Why OpenAI Is Pulling Back On Its E-Commerce Ambitions

Welcome to Memorandum Deep Dives. In this series, we go beyond the headlines to examine the decisions shaping our digital future. 🗞️

This week, we’re looking at a gap the AI industry is starting to run into: the distance between influence and monetization. For the past two years, the assumption was straightforward. If users were already asking AI what to buy, the natural next step was to let them complete the purchase without ever leaving the conversation.

That logic helped fuel one of the most ambitious bets in generative AI: turning chat interfaces into full-stack commerce platforms. With projections pointing to trillions in AI-driven e-commerce, companies raced to compress the entire shopping journey into a single thread, from discovery to checkout. But OpenAI’s quiet retreat from in-chat payments suggests something more fundamental is at play.

The Assumption Behind AI Commerce

Ever since OpenAI released the ChatGPT chatbot, there has been chatter about how the technology will be monetized by the enterprises that developed it. Over time, as adoption increased, a straightforward idea began to take shape. If hundreds of millions of people are already asking AI chatbots for product recommendations, the logical next step is to let them buy without leaving the conversation.

Projections further reinforced this idea, with Ark Investments estimating that AI agents will enable $9T in global e-commerce sales by 2030, or 25% of total global e-commerce sales. This was an opportunity no AI developer wanted to miss, and soon every major AI platform was moving to embed e-commerce in its chatbots.

The result was that, through 2024 and 2025, e-commerce emerged as one of the major battlegrounds in generative AI, and Google, OpenAI, Perplexity, and Amazon all raced to get consumers to start their shopping journeys in AI apps.

Perplexity launched Buy with Pro in November 2024, partnering with PayPal for in-chat checkout. Google soon followed with its Universal Commerce Protocol at NRF in January 2026, backed by Shopify, Walmart, and Target. Driving these launches was the logic that conversational AI could compress the entire shopping funnel: discovery, comparison, decision, and transaction, all inside a single chat thread.

However, OpenAI’s recent retreat from native checkout inside ChatGPT suggests that turning a chatbot into a storefront is far more difficult than anyone anticipated, and that the economics of AI-native commerce may be structurally broken at the point that matters most, which is the moment a user actually has to pay.

OpenAI’s financial imperative

To understand why OpenAI pushed so aggressively into commerce, one has to understand the financial pressure driving it. The company is projected to lose $14B in 2026 alone, with cumulative losses expected to reach $115B through 2029 before it reaches profitability. Only about 5% of ChatGPT’s 800M+ weekly active users pay for subscriptions. That math creates an existential monetization problem: subscriptions alone cannot cover the cost of running inference at this scale.

OpenAI’s response was a two-pronged push into advertising and commerce. In January 2026, the company began testing ads at the bottom of ChatGPT responses for free-tier and Go-tier users in the U.S., with internal projections targeting $1B in ad revenue for 2026, scaling to $25B by 2029, but ads were only half the equation.

On 29 September 2025, OpenAI launched Instant Checkout, allowing U.S. ChatGPT users to buy directly from Etsy sellers inside the chat interface, with over a million Shopify merchants slated to follow. The Agentic Commerce Protocol, co-developed with Stripe and positioned as an open standard for AI-driven transactions, powered the feature.

In the following months, OpenAI signed commerce partnerships with Walmart, Target, PayPal, Salesforce, and, most recently, Amazon to realize its dream of AI-driven transactions.

The plan was to take a fee on each completed transaction, creating a revenue stream that scaled with user intent rather than subscription willingness. For a company burning through cash at an unprecedented rate, commerce looked like the bridge between AI’s enormous costs and its eventual profitability.

The retreat

Six months later, the reality of the dream came knocking, and in early March 2026, The Information reported that OpenAI was walking back native checkout. The reason cited for the backlash, according to internal data, was a pattern that should have been predictable: users were asking plenty of product-related questions but rarely completing purchases inside the chat.

