Inside the AI Graveyard: Why Ambitious Products and Startups Have Failed

Table of Contents

• You might want to know


• Main Topic


• Key Insights Table


• Afterwards...

You might want to know

• Why have so many AI products, startups, and corporate initiatives struggled to survive?


• What can these setbacks teach companies about designing useful, trustworthy, and sustainable AI?

Main Topic

The rapid expansion of artificial intelligence has produced a growing number of products that promise to change how people work, communicate, and use technology. Yet a compelling demonstration or a wave of early attention does not guarantee a lasting business. Some AI companies have closed, some products have been discontinued or absorbed into broader platforms, and others have struggled to deliver on their original promises. Taken together, these cases form an emerging AI graveyard: a record not only of failed bets, but also of the commercial, technical, and human challenges that shape the industry.

Relay, a workflow automation product designed as an AI-powered alternative to Zapier, shut down entirely in August. It allowed users to automate email and task workflows with AI agents. But larger technology companies, including OpenAI and Google, began building similar automation capabilities into their own products. After five years, Relay faced a difficult question: what distinct reason remained for customers to use it as a stand-alone service?

That challenge reflects a wider pattern. According to S&P Global Market Intelligence, about 42% of AI initiatives are ultimately abandoned by their corporate parents. The causes vary. Projects may lack sufficient funding, encounter technical difficulties, struggle to scale, face intense competition, or fail to attract enough users. In some cases, a concept may be technically possible but not useful or distinctive enough to become a sustainable product.

Character.ai, a chatbot platform, filed for bankruptcy just five years after launching. Its difficulties sit alongside the story of Relay and the many AI projects that are abandoned before they become independent businesses. These examples are not simply a list of isolated disappointments. They help illustrate how quickly the market can change and how difficult it is to maintain a clear product identity when larger platforms can reproduce popular features.

A recurring lesson is that AI capability alone is not a durable business advantage: a product also needs a clear purpose, dependable performance, user trust, and a viable path to long-term operation.

Large platforms: consolidation, redesign, and product overlap

Even the biggest AI companies have had products and strategies that did not meet expectations. OpenAI, one of the world’s most valuable private companies, has discontinued stand-alone services, integrated features into ChatGPT, and reversed a major interface change after users objected.

On July 9, OpenAI updated its app in an effort to make ChatGPT a broader “super app.” The redesign brought several experiences together, including separate “Chat,” “Codex,” and “Work” modes, and renamed the familiar version “ChatGPT Classic.” Users criticized the new interface as confusing and cluttered. OpenAI soon rolled back the change and restored the familiar ChatGPT interface. The episode showed that combining more tools in one place does not automatically make a product easier to use.

Other OpenAI products were folded into ChatGPT or discontinued. ChatGPT Atlas, a stand-alone AI-powered web browser, lasted less than a year before being discontinued on August 9. Its most useful features were absorbed into ChatGPT. OpenAI also brought the capabilities of Operator, an AI agent that could browse the web and complete tasks on a user’s behalf, into its broader product direction. Meanwhile, image generation has increasingly been built directly into ChatGPT, reducing DALL-E’s role as a separate destination for creating images.

OpenAI’s video-sharing platform, Sora, faced a different set of pressures. High operating costs and user-retention challenges contributed to its closure in March 2026. The contrast among these cases is instructive: integration can make a feature more accessible, but it may also make a stand-alone product unnecessary; a new product may attract attention, yet still prove too costly or difficult to sustain.

Promises and delays: Siri and Microsoft Recall

Apple’s AI ambitions also faced a gap between the initial pitch and the timeline for delivery. When Apple unveiled Apple Intelligence in 2024, an improved Siri was among its highlighted features. Apple described an assistant that could understand context, recognize what was happening across apps, and complete tasks effectively.

Apple repeatedly delayed the release of the new Siri, citing engineering issues and bugs. The delay later contributed to a $250 million settlement over claims about how Apple marketed the AI capabilities of the iPhone 16. Siri AI eventually appeared in the iOS 27 beta this past July. The new AI-powered version began rolling out to English-language users first this month, with support for other languages expected to follow. The episode demonstrates how ambitious product claims can create expectations that are difficult to meet when technical readiness takes longer than anticipated.

Microsoft’s Recall feature encountered a different problem: privacy. Announced at the 2024 Build developer conference, Recall was presented as an AI-powered “photographic memory” for Windows PCs. It would periodically capture screenshots of a user’s activity so that the user could search through their digital history later.

The announcement triggered an immediate backlash. Critics warned that a searchable archive could contain highly sensitive material, such as passwords, private messages, financial details, and other personal activity. Microsoft delayed Recall for nearly a year to reconsider its privacy and security protections. The debate continued after the redesign: a cybersecurity researcher built a tool capable of extracting and displaying data Recall had captured, reviving questions about whether the underlying risks had been resolved. In this case, the central challenge was not just whether the feature worked, but whether users could trust the way it collected and stored information.

