Why 42% of AI Bets End Up in the Graveyard

Character.ai filed for bankruptcy in early October, just five years after it launched. A few weeks later, the privacy-focused family assistant Ollie said it was shutting down too. TechCrunch AI has started a running list of AI projects and startups that didn’t make it, and it shows a pattern that should worry anyone building in this space.

The key number comes from S&P Global Market Intelligence: about 42% of AI initiatives end up abandoned by the companies that started them. So the AI graveyard isn’t only for small startups. Big companies are filling it too.

🪦 Even the giants are burying projects

What stands out in TechCrunch AI’s list is how many of the failures come from the biggest names in AI.

  • OpenAI tried to turn ChatGPT into a “super app” on July 9, with separate “Chat,” “Codex,” and “Work” modes and a renamed “ChatGPT Classic.” Users said the new design was confusing and cluttered, and OpenAI rolled it back soon after.
  • ChatGPT Atlas, OpenAI’s stand-alone AI browser, lasted less than a year before it was shut down on August 9. Its best features moved into ChatGPT.
  • Sora, OpenAI’s video-sharing platform, shut down in March 2026. It struggled with high operating costs and with keeping users.
  • Apple’s new Siri slipped again and again because of engineering problems and bugs. The delay helped lead to a $250 million settlement over how Apple marketed the iPhone 16’s AI features. The AI-powered Siri is only now rolling out to English-language users.

OpenAI’s case is especially useful. Folding Atlas, Operator, and DALL-E into ChatGPT has worked out. Cramming everything into one redesigned screen didn’t. The lesson is simple: combine products behind the scenes, and don’t break the interface people already know how to use.

📉 Why startups get squeezed out

Startups are dying for different reasons, and two of them stand out.

Getting copied by the platforms. Relay, an AI workflow tool pitched as a Zapier alternative, shut down in August 2026. Once OpenAI, Google, and other big players built similar automation into their own products, Relay couldn’t give people a good reason to use a separate app. Plenty of AI startups are in the same spot. If your product is one feature that a big platform can copy, you’re living on borrowed time.

Being too small to stand out. Ollie launched in June 2026 with real privacy credentials, including SOC 2 compliance. But it was up against bigger rivals like Muse and Instinct. CEO Bill Lennon pointed to trouble standing out, slow growth, and not enough funding. Strong privacy wasn’t enough to win.

⚖️ Character.ai and the cost of legal risk

Character.ai’s collapse is a different kind of failure. On paper, it looked like a winner. It raised $150 million at a $1 billion valuation in 2023, and Google later paid a reported $2.7 billion to license its technology and bring back its founders.

What followed was a pile of legal trouble. Families sued, alleging the chatbots contributed to teen suicides and self-harm. Disney sent a cease-and-desist letter over the use of its characters. The company banned under-18 users from open-ended chats and settled several lawsuits. That still wasn’t enough, and it went bankrupt citing legal, regulatory, and operational headwinds.

In an ironic twist, Disney is now the opening bidder at $20 million, looking to license some of the company’s IP. The sale should close around late November or early December.

This is significant because it shows that safety and IP risk aren’t just PR problems. For consumer AI products, they can sink the whole company.

🧭 What builders and businesses should take from this

The 42% figure is high, but it makes sense. AI is cheap to prototype and expensive to run, defend, and grow. Here’s what I’d take from the list:

  1. Test whether a platform can copy you. If ChatGPT or Gemini could ship your core feature next quarter, you need a moat that isn’t the model: proprietary data, a deep workflow, or a niche audience.
  2. Work out your running costs early. Sora shows that a product people love can still lose money on every user.
  3. Plan for legal and safety risk from day one. That goes double for consumer chat, kids, and anything built on someone else’s characters or content.
  4. Don’t make loyal users relearn your product. OpenAI’s quick rollback shows how fast familiar users push back.
  5. Be ready to fold, not just kill. Moving the best parts of a failed product into one that’s working saves real value.

What comes next

Expect this list to keep growing. As the big platforms absorb more features and legal scrutiny of consumer AI gets tougher, the middle of the market will keep getting squeezed. The survivors will be the ones that own something the big players can’t easily copy.

You can find the full list, including the story of Microsoft’s Recall, at TechCrunch AI.

Scroll to Top