The AI graveyard: what actually killed these products and startups

From a $230 million wearable to OpenAI's own stumbles, here is what the wreckage of the past two years teaches us about where AI money goes wrong.

AI2Day NewsdeskEditor: Lee Brown4 min read
A close-up, edge-to-edge 16:9 photograph of a smartphone screen displaying a payment confirmation in Indian rupees, resting on a wooden desk beside a chai glass
Share

Key points

  • Relay, a workflow-automation startup, shut down on Monday after larger platforms built the same features into their own tools.
  • About 42% of AI initiatives are ultimately abandoned by the companies that build them, according to S&P Global Market Intelligence.
  • Humane raised $230 million for its AI Pin wearable before shutting down in February 2025; HP acquired most of its assets for $116 million.
  • Apple agreed to a $250 million settlement over how it marketed AI features on the iPhone 16 after repeatedly delaying an upgraded Siri.
  • Notion Mail, launched in April 2025, closes on September 22, 2025, after users bypassed it in favour of standalone AI tools.

Relay was a tool that let small businesses automate repetitive email and scheduling tasks using AI agents, software that can carry out multi-step jobs without someone clicking through each step. It was pitched as a friendlier alternative to Zapier. On Monday, it closed for good.

The reason is straightforward: OpenAI, Google and other large platforms built the same automation features directly into products millions of people already use. A small startup selling a standalone tool can't easily compete with that. Relay was five years old.

This isn't unique to Relay. According to S&P Global Market Intelligence, roughly 42% of AI projects are eventually abandoned by the companies that launched them. The causes range from running out of money to never finding enough users willing to pay.

What does the full list look like?

The failures span hardware, software and some of the biggest names in tech.

TechCrunch AI compiled a running list worth examining. A few entries stand out for what they reveal about where bets go wrong.

Product Company Fate Key detail
Relay Independent startup Shut down Monday Outbuilt by larger platforms
Humane AI Pin Humane Closed February 2025 Raised $230M; sold assets to HP for $116M
Rabbit R1 Rabbit Still operating but struggling Sold 100,000 units; early reviews called it unfinished
Notion Mail Notion Closing September 22, 2025 Users chose rival AI agents instead
Huxe Ex-Google developers Shut down May 2026 Podcast-style audio app; outscaled by bigger players
ChatGPT Atlas OpenAI Discontinued August 9 AI browser absorbed into ChatGPT

The Humane AI Pin is probably the starkest cautionary tale in hardware. It was a wearable device meant to replace your smartphone with a clip-on AI assistant. Humane raised $230 million before launch, shipped the product with performance problems, then had to warn customers about a battery fire risk from the charging case. The business closed in February 2025. HP picked up most of the assets for $116 million, meaning investors got back roughly half their money at best. We covered the Pin's collapse on 30 July 2026 in our first look at AI hardware failures.

Rabbit's R1 device, unveiled at CES in January 2024, sold 100,000 units quickly. Reviewers described it as unreliable and thin on useful features. The company is still trying, now focused on a developer-oriented project called Cyberdeck, aimed at people who use conversational AI to write code.

Should you be surprised that OpenAI is on this list?

No. Even the best-funded lab in AI has shelved products. OpenAI discontinued ChatGPT Atlas, a standalone AI-powered web browser, on August 9, folding its features back into ChatGPT. It did the same with Operator, an early AI agent tool. A July 2025 redesign of the main ChatGPT app tried combining Chat, Codex and Work modes into one interface, drew enough complaints that the company reversed course and restored the familiar layout. Our 4 September story found similar friction around OpenAI's GPT-6 Astra rollout, where paying subscribers were locked out of a model the company had declared the start of the AGI era.

OpenAI's video-sharing platform, Sora, shut down in March 2026, weighed down by high running costs and low user retention.

Apple's problems were more expensive. The company promised a smarter Siri when it announced Apple Intelligence in 2024, then delayed the feature repeatedly. That contributed to a $250 million settlement over how Apple marketed the AI capabilities of the iPhone 16. An improved Siri finally appeared in the iOS 27 beta in July 2025.

Survivorship bias is real: we hear about the products that worked. The list above is the part of the story that usually gets skipped.

The honest take: before you build your workflow around any AI tool, check whether the company behind it has a business model that makes sense without a giant platform absorbing it. If the answer isn't clear, keep a backup plan.

© 2026 AI2Day