They compared laptops, hunted for gift ideas, and researched running shoes, then left ChatGPT to buy through channels they already trusted.

OpenAI confirmed the shift, saying it was “evolving” its commerce strategy and moving Instant Checkout into retailer apps, where transactions could happen more smoothly. The wording was careful, but the message was clear. TD Cowen analysts called the move a “stunning admission,” suggesting that the idea that AI platforms would replace apps as the new operating system for online activity may be further away than the industry expected.

Why AI-native checkout keeps failing

The failure of AI-native shopping is not a product design problem; it is a behavioral one. Shopping is a habit built on years of muscle memory: saved credit cards, stored addresses, loyalty points, and the simple comfort of a checkout flow a user has completed dozens of times before.

Asking people to enter payment information inside a chatbot, one they primarily associate with answering questions and generating text, introduced friction where the entire premise was supposed to eliminate it.

Forrester’s March 2026 consumer survey data confirmed this directly: among regular answer-engine users in the U.S., UK, and Canada, completing a purchase within an answer engine was the least-adopted use case. Asking general questions and researching products ranked highest.

Forrester’s analyst framed it bluntly: the checkout moment is proving to be the most difficult to replicate in the AI channel because payments are complex, require heavy compliance, and involve an emotional process where it is easy to lose a customer’s patience, trust, and attention.

OpenAI also ran into infrastructure problems that underscored how far the platform was from being a real commerce player. As of February 2026, the company had reportedly not built a system for collecting and remitting state sales taxes, a fundamental requirement for any entity processing U.S. retail transactions.

Additionally, real-time product data synchronization across millions of retailers proved difficult to scale, and fraud prevention added another layer of complexity that established e-commerce platforms have spent years solving.

The structural issue is that commerce infrastructure is not a software feature that can be bolted onto a chatbot. It is a deeply regulated, operationally intensive system that requires inventory management, tax compliance, fraud detection, returns processing, and merchant support. Amazon, Shopify, and Stripe have spent decades building this, and OpenAI’s shortcut of using a protocol and a handful of partnerships cannot replace decades of learning.

From storefront to search engine

Though OpenAI’s retreat does not end the idea of AI-assisted shopping, it changes the role of chatbots within it. Instead of acting as the store, ChatGPT is becoming a discovery layer that recommends products and sends users to retailer sites to complete purchases.

The shift suggests that while people are comfortable using AI to search and compare, they still prefer familiar checkout systems when money is involved. The broader vision of agent-driven commerce remains alive, but with a more limited scope.

For now, AI may drive traffic and influence decisions, yet the final transaction is likely to remain with established platforms rather than happening inside the chatbot itself.

The harder question

OpenAI’s retreat is a signal about the structural difficulty of monetizing a chatbot beyond subscriptions and API access. If commerce doesn’t work as a revenue stream, and ads are still in early testing with uncertain user tolerance, then justifying its $14B annual burn rate narrows considerably.

The company’s own projections assume advertising and affiliate revenue will eventually monetize the 95% of users who do not pay. But those projections were built alongside a commerce strategy that has now been largely abandoned.

Sam Altman acknowledged the tension in January 2026, saying, “a lot of people want to use a lot of AI and don’t want to pay,” and expressed hope that an ad-supported model could work.

The irony is hard to miss: the company that was supposed to represent a paradigm shift in how software is built and monetized is converging on the oldest business model in tech, showing people ads, while its most ambitious alternative revenue stream failed within six months of launch.

As for the rest of the industry, the lesson is that discovery and transaction are fundamentally different activities, and being good at one does not automatically translate to the other.

So, while chatbots are proving remarkably good at helping users think, compare, and decide, they are far less effective at getting them to pay.

That gap matters because the economics of generative AI depend on turning intent into revenue at scale. If commerce cannot close that loop, and advertising remains uncertain, the industry may have to accept that the most widely used AI products are also the hardest to monetize. For now, AI may reshape how people shop, but the money will continue to flow through the same platforms that controlled it before the chatbot arrived.