The AI Graveyard: Projects and Companies That Didn't Survive

Success stories are no longer the only thing shaping the AI market. Alongside the glossy announcements and massive funding rounds, a growing list is forming of products and companies that have shut down, pivoted away from their original path, or simply failed to deliver what users were promised. According to estimates from S&P Global Market Intelligence, roughly 42% of corporate AI initiatives are ultimately abandoned, for reasons ranging from insufficient funding and technical hurdles to difficulties scaling and weak actual demand.

When the Big Platform Swallows Third-Party Products

The clearest example of this phenomenon is Relay, a workflow automation tool that positioned itself as an alternative to Zapier through AI agents handling email and everyday tasks. After five years in operation, the service shut down entirely in August, finding itself without an independent reason to exist once OpenAI and Google built similar automation capabilities directly into their core platforms.

The lesson here is a familiar one in the software industry: building a product on top of a gap in a major platform is a calculated risk, because the platform can fill that gap whenever it chooses. In a related development, chatbot platform Character.ai declared bankruptcy just five years after launching, despite its broad popularity among younger users.

Even the Giants Stumble

Failures aren't limited to startups. OpenAI itself has run experiments that didn't last, most notably its attempt to turn ChatGPT into an "everything app." In the update released on July 9, the company split the experience into separate modes for chat, coding, and work, and renamed the traditional interface "ChatGPT Classic." The reaction was swift: users described the new interface as cluttered and confusing, and the company reversed the change and restored the familiar design.

By contrast, another strategy did work for the company: folding standalone apps into ChatGPT rather than keeping them separate:

• ChatGPT Atlas: An AI-powered web browser that didn't make it to its first anniversary before being discontinued on August 9, with its key features migrated into ChatGPT.
• Operator: An agent capable of browsing the web and completing tasks, which ultimately ended up merged into the core experience.
• DALL·E: Its role as a standalone image generation platform diminished after the same capability was built directly into ChatGPT.
• Sora: The video-sharing platform shut down in March 2026 due to high operating costs and difficulty retaining users.

Promises Long Overdue: The Siri Case

When Apple Intelligence was unveiled in 2024, the "new Siri" was the star of the show: an assistant that understood context, was aware of what was happening inside apps, and executed tasks efficiently. But Apple repeatedly postponed the launch due to engineering problems and software bugs — a delay that contributed to a $250 million legal settlement over how AI capabilities were marketed in the iPhone 16. The updated assistant didn't actually surface until the iOS 27 beta last July, beginning its rollout to English-language users first, with other languages to follow.

Privacy as a Roadblock: Microsoft's Recall Feature

Microsoft introduced Recall at Build 2024 as a "visual memory" for Windows PCs, periodically capturing screenshots so users could search their own digital history. But the announcement triggered an immediate wave of privacy and security criticism, with experts warning against creating a searchable archive containing passwords, private messages, and financial data. Microsoft delayed the feature for nearly a year to rebuild its protections, yet the controversy resurfaced when a security researcher managed to build a tool that extracted and displayed the data the feature had collected.

Misreading User Behavior

Notion launched Notion Mail in April 2025 as an AI-based email product for organizing and automating inboxes. But the company noticed that its users preferred relying on independent AI agents to manage their mail, and decided to shut the service down on September 22. The message is clear: a genuine need existing doesn't mean your product is the form people want it to take.

AI Hardware: Enthusiasm Isn't Enough

The Humane AI Pin remains the most famous failure in AI hardware. The company raised $230 million and generated enormous media attention around a wearable device meant to replace the phone, but poor performance undermined the experience — compounded by warnings to customers not to use the charging case because of a battery fire risk. Operations ceased in February 2025, and HP acquired most of the assets for $116 million.

As for the Rabbit R1, unveiled at CES in January 2024, the company announced it had quickly sold 100,000 units, but early reviews described it as an unfinished product with unreliable performance and limited integrations. Even so, the company continues shipping updates and is repositioning the device as a controller capable of agentic tasks, while also announcing a new hardware effort called "Project Cyberdeck" aimed at rapid prototyping and coding.

What Can We Learn?

This list is less a record of failure than it is material for learning. The recurring takeaways: build value that major platforms can't easily replicate; respect user expectations before redesigning what they've grown accustomed to; treat privacy as a design requirement rather than a marketing add-on; and finally, don't confuse media hype with sustainable demand. Those who absorb these lessons early spare themselves a spot on a list that grows longer every year.

✦ بقلم فريق دروب أيديا