Microsoft-OpenAI Just Split. Your AI Strategy Should Too.

Microsoft is no longer the exclusive seller of OpenAI's models. The companies amended their partnership on April 27, 2026, capping the revenue-share OpenAI owes Microsoft and freeing both sides to court new partners. For retailers, the practical translation is unglamorous but urgent: ChatGPT and Copilot are no longer one commercial pipe. They are competing surfaces with diverging product strategies, and your agentic commerce feed plan has to reflect that.

TL;DR: The amended Microsoft-OpenAI deal ends Microsoft's exclusivity over OpenAI's models and caps the revenue-share through 2030. ChatGPT and Microsoft Copilot are now separate distribution surfaces with separate product roadmaps, separate protocol stacks (ACP plus UCP on Copilot, ACP-first on ChatGPT), and increasingly separate merchant programs. Treat them as two channels, not one.

What actually changed on April 27?

The new agreement does three concrete things. Microsoft is no longer the exclusive provider of OpenAI's API. OpenAI's revenue-share to Microsoft is now subject to a total cap that runs through 2030. And both parties are explicitly free to pursue independent partnerships, including with cloud and distribution rivals.

Microsoft's official blog frames it as "the next phase." OpenAI's parallel post calls it greater flexibility for both companies. Reuters reporting is blunter: Microsoft has lost its lock on OpenAI's tech, and OpenAI can now negotiate with Amazon, Google, or anyone else.

Until yesterday, you could reasonably model OpenAI distribution and Microsoft distribution as a single supply chain. ChatGPT ran on Azure. Copilot was OpenAI under the hood. The feed you optimized for one tended to work for the other. That assumption is now wrong.

Why this matters for retailers right now

The protocol stack already started forking. Microsoft Copilot's in-app checkout went live with 500,000 sellers in late April, running both ACP (the OpenAI/Stripe protocol) and UCP (the Google/Shopify protocol). ChatGPT runs ACP only. Google's Gemini runs UCP only. The "unified" protocol world stayed unified for about ninety days.

Now the commercial relationship is forking too. OpenAI can build retailer programs that route around Microsoft. Microsoft can ship Copilot features that don't depend on the latest GPT release. Both are likely to do exactly that, because the cap on revenue-share gives each company a financial reason to grow distribution that doesn't pay tribute to the other.

Key stat: 500,000 sellers are live on Microsoft Copilot's in-app checkout as of late April 2026. ChatGPT runs a separate roster of integrated retailers. The two surfaces no longer share a merchant pipeline.

For a mid-market retailer, this means three things. The product feed you push to ChatGPT through ACP is not automatically your Copilot feed. The merchant onboarding programs are separate. And the feature roadmap that determines whether your products appear as a card or a flat text mention is being set by two different product teams with different incentives.

Two surfaces, two feeds, two strategies

The cleanest mental model is now this: ChatGPT and Copilot are like Google and Bing were in 2008. Adjacent, often confused, but with different audiences, different ranking signals, and different commercial relationships with merchants. You optimize for both, but you do not assume one is a free copy of the other.

Here is what differs today, before the split has even fully played out:

Surface	Protocol stack	Merchant program	Checkout model	Audience
ChatGPT	ACP (Stripe/OpenAI)	OpenAI direct integrations + Shopify catalog auto-connect	Discovery and redirect to merchant site	900M weekly users, broad consumer
Microsoft Copilot	ACP and UCP, dual	Microsoft seller program, ~500K live	In-app checkout via PayPal, Shopify, Stripe rails	Embedded in Microsoft 365, enterprise-leaning
Google Gemini	UCP (Google/Shopify)	Shopify, Walmart, Etsy, Wayfair, Target, Ulta, Gap	Native checkout for partner retailers	Search-anchored, mobile-heavy

That is not three flavors of the same channel. That is three channels with overlapping but distinct economics. As of April 2026, Gap launched native checkout inside Gemini, becoming the first major fashion retailer to do so. Ulta did the same a week earlier on the beauty side. Neither launch automatically extends to ChatGPT or Copilot.

