Mapping the AI Graveyard: Startups and Big Tech Features That Failed
From Humane's hardware collapse to OpenAI folding standalone tools, recent closures and pivots highlight the fragile economics and platform risks shaping the AI industry.
The rapid pace of generative AI development often overshadows how many initiatives fail along the way. According to data from S&P Global Market Intelligence, roughly 42% of corporate AI projects are ultimately terminated before or after launch. Driven by intense competition, high infrastructure costs, technical bottlenecks, and weak product-market fit, a growing number of both venture-backed startups and high-profile internal corporate bets are ending up in what has become an expanding AI graveyard.
Recent closures underscore how difficult the landscape has become for standalone platforms. Relay, a workflow automation tool created to rival Zapier, shut down in August after five years when major platform owners like OpenAI and Google built comparable agentic features directly into their suites. More dramatically, conversational platform Character.ai filed for bankruptcy five years after launching. These failures demonstrate that neither early momentum nor notable investor interest provides insulation against changing market realities.
Platform Consolidation and Big Tech Stumbles
Even market leaders with massive capital reserves face misfires. OpenAI has repeatedly re-evaluated its interface and product boundaries. On July 9, the company attempted to turn ChatGPT into a multi-purpose super app with dedicated "Chat," "Codex," and "Work" sections while branding its core experience "ChatGPT Classic." Following severe pushback from users over interface clutter, OpenAI quickly restored the original layout.
OpenAI has also increasingly eliminated separate standalone applications in favor of embedding capabilities into ChatGPT. Its specialized browser project, ChatGPT Atlas, was discontinued on August 9 after less than a year, with core functions integrated into ChatGPT. Web-browsing agent Operator underwent a similar consolidation, while DALL-E's footprint as a standalone destination shrank in favor of direct ChatGPT image generation. Meanwhile, Sora, OpenAI's video-sharing initiative, ceased operations in March 2026 amid steep operating overhead and user retention challenges.
Legacy tech giants have encountered their own technical and public relations setbacks. Apple repeatedly delayed its revamped Siri AI—initially showcased as the cornerstone of Apple Intelligence in 2024—due to software bugs and engineering delays. These postponements contributed to a $250 million legal settlement over marketing representations tied to the iPhone 16. The updated Siri finally surfaced in the iOS 27 beta this past July, beginning its rollout in English.
Similarly, Microsoft faced immediate privacy criticism over Windows Recall, an AI-powered visual timeline introduced at Build 2024 that logged periodic screen captures. The company delayed the rollout for nearly twelve months to implement architectural adjustments, yet security concerns resurfaced after a cybersecurity researcher created an extraction utility proving the feature's captured data remained vulnerable.
Hardware Bottlenecks and Feature Saturation
Dedicated AI consumer hardware has proven especially treacherous. Humane, which secured $230 million in venture funding to launch its wearable AI Pin as a smartphone replacement, shuttered the hardware operation in February 2025. The device suffered from performance limitations and faced safety warnings over battery fire hazards in its charging unit, culminating in HP acquiring most of Humane's assets for $116 million.
Rabbit experienced a similar gap between initial reception and real-world viability. After selling 100,000 units of its R1 assistant following its debut at CES 2024, early users encountered an unfinished operating experience with few practical integrations. Rabbit has since shifted toward positioning the R1 as a task controller and announced "Project Cyberdeck," a specialized portable system intended for vibe-coding.
Software point solutions have not fared much better against shifting platform incentives and server expenses:
- Notion Mail: Debuting in April 2025 to automate email management, Notion scheduled the tool's shutdown for September 22 after discovering users preferred employing separate AI agents.
- Huxe: Created by former Google NotebookLM developers to turn text into conversational podcasts, the platform closed in May 2026 as incumbent players like Spotify expanded native audio generation.
- Yupp: A testing playground that let users compare outputs across more than 800 models from providers including OpenAI, Google, and Anthropic, Yupp shut down in March 2026 after failing to find sustainable product-market fit.
- Figgs AI: Despite attracting over 1 million users for customizable storytelling characters, the service ceased operations after its operators found the server costs of sustaining a free product unmanageable.
What it means for developers
The casualties across the AI ecosystem reveal clear operational lessons for software engineers and project architects. First, building thin software layers around capabilities that foundational model providers can easily add as native features carries high platform risk. Relay and Huxe both collapsed once dominant platforms shipped integrated alternatives.
Second, inference and compute costs dictate product viability. Products like Figgs AI and Sora demonstrate that high user volumes do not guarantee sustainability if operating expenses outpace revenue. For engineers building software today, avoiding heavy fixed infrastructure and rigid vendor lock-in is essential. Developers looking to stay flexible can test and try top AI models cheaply through one API at https://apixoai.online, keeping prototyping lean without overcommitting resources.
Finally, the market remains skeptical of dedicated single-purpose hardware and invasive OS-level logging, as evidenced by Humane, Rabbit, and Microsoft Recall. Developers succeed by meeting users inside their existing environments and delivering dependable, privacy-focused utility rather than novelty form factors.
Source: The AI graveyard: a running list of projects and startups that didn’t make it — TechCrunch AI. Written by the Apixo team from that report.
One key for Claude, GPT, GLM, DeepSeek and more. Pay per token with crypto.
Get your API key
