More than $100 billion flowed into AI companies between 2022 and 2024. A large portion of that capital is now gone. Products shut down. Companies sold off assets at steep discounts. Founders cashed out and walked away from products nobody uses anymore. I track this list because nobody else wants to say it clearly. The failures tell you more about where AI is heading than any product launch announcement will.
The Graveyard Is Bigger Than You Think
The AI investment frenzy started the moment ChatGPT launched in late 2022. Every VC with a checkbook needed an AI portfolio company by early 2023. Valuations went vertical fast. Founders raised on demos, not on products. The money moved quick, the timelines were short, and the pressure to ship was constant.
By 2024, the reckoning started. Companies that raised at billion dollar valuations could not grow into them. Tools that felt unique in 2022 got commoditized by OpenAI and Anthropic the following year. The companies that burned the most cash without building something proprietary started disappearing.
This is not a story about AI failing as a technology. AI is real and it keeps improving. But the money that chased it was not always smart money. And the companies that burned through it are piling up.
The Projects That Didn’t Make It
Start with the biggest failure. IBM spent more than $4 billion building Watson into a healthcare product that was supposed to read medical records and help doctors treat cancer. According to an investigation by STAT News, Watson’s oncology recommendations were often unsafe, unreliable, and based on flawed training data that doctors at major hospitals refused to trust. IBM sold off the Watson Health division in 2022 at a significant loss. Four billion dollars into a product that doctors would not use.
Meta launched Galactica in November 2022 with the goal of building an AI that could write and summarize scientific research. It lasted three days. According to reporting by The Verge, the model generated confidently wrong information about scientific topics so fast that Meta pulled it offline before the week ended. It was a public embarrassment at a company that had already spent heavily on AI research.
Amazon’s Alexa tells a longer and more expensive story. The company announced layoffs of more than 10,000 employees in its devices and services division in late 2022, according to reporting by The New York Times. After a decade of pouring money into the smart speaker product, Amazon never found a business model that made the investment worth it. Alexa became a household name without becoming a profitable product.
Inflection AI raised $1.3 billion in 2023 to build a personal AI companion called Pi. The company had serious talent, serious funding, and a clear vision. Then in March 2024, its CEO and most of its key engineers moved to Microsoft in what amounted to a talent acquisition deal, according to widespread reporting at the time. The original product vision effectively ended that week.
Stability AI was one of the loudest names in generative image AI in 2022 and 2023. By early 2024, the company’s CEO had resigned amid reports of unpaid cloud computing bills and financial instability, according to Bloomberg. The company survived under new leadership but its window as an independent force had closed.
Jasper AI raised $125 million at a $1.5 billion valuation in October 2022 on the strength of AI writing tools. Then OpenAI shipped ChatGPT with similar capabilities at a fraction of the cost, and Jasper had to cut staff and pivot. It has not disappeared, but it never became the company its valuation implied it would be.
The pattern across all of them is the same. Single use tools with no real moat. Expensive compute with no path to profit. Products built on top of models they did not own and could not differentiate from competitors. The companies that survived built proprietary data, workflow advantages, or customer relationships that compound. The ones that failed were expensive wrappers with good marketing.
Video creation is one area where some AI tools have carved out a real business. InVideo AI built a workflow that combines scripting, editing, and production into something creators actually pay for month after month. That is the difference between a product and a demo. The graveyard is full of demos.
What This Means for You
Here is what I would do right now if I were building or investing in AI.
First, avoid anything that is just a wrapper. If your product’s core value is making a big model easier to use for one specific task, you have a six to twelve month window before OpenAI ships that feature natively. I have watched this play out enough times that it is not speculation. It is a pattern.
Second, look for tools with proprietary data or workflow advantages. The companies growing in 2026 own something the big model providers do not. A specific integration. A unique dataset. A customer relationship that gets stronger over time. Those things are hard to replicate. Everything else is temporary.
Third, be careful about building your business workflow on top of a small AI startup. If the company folds or gets acquired, your entire process breaks. Check funding status, look at user growth signals, and make sure the core feature cannot be replicated by a ChatGPT update in six months. AppSumo regularly features lifetime deals on AI tools that have already proven product market fit, which lowers the risk that you are betting on something that will not exist next year.
Every hour you spend building on a dying platform is an hour you lose when it shuts down. The graveyard is expensive for users too, not just founders.
The Bottom Line
The AI graveyard will keep growing. The money is more selective now but that has not stopped bad bets from getting made. Every company that raised on an AI demo in 2023 is facing a reckoning in 2026. Most will not make it. The ones that do will own something real and defensible. The rest will join the list. I’ll keep updating it.
Frequently Asked Questions
What is the AI graveyard?
The AI graveyard refers to the growing list of AI projects, products, and startups that have shut down, sold off assets at a loss, or pivoted so dramatically that the original product no longer exists. It covers everything from massive enterprise bets like IBM Watson Health to well funded consumer apps that ran out of money or relevance.
Why do so many AI startups fail?
Most fail because they built on top of foundation models without creating any proprietary advantage. When OpenAI or Anthropic ships a similar capability natively, the smaller tool becomes redundant overnight. Burning through compute costs before reaching profitability is the other common cause. Both problems compound fast when investor patience runs out.
Which major AI projects have shut down or failed?
IBM Watson Health, Meta’s Galactica, and Inflection AI’s original product are among the highest-profile failures since 2022. Amazon’s Alexa division suffered massive losses and layoffs. Dozens of smaller AI writing and image tools also shut down or pivoted after larger players made their core features free or cheap.
Is AI still worth investing in after all these failures?
Yes, but with sharper criteria. The winners own proprietary data, specific integrations, or customer relationships that compound over time. Single use wrapper tools with no moat are not worth the risk in 2026. The sector is separating fast between real businesses and expensive demos.
How can I avoid building on dying AI tools?
Check funding status, growth signals, and whether the core feature can be replicated by a major model provider update. If the tool’s value disappears the moment OpenAI ships something new, it is not infrastructure. It is a dependency you cannot afford to build on.
