The AI graveyard is growing, but almost everything that dies is killed by a bigger platform, not by the technology

By TechCrunch · October 24, 2026.

TechCrunch has published a roundup of AI projects and startups that have shut down, pivoted or fallen short of expectations. Among the recent cases it cites: Relay, an AI workflow automation tool, shut down in August after five years. Character.ai has filed for bankruptcy five years after its launch.

The context figure, which the outlet attributes to S&P Global Market Intelligence: around 42% of AI initiatives end up being abandoned by the parent companies that back them. Among the causes, the article points to insufficient funding, technical difficulties, competition, scaling problems or weak demand.

The list also includes stumbles by the big players. According to TechCrunch, OpenAI updated ChatGPT on July 9 to merge its “Chat”, “Codex” and “Work” modes, and renamed the traditional version “ChatGPT Classic”. Users criticized the interface as confusing and cluttered, and the company reversed the change shortly afterward. In addition, ChatGPT Atlas, its AI browser, was discontinued on August 9 after less than a year, with its most useful features absorbed into ChatGPT. Operator, the agent that browsed the web, went the same way. Sora, the video platform, shut down in March 2026 due to high operating costs and retention problems, according to the outlet.

Other cases: Notion Mail, launched in April 2025, will shut down on September 22 because, according to the article, its users preferred to use standalone AI agents to manage their email. Huxe, an audio app created by former NotebookLM developers, shut down in May 2026, partly because larger platforms such as Spotify offer similar experiences. Yupp, a comparison tool covering more than 800 models, shut down in March 2026 after failing to reach product-market fit, according to its founders. The roundup also revisits the already well-known cases of the Humane AI Pin (shut down in February 2025; HP bought most of its assets for $116M after it had raised $230M), the Rabbit R1, Microsoft Recall and the delay of the new Siri, which according to the outlet contributed to a $250M settlement over iPhone 16 advertising.

**Our take.** The list mixes failures of different kinds, and it is worth separating them. Some products failed because they were bad (Humane, the early Rabbits), while others did not fail as technology: they were absorbed. Relay, Huxe, Notion Mail and Atlas share a pattern. When an AI capability becomes a feature, it stops being a product, and whoever controls distribution keeps it. Broadly, this is the same dynamic we have already seen with the “plumbing” of agents and platforms: the model matters, but whoever has the audience and the integration wins.

There is a useful paradox here. The OpenAI case shows that not even the biggest players are immune: integrating features into a single product worked, but forcing a confusing “super app” did not. The lesson is about product, not technology. Nor is a 42% abandonment rate a bad sign in itself: in a technology that changes every quarter, quickly abandoning what does not work is a cheap way to learn, provided the lessons are actually learned.

The short-term cost is real. Behind every shutdown are teams that lose their jobs, investors who lose capital and users who lose data or habits (Character.ai, with a highly engaged user base, is the most sensitive example). Consolidation into a few platforms also concentrates power, and that warrants scrutiny.

Even so, our thesis is that this graveyard is not an argument against AI but its selection process. What survives accumulates: useful features end up built into the tools we already use, at lower cost and without us having to think about them. It is the usual path toward abundance: what is a startup today becomes a free feature for everyone tomorrow. Anyone founding an AI company should ask themselves one thing: what happens to my product the day the giant of the moment includes it as standard.