As AI Assistants Fight Over Shoppers, Infrastructure Gets Paid Either Way.

The part of the story that matters here came a decade later. McLean owned the patents on the corner fittings that let the box stack and lock onto trucks, trains and ships. He could’ve licensed them and collected a toll on every container that moved. Instead, he made them available royalty-free so the whole industry could build to one spec. He gave up control of the box and got the world’s ports in return. Sea-Land, the company he built, didn’t need to own the ships, the cranes or the cargo. It needed to be the thing every shipment ran through.

That’s the business model I keep coming back to as the agentic commerce land-grab picks up speed.

The expensive assumption running through most of it is that winning the sale means owning the place where the consumer starts shopping. Build the assistant. Attract the audience. Persuade merchants to connect to yet another way to buy. It’s a big bet, and most of the companies making it will lose.

There’s a second model that gets a lot less attention, even though it’s fundamentally the same one the payments industry has been running for nearly six decades. Build the thing those assistants and merchants need to finish the sale. Make it easy to plug in. Get paid when they use it. The consumer can change where she starts shopping a dozen times without changing whose infrastructure makes the purchase actually happen.

Read More: Why Building AI Agents Is No Longer the Hardest Part of Agentic Commerce

Amazon’s fulfillment move last week is the most visible example of the second model, which is why it’s worth walking through. But the model is the story. Amazon just happens to be this week’s example. Plenty of companies can play this game. A few are already trying. Most haven’t figured out what it takes to win it.

The Best Way to Grow an Infrastructure Business Is to Ask Customers to Give Up Less

On Sept. 24, Amazon announced that U.S. merchants using its Multichannel Fulfillment service can put the Prime badge and fast, free Prime delivery on their own websites for nothing beyond standard fulfillment fees. Merchants keep their payment processor, order management, returns and customer service. Prime membership gets verified after checkout. The shopper never has to sign into Amazon.

It builds on the May launch of Amazon Supply Chain Services, which opened more of the company’s freight, fulfillment and parcel network to other businesses, and sweetens the deal with discounts of 15% to 25% for sellers who route more volume through it.

What they gave away was checkout. Until now, a merchant that wanted the Prime promise on its own site had to take Amazon’s checkout with it. That’s a lot of systems and relationships to rip up for a delivery badge, all while trying to overcome the fox in the henhouse mythos at the same time. Drop that requirement, and the merchant has a lot less to reconsider. A shopper can find a product through any AI assistant, buy it on the brand’s site, pay through the brand’s processor, and the box ships via Amazon. Fulfillment fee in hand.

Read More: AI Can Pick the Shoes but Amazon Still Moves the Box

That’s more or less the McLean move in commerce. Make a capability useful across more transactions by attaching fewer conditions to it. Give up some control over the experience in exchange for the chance to serve more of the market. And monetize the bit that is oh-so-hard and takes oh-so-long to build.

Amazon has run this play before. It opened S3 and EC2 in 2006 and turned the computing it built for itself into a service anyone could buy. That market grew far past what its retail business needed. A fulfillment network built for one retailer can serve brands, marketplaces and orders that start in channels the retailer doesn’t control, as long as the revenue covers the cost of the added capacity. A warehouse full of low-margin orders isn’t automatically a better business.

For Amazon, it’s also a hedge. A company that earns only when consumers choose its storefront has one path to the transaction. A company whose infrastructure serves many storefronts has more. Amazon isn’t indifferent to where the sale happens, and it still has more than half of all online commerce running through its marketplace. And certainly, a fulfillment fee doesn’t replace the marketplace commission, the ad revenue and the shopping data that come with an Amazon.com order. But it’s revenue on orders that would otherwise have belonged to someone else entirely. And it has become all but a no-brainer for merchants to sign onto.

The Card Networks Have Been Running This Play for Fifty Years

If the model sounds familiar, it should. It’s the business the card networks built, and it’s why they’re one of the most valuable players in commerce.

Start with what a network doesn’t do. It doesn’t issue the card. It doesn’t lend the money or hold the deposit. It doesn’t own the consumer, who belongs to her bank, or the merchant, who belongs to its acquirer. It doesn’t run the checkout, in a store or online. Its brand rides in the corner of someone else’s card and inside someone else’s button.

