Clarity for AI buildersSponsored by AI Routing
availability

The AI graveyard: Why some products shut down while others were folded into larger apps

The AI graveyard: Why some products shut down while others were folded into larger apps

TechCrunch’s AI roundup shows how costs, weak demand and platform competition can cut short even well-funded projects.

The publication cites S&P Global Market Intelligence as finding that companies ultimately abandon about 42% of their AI initiatives. The examples in the roundup are not all shutdowns, though. Some products disappeared while their features lived on, some launches slipped, and others stayed on the market after disappointing debuts.

When a feature stops being a product

TechCrunch reports that OpenAI rolled back a ChatGPT redesign that split the app into Chat, Codex and Work modes after users complained that it was confusing and cluttered. But OpenAI kept moving features into ChatGPT: it discontinued its stand-alone Atlas browser and folded in its most useful features, took a similar approach with the Operator web agent, and made DALL-E less of a separate destination by integrating image generation.

Its Sora video-sharing platform, by contrast, shut down in March 2026 after struggling with operating costs and user retention.

For independent companies, the same shift can take away a reason for customers to use a separate app. Relay shut down in August 2026 after larger platforms began offering similar workflow automation in their own tools. Huxe, which turned written material into podcast-style audio, closed in May 2026 as larger services began offering similar experiences to audiences they already had.

Notion Mail faced a different version of the problem: TechCrunch says Notion found that users were turning to separate AI agents to manage their inboxes instead, and the company scheduled the email product for shutdown.

Funding, usage and a sustainable business

Character.ai 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 back its founders. Character.ai later faced lawsuits from families alleging its chatbots contributed to teen suicides and self-harm. It restricted open-ended chats for users under 18 and agreed, along with Google, to settle several suits.

TechCrunch reports that Character.ai filed for bankruptcy in October 2026, citing legal, regulatory and operational problems. The app was still running during the proceedings, with Disney offering $20 million as the opening bidder for a license to some of the company’s intellectual property.

Other closures show why attracting users and sustaining a service are separate tests. Figgs AI reportedly drew more than 1 million users, but its developers said they could no longer afford to keep the service free. Yupp let users compare responses from more than 800 models at its peak, yet its founders said it never found a strong enough market and closed it in March 2026.

At Ollie, a family assistant that brought together calendars, messages and tasks, CEO Bill Lennon cited trouble standing out, growing and raising enough money when he announced the shutdown.

Delays and disappointing launches are not the same as closures

Apple’s promised Siri upgrade was repeatedly delayed, contributing to a $250 million settlement over claims about how Apple marketed the iPhone 16’s AI capabilities, TechCrunch reports. The upgraded assistant later appeared in the iOS 27 beta, with English-language users first in the rollout. Microsoft delayed Recall for nearly a year after a backlash over the sensitive information its searchable screenshots could retain.

Questions persisted after a redesign when a researcher demonstrated that captured data could still be extracted.

The hardware cases ended differently, too. Humane shut down its AI Pin business in February 2025 after performance complaints and a warning that its charging case posed a potential battery fire risk. HP bought most of Humane’s assets for $116 million. Rabbit said it had sold 100,000 R1 devices shortly after launch, but early reviews found unreliable performance and few useful integrations.

Unlike Humane, Rabbit kept updating the device and repositioned it as a tool for controlling a computer. Initial sales measured interest in the device; they did not settle whether it could reliably do what buyers expected.