The ChatGPT App Store is the Largest Unbuilt Consumer Market
When the iPhone first launched in 2007, it had no App Store. It had just 16 pre-installed applications made by Apple. Only a year later did the App Store launch and helped created the greatest consumer product of our generation.
Nearly four years into the ChatGPT era, the consumer AI app boom has yet to arrive. Where are the consumer AI apps?
Apple's App Store ecosystem facilitated an estimated $149 billion in digital goods and services sales in 2025 and for the first time non-games out-earned games. Cumulative payments to App Store developers passed $550 billion in January 2026.
Consumers are getting used to paying for software.
ChatGPT has passed a billion active users. It has built identity, billing, a developer platform, a component library, a directory, and even a review process for "plugins". But there hasn't been a takeoff for a third-party consumer AI software market. Of the 1,600 apps in OpenAI's directory, nearly all belong to companies that already exist elsewhere. No consumer software company has been founded and scaled inside ChatGPT.
The ChatGPT App Store is the largest unbuilt market in consumer software today.
ChatGPT's Blank Prompt Box and The Case for Dedicated Consumer AI Software
ChatGPT's interface is a blank prompt box, and the blank prompt box creates a paradox for consumers and for OpenAI alike.
The model behind it performs an enormous range of specialized jobs. It can prepare a tax return, render a piece of furniture into a photograph of your living room, and, in OpenAI's own favorite example, book a flight. Most people do not know this.
OpenAI knows this and has tried to compensate with suggested prompts on the home screen and in its advertising, but there is a limit to what it can infer from a user's data and a limit to the screen space available to present it. The models are capable of these jobs, but the users who'd pay for them have no idea that these capabilities exist.
Consumer products solve these problems.
They succeed by making one focused job feel obvious and repeatable. The ability to make existing capabilities discoverable and easy is what makes applied consumer AI companies valuable.
Early signs of demand for consumer apps inside ChatGPT.
Write For Me, a GPT consisting of a single instruction telling the model to produce written material, has accumulated over 50 million conversations. It contains no proprietary model, no data and no engineering. The base model produces the same output for anyone who types the instruction themselves. Yet forty-eight million people preferred this packaged version of ChatGPT.
The paradox therefore closes on itself. Founders do not build the dedicated applications that unlock the model's value because those applications look like wrappers, and the model stays underused by consumers because nobody has built the applications. This creates an argument for a storefront inside ChatGPT, where developers build these wrappers cheaply and quickly, utilizing OpenAI's established trust, identity, and billing infrastructure.
The Three Conditions a ChatGPT App Store must Satisfy
The Apple App Store is the best reference case for OpenAI because it is the clearest instance of a platform company declining to build the applications itself. Apple competes with its developers selectively, in Maps, Music and Fitness, but it did not build Uber, Instagram or Slack. It built the conditions under which others could build them and sell them to Apple's users, and it collected a commission.
OpenAI has built the first. It has avoided building the second. The third condition receives the least attention but does the most damage to OpenAI's attempt to become the AI platform company.
OpenAI's Past Attempts on Building this Marketplace
There have been three attempts to build a third-party ecosystem inside ChatGPT between March 2023 and October 2025. The attempts map to the conditions above: the first failed on distribution, the second on monetization, and the third on developer experience.
Attempt I: Plugins, March 2023. Failed on distribution.
Users enabled plugins manually before beginning a conversation and could run no more than three simultaneously. The structural problem is that an application store operates on a pull model, in which users browse and select, while a chat interface operates on a push model, in which users arrive with a task and expect the assistant to determine the tool. OpenAI shut the system down entirely on April 9, 2024.
Attempt II: The GPT Store, January 2024. Failed on monetization.
GPTs improved distribution but introduced no method of payment. The store demonstrated substantial demand for packaged capability. Write For Me, a GPT consisting of an instruction directing the model to produce written material, has accumulated roughly 48 million conversations. Consensus has approximately 18 million. Neither became a business, because OpenAI provided no mechanism to charge for a GPT.
Attempt III: The Apps SDK, October 2025. Failed on Developer Experience.
OpenAI launched with seven partners including Spotify, Canva and Zillow. The engineering is materially stronger than either predecessor, in that applications are MCP servers rendering their own interface inside the conversation, and OpenAI ships a supporting component library. The component library is the constraint. OpenAI's UI guidelines instruct developers to limit an inline card to two primary actions, avoid deep navigation, avoid nested scrolling, and hold carousels to between three and eight items. The ChatGPT composer must remain overlaid on every app, even in full screen. The system supports one interaction pattern, present a result and accept one of two responses, which is why the resulting directory contains no new companies.
On July 9, 2026 OpenAI renamed the App Directory the Plugin Directory, demoting apps back to developer terminology.
OpenAI as the Consumer AI Platform Company
So far, OpenAI's reasons for declining to build a marketplace appear commercial. A revenue share on third-party subscriptions creates a conflict with Apple while OpenAI's own application is distributed through Apple's store. Applications charging twenty dollars a month compete with the twenty dollars OpenAI already collects for Plus subscriptions. Physical-goods commerce monetizes sooner and feeds the advertising business. However, OpenAI still stands to gain more than it loses from building an App Store.
Apps are the only switching cost left once "good-enough" models converge.
Models at the "good-enough" level are converging. With competition from other labs, hyperscalers, and open source, ChatGPT stands to lose its user share among the general population that does not require Astra for most use cases. Inevitably, this level of intelligence commoditizes and OpenAI must compete in a capital war. An application ecosystem adds friction to switching from ChatGPT to Claude.
A user who loves ChatGPT for its custom consumer apps (built only for ChatGPT!) will not have access to them from another model provider. Moreover, the billing, identity layer, and component library can become a significant moat among developers building for ChatGPT.
A hundred thousand developers build more reasons to pay for ChatGPT.
Roughly 50 million of a billion users pay for ChatGPT, a conversion rate near 5%. OpenAI needs more reasons to subscribe than it can generate internally, and a hundred thousand developers building reasons is a larger supply than one product organization.
What Users Stand to Gain
Apps inherit the user's memory. Nothing outside of ChatGPT can.
Every standalone application starts from zero. The tax assistant does not know you moved states, the interior designer does not know you have two children, and each onboarding is a re-interview. An application inside ChatGPT can inherit, with the user's permission, the memory, files and history that ChatGPT already holds. The first time a user opens a benefits-enrollment application it already knows the household. That is a capability no application outside the platform can have, and it is the single largest product advantage the storefront confers. Once again, switching costs from ChatGPT to any other provider increase.
Each job gets the interface it should have had.
The component library, extended beyond the two-action card, gives each specialized job the shape it should have had.
One identity, one bill, one review process.
No new account, no new password, no new twenty-dollar subscription stacked on the existing one. Purchases clear against the billing relationship that already exists, at the price point of an app rather than a service. Every application has passed OpenAI's review and runs inside OpenAI's sandbox.
What Developers Stand to Gain
Auth, billing, hosting and memory are inherited by developers.
A consumer AI company today builds authentication, billing, payments, model hosting, a memory system, and a distribution strategy before it writes the first line of the product. Inside the storefront, all of that is inherited. Developers can just focus on building the best possible AI app.
Together these describe a change in what a consumer software company is. The cost to build has collapsed with Codex. The cost to distribute collapses with the storefront. What remains is taste and iteration, and those are the inputs consumer founders have always had in surplus.