What it does do is set the rules everyone builds to and guarantee the outcome. The merchant knows it’ll get paid. The consumer knows she’s covered if the goods don’t show up or the card gets stolen. The bank knows what it’s owed and when. The network runs the switch that connects thousands of banks to millions of merchants, owns the bad day through chargebacks and zero liability, and takes a few basis points on every transaction that rides its rails. The rails get built once, at enormous cost, and the next transaction costs almost nothing to carry. That’s how you end up with operating margins most businesses would assume were a typo. But would give anything to have.

It’s a great business, which is why everyone has been trying to displace it for twenty years. Pay by bank, real-time payments, wallets, BNPL, stablecoins. Every one of them goes after the fee. None of them has cracked the reason merchants keep paying it. Certainty, everywhere, every time, with somebody standing behind the transaction when it goes wrong. The networks gave up the front end and kept the guarantee. That’s what keeps the consumer coming back and back and back.

Read More: Who Needs Whom? Card Networks, Stablecoins and the Future of Payments

Moving a Box Is a Lot Harder Than Moving a Payment

Here’s the thing. A payment is a message. Authorization, clearing and settlement are bits moving between computers in milliseconds, and the thing being guaranteed is a promise about money that already sits in an account somewhere. The networks built the rules and the switch, and the promise scales.

A box is something very different. The right item has to be in the right building. Somebody has to pick it, pack it and hand it to a carrier. Weather, wrong addresses, damaged goods, porch pirates and returns all show up in the numbers. Every piece has to work every time, and the failures are physical, visible and remembered. Certainty about money took the networks decades of rules, disputes and liability shifts to build. Certainty about delivery takes warehouses, trucks and a lot of people, and there’s no shortcut through any of it. That’s why payments has a handful of viable providers at every layer and fulfillment has one that can attach a membership promise to the box.

Retail learned this the hard way once already. In 1990 Kmart had more stores than Walmart, a national name Walmart didn’t yet have and a bigger head start. Walmart had something else. Distribution centers within a day’s drive of nearly every store, a fleet of its own trucks and a private satellite network that told each warehouse what every register had sold that day. Walmart passed Kmart in sales that year and never looked back. Kmart, which underinvested in the logistics technology that kept Walmart’s shelves stocked, filed for bankruptcy a dozen years later.

Kmart had the stores and couldn’t move the box. The agents have the conversation and can’t move the box either. The reward the consumer is chasing was never the search box or the checkout. It’s the box on the front porch, on the day she was told it would be there.

Read More: Why AI Shopping Is Still Just a Smarter Search Bar

Logistics, as I argued in my January 2026 trends piece, is the story of retail in an agentic world.

Consumers Have Moved the Search and Left the Purchase Where it Was

Our research explains why finishing the sale is where the money is.

The August PYMNTS Intelligence study found that 61 million U.S. adults, nearly one in four, now start retail product research with AI. Fifty-nine percent of AI shoppers do their research on AI where they used to do it on Amazon.

Then they buy at Amazon anyway. Six in ten AI-assisted purchases land there. Big-box and department stores get 20%. Brand sites get 6%, down from 10% before AI shopping showed up. That’s from our August study of 2,191 U.S. consumers.

The same study tells you what it would take to move them. Marketplace shoppers say they’d buy direct for a lower price (54%), shipping as fast as the marketplace (50%) and returns as easy (35%). And here’s the humbling part. Give them identical products, prices, shipping and returns, and 40% still want their agent to buy from the marketplace.

None of that proves Amazon is the only company that can deliver what consumers want. It shows how much work is left after the assistant finds the right product. The merchant has to make the purchase feel like a sure thing, including what happens when something goes wrong. Recommendation is the easy part. Certainty and trust is the product.

Read More: Agentic Commerce Needs Its Febreze Moment

As I explained in my 2026 trends piece, stores become execution points as discovery and decision-making move to agents, and agents continuously test fulfillment performance. A merchant’s ability to deliver becomes part of its ability to compete for the sale. A brand can build that capability, buy it or stitch together some of each.

Infrastructure businesses exist to make the buying option the obvious one.

The Will of the Wallets, or Not

Our latest mobile wallet study, the twelfth in the series, fielded in late August and early September with 2,458 U.S. consumers, is a useful reminder of the difference between the experience a consumer sees and the plumbing a transaction needs to execute.

In stores, wallets are having a moment. Roughly 18% of in-store shoppers used one in the seven days covered by the study, a record. Apple Pay accounts for more than half the category’s growth since 2021. Tapping a phone at the register has become a habit.

