The AI Graveyard: Why Projects and Startups Are Shutting Down
Summary
TechCrunch presents a running record of AI products, startups, launches and corporate initiatives that shut down, pivoted or fell short, framing them as lessons for an industry where many bets do not survive. Relay, a five-year-old AI workflow automation startup positioned against Zapier, shut down after larger platforms such as OpenAI and Google built similar automation into their own products. S&P Global Market Intelligence estimates that about 42% of corporate AI initiatives are eventually abandoned, with funding shortages, technical problems, competition, scaling difficulties and weak user demand among the cited reasons. OpenAI’s examples include a confusing ChatGPT “super app” redesign that was quickly rolled back, the discontinuation of the stand-alone ChatGPT Atlas browser and Operator after their features were absorbed into ChatGPT, the reduced role of DALL-E as image generation moved into ChatGPT, and the shutdown of Sora in March 2026 amid high operating costs and retention challenges. Apple’s promised AI-enhanced Siri was delayed repeatedly because of engineering issues and bugs; it appeared in the iOS 27 beta in July, began rolling out to English-language users in the reported month, and the delays contributed to a $250 million settlement over iPhone 16 marketing claims. Microsoft’s Recall faced privacy and security backlash because it periodically captured screenshots that could contain sensitive information, and a researcher later demonstrated a tool for extracting captured data despite a redesign. Notion Mail, an AI email product launched in April 2025, is scheduled to close on September 22 after users increasingly relied on separate AI agents for inbox work. Humane shut down its AI Pin business in February 2025 after performance problems and a charging-case battery-fire warning; HP acquired most of its assets for $116 million after the company had raised $230 million. Rabbit R1 sold 100,000 units shortly after launch but drew criticism for unreliable performance and limited integrations, while Rabbit continues updating it and developing Project Cyberdeck. Other closures include Huxe, an AI audio app competing with larger platforms; Yupp, a model-comparison playground that once supported more than 800 models but lacked sufficient product-market fit; and Figgs AI, a customizable AI-character platform whose developers said free operation became too expensive. Together, these cases show the range of ways AI products can fail, from being absorbed into larger ecosystems to facing unresolved technical, privacy, cost or demand problems.