AI’s growing graveyard highlights the cost of chasing the next big bet
TechCrunch’s catalogue of abandoned, delayed and absorbed AI projects shows how quickly competition, costs and weak demand can reshape the market.
Relay, an AI workflow-automation startup, shut down on Monday after OpenAI, Google and other larger platforms added similar capabilities to their own products. The five-year-old company had offered AI agents for automating email and task workflows, but TechCrunch reported that it struggled to maintain a reason to exist as a standalone service.
Relay is among a growing list of AI products and initiatives that TechCrunch has described as an “AI graveyard”. The catalogue includes projects that were shut down, delayed, rolled back, pivoted or absorbed into larger platforms — meaning not every example represents a definitive failure. S&P Global Market Intelligence estimates that about 42 per cent of corporate AI initiatives are ultimately abandoned.
OpenAI’s attempted ChatGPT “super app” redesign was rolled back after users criticised its cluttered and confusing interface. The company has also discontinued ChatGPT Atlas, an AI-powered web browser, and folded capabilities associated with Operator and DALL-E into ChatGPT. Its video-sharing platform Sora shut down in March 2026 after reported challenges involving operating costs and user retention.
Apple’s repeatedly delayed AI-powered Siri began appearing for English-language users this month, after first appearing in the iOS 27 beta in July. The delays followed engineering and software problems and contributed to a reported US$250 million settlement concerning claims about how the iPhone 16’s AI capabilities were marketed.
Other cases include Microsoft Recall, whose planned “photographic memory” feature triggered privacy and security concerns after it was announced in 2024, and Notion Mail, which is scheduled to shut down on 22 September. Humane closed its AI Pin business in February 2025, with HP acquiring most of its assets for US$116 million.
The list also includes Rabbit’s R1, which continues to receive updates despite poor early reviews, as well as Huxe, Yupp and Figgs AI. Their reported experiences point to recurring pressures across the sector: larger platforms replicating standalone features, difficulty sustaining operating costs and products failing to attract enough demand.
