The AI Graveyard: What Failed Projects Reveal About the Industry

TechCrunch frames the current wave of AI shutdowns and disappointments as a growing “AI graveyard,” covering products, startups, launches, and bets that shut down, pivoted, or missed expectations. The article points to recent examples including Character.ai’s bankruptcy filing and Ollie’s shutdown announcement. It also cites S&P Global Market Intelligence saying about 42% of AI initiatives are ultimately abandoned by their corporate parents. The takeaway: AI demand is real, but funding, technical execution, scaling, competition, and user adoption still decide what survives.

OpenAI is highlighted as a major player that has still faced product missteps, including a criticized attempt to turn ChatGPT into a broader “super app” before rolling back the redesign. TechCrunch also notes that OpenAI folded features from stand-alone products like ChatGPT Atlas and Operator into ChatGPT, while Sora shut down in March 2026 after struggling with operating costs and retention. Apple’s delayed Siri AI is another marquee example, with repeated delays, an iOS 27 beta appearance, and a $250 million settlement tied to claims about marketed AI capabilities. Microsoft’s Recall feature remains a cautionary case for privacy-sensitive AI, after backlash and renewed security questions.

A recurring theme is that useful AI features can become hard to defend as stand-alone businesses once larger platforms absorb similar capabilities. Relay shut down after bigger companies built AI automation into their own tools, while Notion Mail is scheduled to shut down as users turn to separate AI agents for inbox handling. Huxe, an AI audio app from former NotebookLM developers, shut down as larger platforms expanded similar audio tools. Yupp and Figgs AI show another pressure point: even with novel concepts or user traction, product-market fit and operating costs can be decisive.

The Humane AI Pin and Rabbit R1 illustrate the challenges of turning AI excitement into dependable consumer hardware. Humane raised major funding and drew attention for a wearable AI device, but the product struggled with performance and later faced a charging-case warning before the AI Pin business shut down and assets were acquired by HP. Rabbit R1 launched with heavy hype and early sales, but reviews described it as unfinished, unreliable, and limited in useful integrations. Rabbit has continued updating the device and has announced another hardware project, but the broader lesson is clear: AI hardware needs more than novelty to last.

The strongest signal from the list is that AI products must prove they can retain users, manage costs, protect privacy, and avoid being swallowed by platform features. For founders, defensibility matters: a feature that can be copied or bundled by OpenAI, Google, Microsoft, Apple, Spotify, or another platform may not be enough. For buyers and users, the lesson is to check whether an AI tool has a clear roadmap, sustainable backing, and a practical role in daily workflows. The AI market is still moving quickly, but the graveyard shows that speed alone is not a strategy.