What this signals for the next 90 days

OpenAI now has a legal runway to announce non-Microsoft cloud and distribution deals. Watch for an Amazon or AWS partnership announcement; it is the most-flagged possibility in CNBC's coverage. If it lands, ChatGPT distribution gets a second supply chain that competes with Azure-anchored Copilot distribution.

Microsoft is likely to lean harder into Copilot's UCP support to differentiate from a more ACP-pure ChatGPT. That favors retailers already on Shopify, Salesforce Commerce Cloud, or commercetools, because they get UCP plumbing for free. Pure-DTC brands without that plumbing have to build it themselves.

The mid-market retailers who treat this as a single AI channel are about to discover, eighteen months from now, that they ranked well on one surface and were invisible on the other. The question is which one they were invisible on, and how much traffic they lost while not noticing.

What to do this week

Audit your AI surface inventory. List every channel you currently push product data to: ChatGPT (ACP), Google AI Mode (UCP via Google Merchant Center), Microsoft Copilot, Perplexity, Amazon Rufus. If you cannot list them with the protocol each one consumes, you do not have a strategy yet. You have a default. Start with a product feed review for each surface.
Split your performance reporting. If your analytics rolls up "AI traffic" as a single source, separate ChatGPT, Copilot, and Gemini referrers. Each will trend differently from this week forward. A blended number will hide the divergence.
Validate your Copilot presence. With 500,000 sellers live, the Copilot in-app checkout is no longer a pilot. Confirm whether your products appear there, whether you are integrated through Shopify or Stripe rails, and whether your category competitors are.
Pressure-test your ACP feed. ChatGPT is the surface most likely to evolve in the next 60 days as OpenAI exercises its new freedom. Make sure your ACP-compliant feed has the attribute depth, structured data, and image quality to ride changes you cannot predict. Tools like our AI Readiness Report can flag the gaps before a ranking change exposes them.
Stop assuming protocol parity. The convergence story that ran from January through March was real but partial. ACP, UCP, and MCP are not interchangeable. Plan for a stack where you support each one in parallel, and accept that some channels will lead on features you care about and lag on others.

Frequently asked questions

Did Microsoft lose access to OpenAI's models?

No. Microsoft still has rights to OpenAI's models through the agreement. What changed is that Microsoft is no longer the exclusive seller of those models, and OpenAI is now free to license them to other cloud and distribution partners. Existing Copilot integrations continue.

Does this affect ACP, UCP, or MCP?

Not directly. The protocols are governed separately. ACP is open-source under Apache 2.0 at agenticcommerce.dev, UCP is hosted at ucp.dev with Google and Shopify, and MCP is Anthropic's. The amendment is about commercial terms, not protocol governance, but it accelerates the divergence between ChatGPT and Copilot product roadmaps that consume those protocols.

Should I prioritize ChatGPT or Copilot for product visibility?

Both, with different goals. ChatGPT skews consumer and has 900M weekly users per OpenAI and DemandSage. Copilot is embedded in Microsoft 365 and skews enterprise and high-intent. If you sell to consumers, ChatGPT is the higher-volume surface. If you sell B2B or to enterprise buyers, Copilot's audience composition is more valuable.

What about Gemini and Amazon Rufus?

Both are independent surfaces with their own protocols and programs. Gemini runs UCP and now has native checkout for major retailers including Ulta and Gap. Amazon Rufus is a closed system tied to Amazon's catalog and personalization profile. Treat them as separate channels with separate optimization plans.

Is in-chat purchase coming back to ChatGPT?

There is no signal that it is. OpenAI discontinued Instant Checkout in March 2026. Brands rejected being reduced to anonymous fulfillment centers and lost traffic, customer data, and loyalty. ChatGPT now operates as a discovery and recommendation engine and redirects shoppers to merchant sites for purchase. The amended Microsoft deal does not change that.

The era when one feed strategy covered the whole AI shopping internet ended quietly on April 27. The retailers who notice this week, audit their surface coverage, and split their measurement will spend the next year compounding visibility on whichever channel rewards them. The retailers who keep treating "AI traffic" as a single line item will spend that year wondering why one of their two main pipes went quiet.