The Maze: Walmart's AI-shopping bet is less about making Sparky charming and more about feeding it data that Amazon, search engines, and pure-play marketplaces cannot easily copy. The retailer is turning stores, repeat grocery trips, local inventory, pickup, delivery, membership, marketplace depth, and retail media into one recommendation system. The model can answer. Walmart wants the order.
Walmart is moving AI shopping from prompt to pantry. Its January Gemini partnership puts Walmart and Sam's Club products inside Google's assistant and lets linked accounts use prior online and in-store purchases for recommendations. That matters because grocery is not a one-off query. It is weekly memory. Milk, bananas, detergent, dog food, school snacks, and last Tuesday's forgotten cereal become training signals for the next cart. Walmart can bring those signals from physical stores into AI shopping without asking the customer to rebuild intent from scratch.
Sparky is the handoff layer, not just a chatbot. Walmart's AI team has described a future where third-party LLMs can pass the customer into Sparky, then Walmart decides which products appear and in what order. That is the real control point. In the March TMT transcript, Walmart tied Sparky to prior store and online history, almost 0.5 billion marketplace and store items, and a typical supercenter with roughly 150,000 to 180,000 items. Search used to sell shelf position with keywords. Agentic commerce sells trust in the answer.
The store network turns AI promises into operational promises. Walmart's Q1 FY27 materials show why the offline base matters. Walmart U.S. ecommerce grew 26%, delivery grew 45%, and more than 36% of U.S. store-fulfilled deliveries arrived in under three hours. The company also said it delivered more than 3.5 billion units same or next day globally in the quarter. That gives AI shopping a physical back end: recommend the item, know where it sits, offer pickup or delivery, then close the cart before the customer opens another app.
The economics reach beyond the basket. Sparky weekly active users more than doubled in the quarter, and Walmart said Sparky users had average order value about 35% higher than non-Sparky customers. Global advertising grew 37%, while Walmart U.S. advertising grew 36%. Those two lines belong together. If Walmart keeps control of product ranking inside assistant-led journeys, AI shopping can protect retail media from becoming an external model's toll road.
The awkward part is trust. Walmart can personalize because it sees what households actually buy, not only what they search. That is powerful. It is also sensitive. The clean version is convenience: reorder what runs out, switch between store and delivery, and use membership benefits without hunting. The messy version is a retailer deciding which product the agent says first. AI commerce may look conversational. Underneath, it is still shelf space with better memory.
Why it matters: AI shopping will not reward the company with the prettiest assistant. It will reward the company with the richest permissioned customer graph and the best fulfillment certainty. Walmart's old advantage was density: stores near households. Its new advantage may be context: stores inside the data loop. For retailers, the lesson is brutal and useful. In agentic commerce, inventory, identity, loyalty, delivery, and ads stop being separate departments. They become one answer engine.
