OpenAI loses to Qianwen; the chat window can't display the checkout counter

Author: Kaori

In March 2026, OpenAI officially announced the abandonment of Instant Checkout.

Just six months earlier, this feature had been touted by Silicon Valley media as a key component in establishing the future of Agentic Commerce, with Sam Altman even depicting it on multiple occasions as a crucial money printer for OpenAI. Now it has died due to nearly zero actual transaction conversions.

Interestingly, on the Pacific side, Alibaba’s Qianwen AI shopping has fully launched testing. When you say “I want a cup of milk tea” in the chat box, a delivery person can knock on your door half an hour later.

Both are AI, both are chatbots, yet one is a failed fantasy while the other is a powerhouse of real consumer spending.

The same path has led to completely different outcomes; the difference is not because one model is smarter than the other.

When we talk about AI shopping, what exactly are we discussing?

In the second half of 2025, this was the hottest track. ChatGPT had 900 million weekly active users; even if only one ten-thousandth made a purchase in conversation, the revenue could be staggering.

But in the six months from the grand launch of Instant Checkout to its cancellation, many harsh stories unfolded.

Those Without Shelves Cannot Open Supermarkets

Let’s turn the clock back to September 2025, the day OpenAI launched Instant Checkout, when the entire retail sector was in a frenzy.

Shopify President Harley Finkelstein called it the new frontier of online retail.

On the launch day, Etsy’s stock jumped 16%. Partners quickly came on board, allowing American sellers’ products to appear directly in ChatGPT, even helping merchants bear transaction costs to get the service up and running quickly.

Walmart listed about 200,000 products.

PayPal planned to integrate its wallet into ChatGPT checkout and, as part of the partnership, promised to increase its procurement of OpenAI’s API and enterprise ChatGPT subscriptions.

Stripe and OpenAI co-developed the Agentic Commerce Protocol, attempting to set industry standards for AI agent transactions.

But six months later, the bubble burst.

OpenAI had promised to integrate over 1 million Shopify merchants, but only a pitiful 30 were actually online. The dedicated landing page created by Shopify for ChatGPT has now silently redirected back to the homepage.

What’s worse, internal data from OpenAI indicated that although many users browsed and compared prices in ChatGPT, almost no one actually placed an order in the chat interface.

According to data from Walmart, the conversion rate for jumping back to the retailer’s website for checkout is three times that of staying in ChatGPT. Forrester’s survey also confirmed this; among users who frequently use AI Q&A engines, completing purchases within the engine has the lowest adoption rate.

Why doesn’t shopping in ChatGPT work? Because OpenAI attempted to act as an e-commerce platform without owning any commercial infrastructure.

The most superficial reason is habit. People use ChatGPT like they use Google—to search for information and make comparisons. Once they find what they want, they go to a trusted site to swipe their card.

Asking users to input their credit card number in a chat interface just feels unsafe. Users are willing to have AI help them choose skincare products, but when it comes time to pay, that cold chat box can’t provide the reassurance they need.

Moreover, even if users dared to pay, OpenAI couldn’t handle it.

As of February 2026, OpenAI had not even set up a system to collect and remit sales tax across states in the U.S. This is infrastructure that Amazon and eBay took years to develop.

Not to mention fraud detection, returns processing, and consumer protection compliance. Ensuring real-time accuracy in price, inventory, and shipping information across millions of products is not just a matter of writing a few lines of pretty code; it’s a quagmire.

Forrester analysts pointed out that Instant Checkout had frequent errors during its launch period, didn’t support multi-item shopping carts, lacked promo codes, and even had opaque shipping information.

The most awkward situation isn’t for OpenAI itself, but for the partners that were brought in.

PayPal not only invested engineering resources for integration but also committed to increasing its procurement of OpenAI’s API and enterprise subscriptions. Now the returns on the shopping side have evaporated, yet the procurement commitments still stand.

According to insiders, PayPal and OpenAI are evaluating how to continue their relationship.

Etsy’s situation is not easy either. It previously absorbed transaction costs for merchants, and now it has to start from scratch to create its own ChatGPT app, with the fee structure still not finalized.

An Etsy spokesperson said they are not even clear if OpenAI will charge for in-app transactions.

Stripe’s situation is relatively better, as it was already handling OpenAI’s consumer subscription payments, and this revenue stream does not rely on shopping functionality. But most partners do not have such a cushion.

For a company that is shifting focus towards enterprise customers, this flip-flopping partnership model poses a significant risk.

Alibaba Did It, But It’s Both a Gift and a Shackles of Ecology

On January 15, 2026, while OpenAI’s Instant Checkout was still struggling, something happened on the other side of the globe. Alibaba held a press conference in Hangzhou, announcing that Qianwen App would fully integrate with Taobao, Alipay, Taobao Flash Sale, Fliggy, and Amap.

At the press conference, Wu Jia, President of Qianwen’s C-end business group, said into his phone, “Help me order 40 cups of Bawang Chaji’s Boya Juexian,” and half an hour later, a rider arrived.

