Hype or major investment alone is not enough to ensure the survival of an artificial intelligence project. That is what a list compiled by TechCrunch of projects and products that were shut down, changed direction, or failed to fulfill the promises accompanying their launches demonstrates. The list draws on multiple cases, including startups and products within major companies, as well as an estimate from S&P Global Market Intelligence stating that approximately 42% of artificial intelligence initiatives are ultimately abandoned by the companies that own them.
The reasons vary from one project to another, but include lack of funding, technical challenges, difficulty scaling, competition, and weak demand. In other cases, the product does not disappear entirely; instead, its functions are integrated into a broader service or its position is redefined after it becomes clear that the standalone product does not have sufficient justification for continuing.
When the Larger Platform Becomes a Direct Competitor
Relay, a product for automating workflows using artificial intelligence agents, was shut down after five years of work. It enabled the automation of email and daily tasks as an alternative to Zapier, but faced a fundamental problem as OpenAI, Google, and larger platforms developed similar automation tools within their existing products. In practice, it was no longer easy for an independent product to defend its position when the same function reached tools customers were already using.
The integration pattern was repeated within OpenAI. The company discontinued ChatGPT Atlas, an independent AI-powered browser, after less than a year on August 9, while moving its most prominent functions to ChatGPT. The company took a similar path with Operator, an agent capable of browsing the web and carrying out tasks on behalf of the user, while image-generation functions became more deeply integrated into ChatGPT, reducing DALL-E’s independent role. The Sora video-sharing platform was also shut down in March 2026 amid high operating costs and challenges retaining users.
Rapid Launches Do Not Make Up for Product-Market Fit
Notion launched Notion Mail in April 2025 as an AI-focused email product for organizing the inbox and automating certain tasks. However, the company observed that its users were relying on separate AI agents to manage email, and the service is therefore scheduled to be discontinued on September 22.
Yupp, meanwhile, offered a free arena for comparing the answers of hundreds of models side by side, with the ability to vote on results and earn cryptocurrency. At its peak, it enabled users to test more than 800 models, including models from OpenAI, Google, and Anthropic, but it was shut down in March 2026 after its founders said it had not achieved strong product-market fit. Figgs AI, which enabled the creation of customized artificial intelligence characters for role-playing and stories, faced a different problem; although it attracted more than one million users, its developers said keeping the service free had become unaffordable, so it was shut down after operating between 2023 and 2024.
Hardware and Big Promises Under Scrutiny
Humane AI Pin stands out as one of the most prominent artificial intelligence hardware failures. The company raised $230 million and sought to provide AI functions through a wearable device instead of a traditional phone, but the product suffered from major performance problems. Humane then warned customers not to use the charging case because of a potential battery-related fire. The AI Pin business was shut down in February 2025, and HP acquired most of the company’s assets for $116 million.
Rabbit R1 faced a less definitive fate. The device, unveiled at CES in January 2024, attracted broad interest, and Rabbit said it sold 100,000 units shortly after launch. However, early reviews described it as unfinished and pointed to unreliable performance and a limited number of useful integrations. Rather than discontinue it, the company continued updating it and presenting it as a tool for controlling computers and carrying out agentic tasks. It also announced a new hardware project called Project Cyberdeck, dedicated to a portable device built around the idea of vibe-coding.
What Changes in Practice?
These cases show that evaluating an artificial intelligence project should not depend solely on the number of models, the size of its funding, or initial sales. Its survival also depends on having a function that major platforms would find difficult to copy, recurring demand, affordable operating costs, an understandable user experience, and convincing privacy and security safeguards.
This became clear in the trajectory of Apple’s new Siri. When Apple introduced Apple Intelligence in 2024, it promised an assistant that understood context, connected what was happening across apps, and carried out tasks effectively. But Apple repeatedly delayed the launch because of engineering problems and errors, and the delay contributed to a $250 million settlement concerning allegations related to how the artificial intelligence capabilities of the iPhone 16 were marketed. The AI-powered Siri appeared in the beta version of iOS 27 in July and began rolling out to English-language users this month, with support for other languages expected later.
Microsoft Recall’s experience reveals that the problem may lie in trust rather than technical capability. The feature, announced at the 2024 Build conference, was designed to record periodic snapshots of what the user was doing on a Windows device so they could search their digital history. This immediately raised concerns about creating a searchable archive containing passwords, private messages, and financial data, so Microsoft delayed the feature for nearly a year to reconsider its security and privacy controls. Even after it was redesigned, a cybersecurity researcher reignited questions after building a tool to extract the data collected by Recall.
From this perspective, the “AI graveyard” is less a definitive list of failures than a record of the practical limits of a rapidly expanding industry. Some projects have truly ended, some have moved into another product, while projects such as Rabbit R1 continue to evolve despite their initial setbacks. Therefore, the most important question for every new launch remains whether it offers sustainable value that the owning platform or a larger competitor cannot easily provide.