Online is a different story. About 14% of shoppers used a wallet in that same week, but only 9% used one for their last online purchase, unchanged from 2024. Apple Pay was 3% of last online purchases, PayPal 2%, Google Pay 1%. People try a wallet online and like it fine. They just don’t make it the default, because the merchant already has a card on file and the consumer already has a way to buy that works.

The reasons for using a wallet online are shifting, too. Since 2023, “fast” has fallen nine points as the top reason. “Easier to track purchases” is up seven. “Better security” is up four. Speed still matters, but the value consumers describe now lives around the payment and after it: the record, the protection, the fix when something breaks.

That’s a map of which problems still need solving, and a warning against measuring an infrastructure business by the share of a visible button. The right measure is the transactions a company enables, the revenue it earns on them and the work it takes off its customers’ hands. Which brings us to the companies trying to do exactly that.

A Connection Can Be Live and Still Not Matter

Fulfillment makes the infrastructure argument easy to see because a box eventually has to show up. But agentic commerce depends on a chain of things going right before the box: accurate product information, available inventory, a merchant that recognizes the agent’s authority, a payment that’s accepted, an order that’s recorded, a problem that gets resolved. One broken link undoes everything before it. Every link is a place where a company can become necessary across many merchants and many agents.

PayPal wants that role. Agent Ready and Store Sync bundle merchant connectivity, catalog distribution and payments. PayPal owns the money’s bad day, with 439 million accounts, buyer protection and a dispute process. It still can’t promise the box. PayPal has to give merchants a reason to expect more completed sales and consumers a reason to change how they already buy. Our wallet data makes that a hard sell. PayPal is about 2% of last online purchases. That figure doesn’t capture its processing business or its agentic services, but it puts the consumer side of the challenge in perspective. Connecting a catalog to an assistant doesn’t create demand for the catalog. Adding a payment option doesn’t make shoppers pick it. I’d want to see repeat purchases and real merchant volume before calling this a business with momentum.

I feel the same way about Muse. Browsing and buying is a capability. Getting the consumer to trust it with the next purchase is the business. Our AI shopping data shows how easily consumers move the conversation while leaving the transaction somewhere familiar. Until there’s evidence of a repeat shopping habit, Muse doesn’t get much weight from me in the race to shape commerce.

We’d love to be your preferred source for news.

Please add us to your preferred sources list so our news, data and interviews show up in your feed. Thanks!

Add as Preferred Source

Read More: Is Muse Really the Answer to Meta’s AI Woes?

The point across both is the same. A layer can be essential and competitive at the same time. Being essential creates demand. Differentiation and the availability of substitutes decide what a provider gets to charge. And a connection can be technically live and commercially irrelevant. The infrastructure opportunity is real, but somebody has to use what you built, get value from it and come back.

Shopify Learned That the Button Was Never the Business

Shopify is where the infrastructure argument gets real for merchants, because that’s where Amazon started. At launch the program runs through the Amazon MCF and Buy with Prime app for Shopify, with an API and other integrations promised for later. So for now this is a Shopify story, which makes it a direct-to-consumer story. The DTC brands that live on ChatGPT and Gemini referrals, that built their businesses on owning the customer and skipping the marketplace, are exactly the merchants whose checkout Amazon just left alone and whose inventory it now wants.

Think about the choice from the brand’s side. According to PYMNTS Intelligence data, brand sites get 6% of AI-assisted purchases today, down from 10%. Half of marketplace shoppers say they’d buy direct if shipping were as fast, a third if returns were as easy. An agent asked to get something to a consumer by Friday, free, will favor the merchant that can promise it, and the Prime badge is the only promise an agent can trust without checking. A DTC brand on Shopify can keep running its own logistics and lose that comparison to a Prime-badged competitor, or it can put inventory in Amazon’s building and convert like Amazon does. That’s a choice for a little while. The brands that spent a decade defining themselves as the alternative to Amazon will end up with more of their sales as Amazon-fulfilled storefronts with their own logo on the box. Mostly because so many of them already sell there.

Follow that a few turns and it doesn’t stop at fulfillment. Once the inventory sits in Amazon’s building and the badge sits on the brand’s page, the brand is running an Amazon storefront with extra steps. It’s paying Shopify for the checkout, paying Google or OpenAI for the traffic and paying Amazon for the only part the consumer cares about. At some point the CFO asks why the brand isn’t selling more on Amazon.com, where the same inventory converts at nearly 10% against Google Shopping’s 2% and the delivery promise is native rather than borrowed. Since they are already there anyway. With Amazon’s promise heading toward same-day, that question could get harder to answer every year. The end state of “delivered by Amazon” is “lives on Amazon,” with the brand’s own site as the place the agent sends people who asked for it by name.

