The AI graveyard: a running list of projects and startups that didn’t make it

The Expensive Reality of AI Burnout

I have spent the last decade watching founders chase cycles. Whether it was the early days of SaaS or the chaotic heights of the crypto boom, the pattern is always the same: a frantic rush to build followed by a silent, painful culling of the herd. Right now, we are watching the AI graveyard fill up in real time. This isn't just about small teams running out of runway; we are seeing massive projects from the industry giants hit a wall that no amount of capital seems able to climb.

When we look at the list of recent failures, it is easy to mock the ambition. But for those of us actually building, these shutdowns are data points. They tell us where the limits of current models lie and where the market is losing patience with vaporware. The era of the waitlist as a product is ending, and the era of the graveyard is beginning.

The Siri Paradox

Apple has always been the master of waiting until a technology is mature before slapping a polished UI on it and claiming they invented it. But with the recent delays and setbacks in their revamped, AI-driven Siri, the strategy is showing cracks. They are struggling to bridge the gap between a voice assistant that sets timers and a true agent that understands intent. The problem isn't just code; it is the fundamental friction of trying to make a closed ecosystem play nice with generative models that are inherently unpredictable.

For builders, the Siri struggle is a warning. If a company with Apple’s cash reserves and hardware integration can’t ship a seamless agentic experience on schedule, you should be very skeptical of any pitch deck promising a universal AI assistant built by a team of three in a garage. The integration layer is where the bodies are buried.

OpenAI and the Super App Fatigue

OpenAI has been the North Star for this entire movement, but even they aren't immune to the graveyard. Their pivot away from certain consumer-facing experiments and the messy rollout of their supposed super app indicates a lack of focus. They are trying to be the platform, the model, and the interface all at once. History suggests you can usually only win at one of those at a time.

We are seeing projects inside OpenAI get shelved not because they don't work, but because they don't scale profitably. This is the pivot from research lab to corporation. When a project hits the graveyard at a place like OpenAI, it is usually a sign that the cost of compute has finally outweighed the potential for user acquisition. As a founder, you need to look at your unit economics today, not in some imaginary future where tokens are free.

Why Most AI Startups Are Features, Not Companies

The graveyard is currently littered with wrapper startups. These are the companies that built a slightly better UI for a specific prompt and called it a business. As soon as the underlying model providers (Anthropic, Google, OpenAI) updated their core product, these startups became redundant overnight. They didn't build a moat; they built a temporary bridge.

• Lack of proprietary data access.
• Over-reliance on a single API.
• High churn because the novelty wears off in 30 days.
• Zero workflow integration.

If your product can be replaced by a system prompt update from a Tier 1 lab, you aren't building a company; you're building a feature that is destined for the obituary section. I’ve said this to every founder I advise: if you don’t own the data or the distribution, you don't own the business.

The Compute Ceiling

We have to talk about the physical limits of this boom. Many projects are failing simply because they cannot get the chips or the power required to train the next iteration. The graveyard is full of ambitious teams that had the talent but couldn't secure the H100s. We are entering an era of compute inequality, where the graveyard is populated by those who were priced out of the hardware market.

This is where I get skeptical about the narrative that AI will solve everything. If the barrier to entry is a billion-dollar server farm, we are moving away from the democratic nature of the early internet and toward a feudal system. The projects dying right now are the ones that tried to play the scale game without the bankroll to back it up.

What Builders Should Do Now

The smartest people I know aren't trying to build the next LLM. They are looking at the graveyard and seeing what failed. Most failed because they were too broad. They tried to be everything to everyone. The survivors are going to be the ones who go narrow and deep. They are building for specific industries—law, construction, logistics—where the general-purpose models are too noisy or unreliable.

The graveyard is the best textbook a founder can buy. It shows you exactly where the hype met reality and lost.

Stop looking at the success stories on X (formerly Twitter). Go look at the post-mortems of the AI startups that folded this month. You will find that they almost all suffered from the same three things: high burn, low retention, and a product that was essentially a fancy skin on someone else's intellectual property.

Takeaway

The culling of AI projects is healthy, even if it’s painful for those involved. It clears out the noise and forces us to focus on utility over hype. If you are building today, your goal shouldn't be to avoid the graveyard through more funding. Your goal should be to build something so deeply integrated into a specific user's workflow that it's impossible to kill. Don't build a wrapper. Build a system.

Read the original at TechCrunch AI →