff?
- Skills. Have we budgeted for the engineers or operators who will keep it running, and can we hire or train them?
- Migration. Can we switch over without stopping the business, and have we planned the transition period’s lower output?
- Vendor sprawl. How many vendors and integrations will we own, and who is accountable when they disagree?
- Exit plan. If it underdelivers, what can we reuse, resell or write off, and at what point do we decide?
- Security and resilience. Is the system covered by patching, access control, monitoring, incident response and vendor-risk review like any critical application?
If your total is low because of questions 5 and 6, revisit the numbers with the method in our AI automation ROI guide before you sign.
7. What to watch next
Three signals will show whether Walmart’s bet is turning. Watch Symbotic’s results for robot repair rates and progress on the Dallas in-store test, where reports already disagree on timing. Watch how quickly the Stockton and Carnesville next-generation centers ramp, since purpose-built sites should show the gains retrofits struggle to deliver. And watch Walmart’s commentary on automation spending as it peaks this year and next: if costs fall and throughput rises in 2027, the friction will look like a necessary toll. If not, expect harder questions from investors, and a useful warning for everyone else rushing to automate.
Frequently asked questions
What problems is Walmart having with warehouse robots?
According to a Wall Street Journal report summarized by Quartz and Supermarket News in October 2026, Walmart’s cardboard cases were too large for automated systems, small wheeled robots broke down often enough to disrupt work, dust caused cascading stoppages, oversized items still need manual handling, and automated sites cost far more to power and staff.
How much does it cost to power an automated Walmart warehouse?
People familiar with the matter told the Journal that power bills at automated grocery warehouses can exceed $800,000 a month at peak, compared with about $250,000 for predominantly manual operations, roughly three times as much.
What is Walmart’s relationship with Symbotic?
Symbotic, based in Wilmington, Massachusetts, is Walmart’s main automation partner for non-perishable goods and has worked with the retailer since 2017. Walmart holds a 12.6% stake. Symbotic says fewer than 10% of its robots were under repair at any time last year and about 5% now.
Is Walmart scaling back automation?
No. Walmart opened its fifth next-generation fulfillment center in Stockton, California, this month, has a sixth planned in Georgia, and expects automation spending to peak this year and next. Its leadership says waiting for a perfect system would take too long, especially with Amazon ahead.
What can small and mid-sized businesses learn from Walmart’s robot problems?
Budget for the hidden layers: changing inputs, handling exceptions, running costs such as energy and specialist staff, migration downtime and vendor integration. Test reliability at real volume, keep a manual fallback, and set a clear exit point before you commit.
Does this apply to AI software, not just robots?
Yes. AI agents and automation software face the same pattern: capable tools slowed by messy data, legacy systems, exceptions that land on people and running costs such as inference and review time. The same readiness questions apply whether the automation is physical or digital.
Bottom line: Walmart’s robots are not failing; the world they were dropped into is fighting back. Boxes, dust, old buildings, power bills and exceptions are where automation business cases are won or lost, for warehouse robots and AI agents alike. Plan for that friction up front and automation can still pay. Ignore it and even a company of Walmart’s size ends up writing cheques it did not budget for.