Shopify is the cautionary tale inside that story. It spent a decade and a lot of ambition trying to become the place consumers shop, the anti-Amazon storefront, and built the Shop app to prove it. Consumers never showed up there to browse. What they used was Shop Pay, one of the best checkout experiences on the web and exactly that, a button. Shopify became the payments layer for a million independent stores, and now Amazon has made those stores’ delivery layer its own, through a Shopify app, without asking Shopify for anything. The button was amazing. But in an agentic world, the button isn’t the business.

Read More: What Happens to Stores When AI Agents Do the Shopping?

Walmart shows how far the infrastructure proposition reaches. Since May 2025 Walmart has let its marketplace sellers fill Walmart.com orders out of Amazon’s warehouses through the same MCF service, as long as the box is plain and Amazon’s own vans don’t make the last-mile delivery. If you’re wondering why there are so many white unmarked vans delivering boxes marked Amazon on the shipping label, now you know why. If Amazon can fulfill for Walmart, the list of who it can’t fulfill for gets short. Necessary to everyone, visible to no one. That’s the infrastructure position.

The Handoffs Are Where the Money Is

Ask an agent to find a particular coffee maker under $100 and get it here by Friday. Simple request. The consumer described an outcome. Completing it takes several businesses agreeing on the same facts and carrying them intact from one system to the next.

The price has to hold. The inventory has to be there when the order is placed. Shipping can’t push the total past the budget. The delivery estimate has to be true for that address. The merchant needs to know what the agent was allowed to do. And if the coffee maker sells out, somebody needs a rule about substitutes.

Read More: Department Stores of the Future Are AI Agents

A good recommendation solves one piece of that. The handoffs decide whether the rest survives contact with the transaction. An agent that finds the perfect product and loses the promised price or delivery date before checkout has just made more work for the shopper.

This is where becoming infrastructure creates a market bigger than any one company’s audience. A provider that helps merchants keep inventory data accurate, validate an agent’s spending authority or preserve an order’s terms across systems can sell that to every competing agent at once. Its value is that the handoff works every time. And delivery reliability you can take to the bank.

Merchants Still Have to Do the Math

The merchant’s side of this gets less attention than it deserves. Keeping their processor and checkout removes one kind of friction. Putting their inventory in someone else’s warehouse creates a different set of commitments. There are the fulfillment costs, inventory placement, operating requirements and what it would cost to switch later. Volume discounts improve today’s economics and deepen tomorrow’s dependence at the same time, and every order that arrives on time deepens the shopper’s relationship with the company that delivered it at least as much as with the brand that sold it.

Read More: The Value of Consumer Certainty in Driving Digital Transformation

Providers want customers to rely on them. Customers want reliable service and the freedom to switch, negotiate and stand out. The winning proposition delivers enough ongoing value to make the dependence worth it, and it shows up in the order economics. The conversion, cost to serve, repeat purchases and the ability to fix what breaks. More sales are a bad bargain when winning and fulfilling them eats the margin.

Becoming Necessary Takes More Than an Integration

Now back to where we started. McLean’s containers didn’t win because they were clever. They won because every port, railroad and trucker could build to the same spec without asking his permission, and because the savings were too big to ignore. The card networks did the same thing with money. Standardize the transaction, give up the front end, own the guarantee, get paid on the flow on billions of transactions.

That’s the test for everyone claiming the infrastructure position in agentic commerce. Solve a problem that gets bigger as more agents, merchants and consumers show up. Delivery certainty. Trusted authorization. Accurate inventory. An order that can be completed and reconciled across systems that were never designed to talk to each other. Make it easy to adopt. Keep the service levels up. Make the economics work on both sides. Eliminate the friction to adopt. Then do it again next quarter, because infrastructure status has to be earned over and over.

The company that owns the shopping conversation and the company that executes the order can both build something valuable. But right now, only one of them gets paid no matter where the consumer starts. Delivery decided retail once. It might be about to decide it again when agents do the shopping.

Until NEXT time.

Join the 21,000 subscribers who’ve already said yes to what’s NEXT.