If we compare OpenAI’s Instant Checkout with Qianwen’s AI shopping, the reason Qianwen succeeded is not because one big model is smarter than the other; the difference lies in who owns the entire chain from finding products to receiving packages.

Taobao’s hundreds of millions of SKUs, Alipay’s native payments, and the logistics network of the “three links and one reach” are all readily available.

When you say in Qianwen, “I want to hike Mount Siguniang next week, what equipment do I need?” it can directly provide a list, and with a click on the card, you can place an order very smoothly, because there are no inter-company data authorization negotiations or profit distribution disputes involved.

This is why domestic players like DeepSeek and Kimi cannot achieve this. No matter how strong their reasoning capabilities or long-text processing abilities are, without shelves and payments, they can only throw you a link.

Alibaba has integrated shelves into the chat box, while OpenAI attempted to make the chat box pretend to be a shelf.

But can this really be considered a perfect final form?

Alibaba can do it because it has dragged an extremely heavy commercial ecosystem into the large model, but when Qianwen acts as both referee and athlete, can its recommendations remain objective?

If I ask Qianwen which phone is good, will it prioritize recommending a product on Taobao for commercial reasons? When an AI loses its neutral search position and becomes a super sales assistant for its own e-commerce platform, can it still be considered a general-purpose brain? Essentially, this is using a heavy old ecosystem to hijack a new technology entrance.

Qianwen’s “success” is both a gift and a shackle of the ecosystem.

$50 Billion and the Elephant in the Room

OpenAI has also realized that it lacks this heavy ecosystem, but that is not the only reason for its abandonment. The real elephant in the room is Amazon.

At the end of February, Amazon announced a $50 billion investment in OpenAI, becoming the exclusive third-party cloud provider for its enterprise platform Frontier.

When your biggest backer is a giant that holds 40% of the U.S. e-commerce market and is actively promoting its own AI shopping assistant, continuing to develop a checkout system within the app makes you look very out of touch.

Moreover, this money itself is a powder keg. Microsoft believes that hosting the Frontier platform on AWS violates its exclusive cloud agreement with OpenAI and is considering legal action. OpenAI’s lawyers are using technical terms like “stateful architecture” to circumvent the spirit of the contract.

To survive in the competitive landscape dominated by giants, OpenAI must make choices.

In mid-March, OpenAI’s business CEO, Fidji Simo, announced a significant strategic shift at an all-hands meeting: “We cannot miss this moment due to side projects.”

What has prompted OpenAI’s urgency is Anthropic’s rapid rise in the enterprise market. The Claude Code and Cowork products have made Anthropic the preferred choice for enterprise customers, and Ramp credit card data shows that new enterprise customers are choosing Anthropic at three times the rate of OpenAI.

Last year, OpenAI spread itself too thin: Sora video generation, Atlas browser, Jony Ive hardware devices, e-commerce functionality, advertising business, adult mode.

Now, they must refocus on two core battlefields: coding tools and enterprise customers. After all, making money from enterprise customers is much more stable than scraping for transaction fees in a chat box.

OpenAI expects that enterprise customers will contribute half of its total revenue this year, up from the current 40%. To achieve this, it plans to double its workforce from 4,500 to 8,000, with new hires focused on engineering, research, sales, and product development.

In San Francisco, OpenAI signed a new lease, expanding its office space to over 1 million square feet.

The Real Battlefield of AI Shopping Is Not at the Cash Register

OpenAI’s retreat does not mean that AI shopping is dead; on the contrary, the upper half of the funnel has been completely reshaped.

More than half of American consumers have become accustomed to letting AI help them make decisions. People no longer search for floor cleaners and flip through ten pages of ads but directly ask, “Which one has the best cost-performance ratio?” Discovering, researching, and comparing are all being pushed forward, and the value of retailers’ own channels is being ruthlessly eroded.

However, the last mile of the transaction loop does not require smarter models but a more complete infrastructure.

OpenAI has clearly stated that it will prioritize product search and discovery next, and will add ads in ChatGPT. This is its way of monetizing at the discovery layer, which is much more realistic than trying to create a checkout system.

Ultimately, the entity most likely to replicate Alibaba’s path in the U.S. is not OpenAI, but Amazon. It has user profiles, product graphs, payment channels, and fulfillment infrastructure.

OpenAI’s attempt to create its own e-commerce loop has failed, and after receiving money from the largest e-commerce platform, it may ultimately become the traffic entrance for this e-commerce platform.

In China, Alibaba’s full-stack advantage has allowed Qianwen to take a different path, but this path is currently only viable for Alibaba.

Qianwen President Wu Jia made a key statement: The comprehensive Agent has a very strong competitive edge, and vertical Agents are increasingly proving to be a transitional product. In the future, there won’t be many AI applications that can stand alone as entrances.

Translated into business language, this means that in the future, those platforms that already possess a complete ecosystem will be the ones capable of closing the AI shopping loop, rather than AI companies that start from scratch.

You can’t grow a checkout counter in a chat box, but if the checkout counter is already in your store, casually placing a chat-enabled AI next to it feels entirely natural. This is the most important lesson of AI shopping in 2026.