Products that struggled to find a lasting role

Notion Mail launched in April 2025 as an AI-focused email product intended to help people organize and automate their inboxes. The company found, however, that users were turning to separate AI agents to manage email instead. Notion Mail is scheduled to shut down on September 22. Its story highlights a practical test for specialized AI tools: users must see enough value in adopting a separate product rather than relying on other services they already use.

Huxe took a different approach to information consumption. Built by former developers of Google’s NotebookLM, it transformed written material into conversational, podcast-style audio. The company shut down in May 2026, partly as larger platforms began offering similar experiences to established audiences. Spotify, for example, has been expanding its own AI-powered audio tools. A smaller product can have an appealing concept and still struggle when broad platforms bring comparable features to users at scale.

Yupp offered a free way to compare the responses of hundreds of AI models side by side. Users could vote on answers and earn cryptocurrency through the platform. At its peak, it enabled people to test more than 800 models, including models from OpenAI, Google, and Anthropic. According to its founders, however, Yupp never achieved strong enough product-market fit to survive. The platform shut down in March 2026. Its large selection of models was distinctive, but breadth alone was not enough to establish a sustainable service.

Figgs AI operated from 2023 to 2024, allowing users to create and interact with customizable AI characters for role-play and storytelling. The platform reportedly attracted more than 1 million users. Its developers said that keeping the service free became too expensive to sustain, and they shut it down. Figgs AI’s experience underlines the difference between attracting an audience and finding a business model that can support continued operation.

AI hardware: attention is not the same as usefulness

The Humane AI Pin became one of the most widely discussed failures in AI hardware. It was designed to deliver AI capabilities through a wearable device instead of a conventional smartphone. Humane raised $230 million from investors and attracted considerable attention, but the product struggled with performance. The situation became more serious when Humane warned customers to stop using its charging case because of a potential battery fire risk.

Humane shut down its AI Pin business in February 2025. Most of the company’s assets were acquired by HP for $116 million. The case shows how hardware products face multiple hurdles at once: they must be reliable, safe, and useful enough to justify becoming part of a person’s daily routine. Strong interest before launch cannot compensate for problems that users encounter after purchase.

The Rabbit R1 also received substantial attention. Unveiled at CES in January 2024, it was presented as an AI companion capable of carrying out tasks on behalf of its owner. Rabbit said it had sold 100,000 units shortly after launch. However, early reviews described the device as unfinished, with unreliable performance and a limited number of useful integrations.

Rabbit has continued updating the R1 rather than abandoning it. The company has repositioned the device as a computer controller capable of performing agentic tasks. It also announced a new hardware project called “Project Cyberdeck,” intended to create a portable device designed for vibe-coding. These developments do not erase the product’s early shortcomings, but they show that some companies respond to weak initial results by changing a product’s direction and continuing to experiment.

What the AI graveyard reveals

These cases point to several recurring difficulties. One is product differentiation: as AI features become part of widely used platforms, independent tools may lose their reason to exist. Another is execution. Delays, unreliable results, limited integrations, and confusing interfaces can erode confidence even when the underlying idea is attractive.

Trust and safety are also central. Microsoft Recall faced objections over sensitive data, while Humane’s charging case raised a potential fire risk. A product that handles personal information or introduces new hardware into everyday life must address foreseeable risks clearly and convincingly. User acceptance depends not only on what a system can do, but also on whether people feel safe using it.

Finally, financial sustainability and product-market fit remain essential. Huxe faced competition from platforms with larger audiences; Figgs AI could not afford to keep its service free; Yupp’s founders said the product did not achieve a strong enough fit with market demand; and Sora encountered high operating costs and user-retention challenges. Across these different situations, attention, technical novelty, and early adoption are not reliable substitutes for repeat use and a workable business model.

Key Insights Table

Afterwards...

The AI graveyard should be read as a source of evidence, not merely as a scorecard. Each discontinued product offers clues about the conditions required for AI to deliver lasting value. Future work should explore better ways to measure whether AI systems solve meaningful problems, how users behave after the novelty fades, and what forms of deployment remain financially and technically sustainable.

Further research into privacy-preserving design, security testing, dependable agents, accessible interfaces, and responsible hardware could help developers avoid repeating familiar mistakes. Companies should also examine how AI features fit into existing routines, whether users can understand and control them, and how products can maintain a distinct purpose as larger platforms expand.

The next phase of AI progress will depend not only on more capable models, but on careful engineering, transparent safeguards, and products that earn a lasting place in people’s lives. Learning from discontinued projects can help the industry direct investment and research toward systems that are useful, trusted, and built to endure.