AI Graveyard: AI Projects and Startups That Shut Down

7 min read · Last updated: October 25, 2026

The AI graveyard is a running list of AI products, startups and launches that shut down, pivoted or missed expectations. TechCrunch's October 2026 version includes Relay, Sora, ChatGPT Atlas, Notion Mail and Humane's AI Pin. The pattern is consistent: a bigger platform ships the same feature, or the running costs outgrow the users.

Came here from Instagram, LinkedIn, X or YouTube? Here is the full source list you came for. Everything below is built from these, nothing else.

The list

• The AI graveyard: a running list of projects and startups that didn't make it, TechCrunch, October 24, 2026

Key takeaways

• According to S&P Global Market Intelligence, about 42% of AI initiatives are ultimately abandoned by their corporate parents.
• Relay, a five-year-old Zapier alternative, shut down in August after OpenAI, Google and others built similar automation into their own tools.
• ChatGPT Atlas lasted less than a year and was discontinued on August 9. Sora shut down in March 2026.
• Humane raised $230 million, shut its AI Pin business in February 2025, and its assets sold to HP for $116 million.
• Not every entry is dead. Apple's Siri AI, Microsoft's Recall and Rabbit's R1 were delayed, criticised or oversold, then kept going.

What is the AI graveyard, and why keep a list?

An AI graveyard is a record of AI products that were shut down, pivoted or fell short of expectations. TechCrunch frames it less as a scorecard of flops and more as a record of what these attempts can teach the industry. According to S&P Global Market Intelligence, about 42% of AI initiatives are ultimately abandoned by their corporate parents.

Think of a small bakery on a street. It does well because it is the only place selling good bread. Then the supermarket at the corner opens its own bakery counter. The bread is not worse. The supermarket just has the foot traffic already.

That is the shape of most entries below. A stand-alone AI product wins attention first. Then a platform with an existing audience adds the same thing as a feature.

TechCrunch lists the reasons AI bets fail: insufficient funding, technical challenges, competition, difficulty scaling, or weak user demand. Most of the cases below combine two or three of them.

How does an AI product end up in the graveyard?

Most follow five steps: a launch with attention, a bigger platform building the same thing, users drifting to the alternative, costs or retention biting, and then a shutdown or fold-in. Each step below maps to a real case from the list.

1. Launch with a narrow job. Relay let users automate email and task workflows with AI agents. Notion Mail launched in April 2025 as an AI-focused product to organise and automate inboxes.
2. A bigger platform builds the same thing. As OpenAI, Google and others built automation into their own tools, Relay struggled to keep a reason to exist as a stand-alone product. Huxe, an AI audio app that turned written information into podcast-style audio, faced larger platforms offering similar experiences to big existing audiences. TechCrunch names Spotify, which has been expanding its own AI audio tools.
3. Users route around you. Notion noticed its users were handling their inboxes with separate AI agents instead of Notion Mail.
4. Costs or retention bite. Sora struggled with high operating costs and user-retention challenges. Figgs AI reportedly drew more than 1 million users, but its developers said keeping it free became too expensive.
5. Shutdown or fold-in. Notion Mail was scheduled to shut down on September 22. ChatGPT Atlas was discontinued and its most useful features absorbed into ChatGPT.

Notice that step 2 is not about quality. Relay's problem was not a broken product. It was a missing reason to exist alone.

Which AI projects have shut down?

TechCrunch's list covers at least ten products that ended, were folded into another product, or collapsed. This table keeps the facts to what the article states.

Project	What it was	What happened	When
Relay	AI workflow automation, a Zapier alternative	Shut down entirely	August
Character.ai	Chatbot platform	Filed for bankruptcy five years after launch	Recent
Sora	OpenAI's video-sharing platform	Shut down after high costs and weak retention	March 2026
ChatGPT Atlas	Stand-alone AI web browser	Discontinued, features moved into ChatGPT	August 9
Operator	AI agent that browses and completes tasks	Folded into ChatGPT	Not dated in the source
Notion Mail	AI-focused email product	Scheduled to shut down	September 22
Humane AI Pin	Wearable AI device	Business shut down, assets bought by HP for $116M	February 2025
Huxe	AI audio app from former NotebookLM developers	Shut down	May 2026
Yupp	Free playground to compare 800+ AI models	Shut down, no strong product-market fit	March 2026
Figgs AI	Customisable AI characters	Shut down, free tier too expensive	Operated 2023 to 2024

Two details stand out. First, even OpenAI shut down several stand-alone apps and folded their features into ChatGPT. TechCrunch says that strategy has worked out better than its redesign did. It has also built image generation directly into ChatGPT, shrinking the role of DALL-E as a separate destination.

Second, Humane is the clearest hardware case. It raised $230 million and got significant attention, but the product struggled with performance. Humane also warned customers to stop using the charging case over a potential battery fire risk.

Which AI launches missed expectations but survived?

