What the Walmart–OpenAI Arc Actually Showed

Walmart's announcement of its OpenAI partnership on 14 October 2025 had the shape of a typical big-company AI press release: ambitious language about "AI-first shopping experiences," references to agentic commerce and conversational interfaces, a quote from the CEO about running towards the future. Standard issue.

The specifics were more interesting. US customers could soon shop Walmart through ChatGPT using Instant Checkout. Corporate teams would get access to ChatGPT Enterprise. Walmart was among the first retailers to take up OpenAI's new certification programme. And Sparky (Walmart's own AI shopping assistant, built out through 2024 and 2025 to handle everything from meal planning to restocking household essentials) was being integrated more deeply into the overall experience.

The partnership put the second-largest online retailer in the US inside the same distribution channel that Etsy and Shopify merchants had entered two weeks earlier. That's a different kind of commercial signal. When independent Etsy sellers join a new platform, it's market discovery. When Walmart joins, it's a considered bet.

Why Walmart's Scale Changed the Conversation

Walmart at scale is a different proposition from independent Shopify sellers. For Walmart to decide that ChatGPT is a commerce channel worth investing in, with a direct integration, an enterprise deployment across corporate teams, and its CEO publicly framing it as the direction of travel, means the agentic commerce thesis was being stress-tested by an operator with the data and the analytical capability to actually evaluate it.

McMillon put it plainly in the October press release: "For many years now, eCommerce shopping experiences have consisted of a search bar and a long list of item responses. That is about to change." Whether you find CEO announcements credible or not, the operational commitment behind the statement was real. Walmart's 1.6 million US associates make it the largest private-sector employer in the country; deploying ChatGPT Enterprise across corporate teams, and rolling out AI literacy training across the workforce, signals a belief that AI tooling will change how retail operations work internally, not just how the customer-facing experience looks.

Worth noting that Walmart was one of several named early partners for OpenAI's new certification programme, alongside John Deere and Accenture, among others. The press release describes Walmart as "among the first to embrace OpenAI Certifications." Accurate, but not exclusive.

The Sparky Question

The more intriguing element of Walmart's AI positioning was always Sparky. The official Walmart description is that the platform will "learn and predict customers' needs, turning shopping from a reactive experience into a proactive one." The ambition: a system that anticipates what you're about to need, surfaces it before you've articulated the need, and handles the transaction on your behalf.

That is, roughly, the Amazonian dream pursued through a different architecture. Amazon built recommendation systems on top of an explicit search-and-browse model: pull-based, reactive, triggered by intent. Sparky's stated vision is more conversational and more proactive. Whether it delivers on that ambition is a separate question. The aspiration, as of October 2025, was clearly well-funded and seriously meant.

The vendor framing would tell you this is a category-defining moment. The more considered view is that Walmart has an unusually strong starting position for this kind of AI because of the depth of its first-party data: purchase history across grocery, hardlines, fashion, pharmacy, and fuel, from a customer base that shops regularly, at scale, across income demographics. Amazon has more e-commerce breadth; Walmart has more real-world shopping depth. That underlying data asset is what makes the Sparky thesis plausible rather than just aspirational.

What Actually Happened Next

The more instructive part of the Walmart–OpenAI story is what came out in March 2026, when reporting from around the Morgan Stanley Technology, Media and Telecom conference revealed that Instant Checkout had been quietly wound down.

Reading the post-announcement coverage: conversion rates for products sold directly inside ChatGPT were three times lower than for those that rerouted users to a retailer's own website. Walmart's EVP of AI acceleration, Daniel Danker, described Instant Checkout as "a very temporary moment in time." OpenAI, for its part, acknowledged that the initial Instant Checkout design lacked the flexibility retailers needed, and pivoted to focus its Agentic Commerce Protocol on product discovery rather than checkout, letting retailers own their own purchase journeys.

Walmart's response was not to step back from AI commerce. It was to step forward differently: embedding Sparky directly into ChatGPT and Google Gemini rather than relying on OpenAI's generic checkout infrastructure. The distinction matters. Rather than renting shelf space inside someone else's agent, Walmart's own assistant sits inside the AI platform, retaining direct access to Walmart's product catalogue, pricing, and customer data.

Danker's framing of the pivot: "Shopping can start anywhere now, whether that's Walmart or a question in ChatGPT. Our approach is simple: wherever it begins, customers still get the same personalised Walmart experience." Early pilot data from the new Sparky integration, according to TechBuzz.ai, suggested purchase completion rates roughly 70% of those on Walmart.com directly. Substantially better than Instant Checkout, though this figure comes from unconfirmed internal data and should be read as directional rather than definitive.

The Pattern Worth Watching

The Walmart arc follows a recognisable shape: initial excitement about an AI platform partnership, discovery that the default integration underconverts, then a strategic pivot towards proprietary AI capability embedded within the platform rather than dependent on it. It will probably repeat across retailers through 2026 and beyond.

This is not a new problem in new clothes. The history of retail technology is full of early integrations with distribution platforms that looked strategically compelling until the economics came in: marketplaces that extracted too much margin, ad networks that converted below threshold, fulfilment partnerships that handed over customer data without proportionate return. The pattern with AI commerce channels has the same structural logic. The platform owner has an incentive to generalise; the retailer has an incentive to differentiate. Those incentives eventually diverge.

The retailers that do well in agentic commerce will be the ones who own their agent, own their data, and make that agent available wherever consumers happen to be asking questions. Not the ones who outsourced both to a platform and hoped for the best. For UK retailers watching from the sidelines, the lesson from Bentonville is fairly direct: if you have first-party customer data that's genuinely deep (a loyalty programme, an own-label range, a digital infrastructure that knows what the customer bought last week and the week before), then you have something worth protecting and worth deploying.

Tesco, Sainsbury's, M&S: the data asset is there. The question is whether the product development capability and the operational tolerance for risk-taking are there alongside it. That combination is harder to assemble than it looks. It's also what's actually being competed on. Not AI platforms. Not vendor partnerships. The underlying data infrastructure and the capacity to act on it.

---

Update — March 2026

The October announcement played out as outlined above. By March 2026, Walmart had wound down OpenAI's Instant Checkout and embedded Sparky directly inside ChatGPT and Google Gemini. The conversion problem turned out to be a genuine friction point rather than early-stage teething: users asking an AI to make a purchase and not completing it. The lesson Walmart drew is that integrating your own agent into an AI platform works better than depending on the platform's generic checkout infrastructure. That is now the working model for the Sparky rollout, which reached ChatGPT Plus and Gemini Advanced subscribers in late March 2026.

Sources: Retail Dive, TechBuzz.ai, CNBC

---

The retailers who own their agent own the conversation. The ones who don't are renting shelf space on someone else's terms.