The AI Graveyard Is Growing: What Failed Products Reveal

Introduction

The AI industry tends to celebrate launches, funding rounds, and valuations. Yet an equally important story is taking shape on the other side of the market: products are being shut down, folded into larger platforms, or falling short of their original promises. TechCrunch’s running “AI graveyard” shows that these outcomes affect both startups and some of the world’s largest technology companies.

Key takeaways

• Platforms are absorbing stand-alone tools. Relay was built as an AI-powered alternative to Zapier, automating email and task workflows with agents. As OpenAI, Google, and other major companies added similar capabilities to their own products, Relay struggled to justify remaining independent and shut down in August. Notion Mail is also scheduled to close, with the company noting that users were turning to separate AI agents for inbox management.
• Integration is not always a product strategy that works. OpenAI briefly redesigned ChatGPT as a broader “super app,” combining several modes in one interface. Users found the experience confusing and cluttered, and the company rolled it back. Other products followed a different path: Operator and ChatGPT Atlas were discontinued as their most useful capabilities were absorbed into ChatGPT, while image generation has increasingly been built directly into the chatbot rather than centered on DALL-E as a separate destination.
• Promises can take years to reach users. Apple presented a more capable Siri as a major part of Apple Intelligence, promising contextual understanding and cross-app task execution. Engineering problems and bugs repeatedly delayed the release. Microsoft’s Recall faced a different obstacle: its proposed “photographic memory” for Windows could preserve sensitive information through periodic screenshots, triggering immediate privacy and security criticism. Microsoft delayed and redesigned the feature, but questions about its protections remain.
• AI hardware still lacks a reliable formula. Humane’s AI Pin drew significant attention and investment, but performance problems and a warning about a possible battery fire involving its charging case ended the business. HP later acquired most of the company’s assets. Rabbit’s R1 attracted strong early interest, yet reviews described it as unfinished and unreliable, leaving the company to keep updating the device and repositioning it around agentic tasks and computer control.
• Users do not automatically create a sustainable business. Figgs AI reportedly attracted more than one million users, but its developers said the cost of keeping the service free became unsustainable. Model-comparison platform Yupp shut down after failing to find strong product-market fit. Huxe, an app that converted written material into conversational audio, was also squeezed as larger platforms introduced similar experiences.

What the graveyard means

S&P Global Market Intelligence estimates that about 42% of AI initiatives are eventually abandoned by their corporate parents. The reasons range from insufficient funding and technical barriers to competition, scaling problems, and weak demand. The figure suggests that high attrition is not simply the result of a few poorly managed teams. AI capabilities are changing quickly, while products still need time to prove that customers will use and pay for them.

For startups, a single visible feature can be reproduced by a model provider or a large platform. Defensibility may instead come from a deeper workflow, proprietary data, or a durable customer relationship. For large companies, putting more AI features inside one application is not automatically better either: complexity, reliability, privacy, and user trust can determine whether adoption lasts. Character.ai’s legal and safety troubles further show that consumer AI must address youth protection, content responsibility, and intellectual-property concerns alongside model performance.

The AI graveyard should therefore be treated less as a scoreboard of failures than as a record of market lessons. It shows which use cases lacked durable demand, which cost structures did not work, and which promises were not ready for everyday life. The next stage of competition may depend less on launching the most AI features and more on making them stable, simple, safe, and worth using over time.

TechCrunch AI