Some launches stumbled without dying. They were delayed, rolled back or criticised, and the products are still going. Four examples from the list show different ways to miss.

OpenAI's super app. On July 9, OpenAI updated ChatGPT to combine separate Chat, Codex and Work modes and renamed the familiar version ChatGPT Classic. Users found the new interface confusing and cluttered. OpenAI rolled the change back soon after.

Apple's Siri AI. Apple unveiled Apple Intelligence in 2024 with an improved Siri that understood context and got tasks done. Apple repeatedly pushed the release back, citing engineering issues and bugs. The delay contributed to a $250 million settlement over how Apple marketed the AI capabilities of the iPhone 16. Siri AI appeared in the iOS 27 beta in July and is rolling out to English-language users this month.

Microsoft's Recall. Announced at Build 2024 as a "photographic memory" for Windows PCs, Recall took periodic screenshots so users could search their history. Privacy and security backlash followed, and Microsoft delayed it for nearly a year. A cybersecurity researcher recently built a tool that extracts data Recall captured, reviving questions about the fix.

Rabbit R1. Unveiled at CES in January 2024, Rabbit said it sold 100,000 units shortly after launch. Early reviews called it unfinished, with unreliable performance and few useful integrations. Rabbit kept updating it and announced a new hardware project, "Project Cyberdeck", aimed at vibe-coding.

The lesson from Rabbit is plain: strong initial sales did not mean a strong product.

When should you build a stand-alone AI product, and when should you not?

Build stand-alone only if a big platform cannot easily copy your core job. If your product is one feature a larger tool could add, expect to be absorbed. This is my reading of the list, not a claim from TechCrunch.

Signal	Lean stand-alone	Lean feature or integration
Who ships the same thing	Nobody with a big audience	OpenAI, Google or Spotify-scale platforms already do
What users do today	Come to you for the job	Already use a separate AI agent for it (the Notion Mail case)
Cost per free user	Low, or you charge from day one	High and unsold (the Figgs AI case)
Retention	Users return on their own	Novelty drops off (the Sora case)
Fit with the product	Solves a real, repeated problem	Needs the hype to sell (the Rabbit R1 case)

If you are early, the cheaper move is to test the idea as a feature inside a tool people already use. OpenAI did the reverse. It built separate apps, then pulled them back into ChatGPT.

If you are building anything in this space, our guide to the 7 deadly mistakes killing 97% of AI startups covers the earlier-stage version of the same problem. For the wider hype question, see The GenAI hype: a revolution, an evolution or a $1 trillion smart illusion.

How can you check an AI idea against the graveyard?

A simple checklist turns the five death steps into questions you can answer before you build. This sketch is my own rule of thumb, not a method from the source. The weights are guesses, so adjust them.

type Idea = {
  platformCouldShipIt: boolean;   // Relay, Huxe
  usersAlreadyUseAgents: boolean; // Notion Mail
  freeTierIsExpensive: boolean;   // Figgs AI, Sora
  retentionUnproven: boolean;     // Sora
  needsHypeToSell: boolean;       // Rabbit R1
};

function graveyardRisk(idea: Idea): "low" | "medium" | "high" {
  const hits = Object.values(idea).filter(Boolean).length;
  if (hits >= 3) return "high";
  if (hits >= 1) return "medium";
  return "low";
}

graveyardRisk({
  platformCouldShipIt: true,
  usersAlreadyUseAgents: true,
  freeTierIsExpensive: false,
  retentionUnproven: true,
  needsHypeToSell: false,
}); // "high"

A "high" result does not mean do not build. It means you need a plan for step 2, the day a bigger platform ships your feature.

Also keep in mind that code is rarely the hard part. We wrote about that in Code was never the hard part.

FAQ

What is the AI graveyard?

It is a running list of AI products, startups and launches that shut down, pivoted or significantly missed expectations. TechCrunch published its version on October 24, 2026. It includes Relay, Sora, ChatGPT Atlas, Notion Mail, Humane's AI Pin, Huxe, Yupp and Figgs AI.

Why do AI startups shut down?

TechCrunch lists insufficient funding, technical challenges, competition, difficulty scaling and weak user demand. In its examples, the most common cause is a larger platform building the same feature into a product that already has the audience, as with Relay and Huxe.

What percentage of AI projects get abandoned?

According to S&P Global Market Intelligence, cited by TechCrunch in October 2026, about 42% of AI initiatives are ultimately abandoned by their corporate parents.

Did Apple's new Siri ever launch?

Yes, after repeated delays. Siri AI appeared in the iOS 27 beta in July 2026, and the AI-powered version is rolling out to English-language users in October, with other languages expected to follow.

Before you build your next AI idea, run it through the checklist above and try the free tools on Art of Code. If you are weighing where AI is heading for your own career, read AI replaced coding: the skills that survive in 2026.

Written by Adithya Guttha, Founder of Art of Code.