AI Apps on the App Store: The Growth Is Real, and So Is the Crowd

Two curves on the App Store are going up at the same time, and AI is behind both. The first is money: people are paying for AI apps at a rate the store has not seen for a new category in years. The second is supply: more new apps are being released than at any point in roughly a decade, largely because AI tools made them cheaper to build.

If you are planning an AI app, the first curve is the reason to do it and the second is the reason it is harder than it looks. This article goes through the numbers behind each, what the surge has done to App Review and to Apple's rules, and what I would build differently because of it.

Sensor Tower's State of Mobile 2026 put global in-app purchase revenue for 2025 at $167 billion and reported that non-game apps out-earned games for the first time. The firm credits generative AI services for a large part of that shift.

The same firm's State of AI 2026 report projected that in-app revenue from AI apps would pass $4 billion in the first half of 2026, up 36% on the previous half-year, and that time spent in generative AI apps would roughly double year over year, from 17.2 billion hours to 36 billion. Those are projections published in June, not final counts. On the consumer side, ChatGPT was the most downloaded free iPhone app in the US in 2025, ahead of Threads and TikTok.

The headline numbers hide how concentrated the category is. According to Sensor Tower, ChatGPT, Google Gemini and DeepSeek accounted for nearly 90% of all time spent in AI assistant apps in the first quarter of 2026. A new general-purpose chat app is competing for the remaining tenth against every other assistant, and all three leaders have a free tier.

The more useful number for anyone building a product is a different one from the same report: apps that mention AI in their store listing were on track for nearly 10 billion downloads in the first half of 2026, spread across health, finance, education and utilities. AI has turned into a feature that users expect inside ordinary apps. The opening for a new product is a specific job done well with AI inside it. Another chat window is not one.

Now the supply side. Appfigures data reported by TechCrunch showed worldwide app releases up 60% year over year in the first quarter of 2026 across the App Store and Google Play, and up 80% on the iOS App Store alone. Early April was running at 104% across both stores. The Information reported an 84% jump in App Store submissions in a single quarter, which The Next Web described as the largest in ten years.

The obvious explanation is AI coding tools, and the timing fits, although nobody has measured how many of those apps were built that way. Games still make up most new releases. What is not in doubt is the consequence: the number of apps grew far faster than the number of people looking for them. Being found was already the hard part of the App Store, and it got harder.

In March 2026 developers began reporting review waits of a week to a month, against the usual one to two days. Apple has disputed how widespread the delays were. The practical lesson is the same either way: a launch date that depends on a 24-hour review is a risk. Put slack between submission and any marketing you cannot move.

Apple also started enforcing an old rule more strictly. Guideline 2.5.2 says an app may not download or execute code that changes its features after review. In mid-March Apple blocked updates to app-building tools including Replit and Vibecode on that basis, and at the end of the month it removed an app called Anything from the store. Apple has no rule against apps written with AI. The rule is about apps that generate and run new functionality on the device after approval. If your product idea involves users creating mini-apps or automations with a prompt, read 2.5.2 before you write any code.

In November 2025 Apple revised guideline 5.1.2(i) to name AI directly: 'You must clearly disclose where personal data will be shared with third parties, including with third-party AI, and obtain explicit permission before doing so.' Sharing data without permission was already forbidden. Naming AI made it something reviewers check for.

In practice, if your app sends anything personal to OpenAI, Anthropic, Google or any other outside model, the user has to see a clear explanation and agree before it happens. A line in the privacy policy is unlikely to cover it, and developers have reported rejections where the consent screen existed but reviewers judged it unclear. The cheapest way to satisfy the rule is to not send the data at all, which is one of the reasons I push features onto the device where the model allows it. I went through what that looks like in the guide to Apple Intelligence and on-device AI.

RevenueCat's 2026 State of Subscription Apps report, covered by TechCrunch, is the least comfortable reading in this article for anyone with an AI app. AI apps now make up about 27% of subscription apps in its dataset. They are reported to earn noticeably more per paying user than other apps. They also lose subscribers faster: annual retention was 21.1% for AI apps against 30.7% for the rest, annual subscribers cancelled 30% faster at the median, and refund rates were 20% higher.

My reading is that novelty sells the first payment and only repeated usefulness sells the renewal. An app whose main appeal is 'it has AI' gets tried, produces one impressive result, and is cancelled before the second billing date. There is a cost side as well. A cloud model charges you every time a user does something, so the heaviest users of a flat-price subscription are the least profitable ones. That has to be in the pricing model before launch, not discovered from the first invoice.

Start from the job, and let AI be how it gets done. People renew for a result they would miss: a finished document, a logged meal, a translated conversation. If the pitch still works after you delete the word AI from it, the product probably has a reason to exist.

Decide per feature where the model runs. On-device inference costs nothing per request, works offline, and avoids the third-party consent screen. Cloud models are stronger and cost money on every call. Most products end up with a mix, and choosing it deliberately affects margin, privacy and review all at once.

Plan for review and for being found. Build the consent flow before the first submission, leave room in the schedule for a slow queue, and treat the store listing as part of the product. With this many new apps arriving, the screenshots and the first line of the description do more work than they used to. I covered that stage in From Idea to App Store.

Watch retention from the first cohort. Downloads and trial starts arrive early and look good. The renewal number arrives late and decides whether there is a business. Instrument it from day one.

This is the work I do: native iOS apps with AI where it earns its place, taken from the idea through review to the store. If you are weighing an AI app and want a straight read on whether it has room in this market, tell me what you are building.