The AI graveyard: a running list of projects and startups that didn’t make it

TechCrunch has assembled a running record of AI products, startups and corporate initiatives that have shut down, pivoted or fallen well short of their launches. The piece, framed as a graveyard rather than a scorecard, arrives alongside an estimate from S&P Global Market Intelligence that roughly 42% of AI initiatives are eventually abandoned by the companies that created them. The stated causes range from insufficient funding and technical difficulty to competition, problems scaling and weak user demand.
Character.ai features prominently. Founded in 2021 by former Google researchers Noam Shazeer and Daniel De Freitas, the consumer app let people chat with customizable bots, many modeled on celebrities and fictional characters. It raised $150 million at a $1 billion valuation in 2023. The following year Google reportedly paid $2.7 billion to license its technology and bring its founders back in-house, leaving the startup to continue without them. Legal trouble followed: families sued alleging chatbots contributed to teen suicides and self-harm, and Disney sent a cease-and-desist over unauthorized character use. The company barred under-18 users from open-ended chats late last year and agreed to settle several suits, along with Google, in early 2026. It filed for bankruptcy in early October, citing legal, regulatory and operational headwinds. Disney is now the opening bidder, offering $20 million for a license to some of its intellectual property; the app continues to run during the process, with a sale expected by late November or early December.
OpenAI supplies several entries. On July 9 it restructured ChatGPT into a broader "super app" with separate Chat, Codex and Work modes and renamed the familiar version ChatGPT Classic; users found the interface confusing and cluttered, and the change was rolled back. A more successful pattern has been folding standalone products into ChatGPT itself. The Atlas browser, a standalone AI-powered web browser, was discontinued on August 9 after less than a year, with its useful features absorbed into ChatGPT. Operator, an agent that browsed the web and completed tasks for users, met the same fate, and image generation has increasingly moved directly into ChatGPT, shrinking DALL-E's role as a separate destination. The video-sharing platform Sora shut down in March 2026 after struggling with high operating costs and user retention.
Apple's long-promised Siri overhaul also appears. Apple Intelligence was unveiled in 2024 with an improved Siri as a headline feature, pitched as understanding context, knowing what happens across apps and completing tasks. The release slipped repeatedly over engineering issues and bugs, and the delay contributed to a $250 million settlement over how Apple marketed the AI capabilities of the iPhone 16. The new Siri finally arrived in the iOS 27 beta this past July, rolling out to English-language users this month, with other languages expected later.
Among smaller companies, Relay, an AI workflow automation tool positioned against Zapier, shut down entirely in August 2026. It automated email and task workflows with AI agents, but as OpenAI, Google and other large platforms built similar automation into their own tools, the five-year-old product struggled to justify standing alone. Ollie launched in June 2026 as one of the first mainstream apps to achieve SOC 2, a privacy compliance standard, and aimed to pull family information scattered across school messages, calendars, task lists and group chats into one synced assistant delivered by text. Co-founder and CEO Bill Lennon announced a November 6 shutdown, citing an inability to stand out, failure to grow and insufficient capital, with Muse and Instinct named as heavyweight competitors. Huxe, an AI audio app from former NotebookLM developers that turned written material into podcast-style audio, closed in May 2026 as larger platforms offered similar experiences to existing audiences, Spotify among them. Yupp, a free playground for comparing responses from hundreds of models with voting and cryptocurrency rewards, tested more than 800 models at its peak, including from OpenAI, Google and Anthropic, but its founders said it never reached strong enough product-market fit; it shut down in March 2026. Figgs AI, which ran from 2023 to 2024 and let users create and role-play with custom characters, reportedly drew more than 1 million users but became too expensive to keep free.
Two internal corporate bets illustrate the same pressures. Microsoft announced Recall at its 2024 Build conference as an AI "photographic memory" for Windows PCs that periodically screenshots user activity for later search. It drew immediate backlash over the risk of a searchable archive of passwords, private messages and financial details, and Microsoft delayed it for nearly a year to rework protections. Controversy has persisted: a researcher recently built a tool that could extract and display captured data, reviving questions about whether the underlying problem is fixed. Notion Mail, launched in April 2025 as an AI-focused email product, is scheduled to shut down on September 22 after the company observed users turning to separate AI agents for their inboxes instead.
Hardware is represented by two of the most publicized launches. Humane's AI Pin raised $230 million and promised AI through a wearable rather than a phone, but suffered performance problems and a warning that customers should stop using its charging case over potential battery fire risk; Humane shut the business in February 2025, with HP acquiring most assets for $116 million. The Rabbit R1, unveiled at CES in January 2024, sold 100,000 units shortly after launch, yet early reviews called it unfinished, with unreliable performance and few useful integrations. Rabbit has kept updating it as a computer controller for agentic tasks and announced a new hardware effort, Project Cyberdeck, aimed at portable vibe-coding.
Read together, the pattern is less about bad ideas than about distribution. Several of these products did something useful and were then absorbed into a larger platform that already had the users — the browser, the agent, the automation layer. For a small team, that makes defensibility the central design question: whether a feature is easy for a platform owner to replicate, and whether the product owns a workflow or relationship that survives replication. Regulation and litigation are a second axis; Character.ai's trajectory suggests safety and IP exposure can compound into an existential cost. The failure modes are also unevenly documented, and a public list skews toward companies loud enough to be noticed, so the 42% figure is best treated as a rough signal rather than a precise rate.
Why it matters: teams building AI products now have a concrete set of precedents showing where standalone tools lose to platform bundling, and where legal and safety exposure can turn a fast-growing consumer app into a bankruptcy filing. For developers, the lesson is to plan for absorption — build on interfaces and data others cannot trivially copy, and price for the possibility that a model provider ships your feature next quarter. Product teams evaluating long-term vendors should also weigh durability, since a service that looks permanent can be rolled back or shut down within a year.
Based on reporting from the original publisher. Visit the source for full context and later updates.
Publisher excerpt
From Apple's repeatedly delayed Siri AI to OpenAI's messy "super app" launch, here's a look at the AI projects that shut down or missed expectations.