The AI Graveyard: 99 Startups Dead, US$ 15 Billion Burned
What looked like a golden age of innovation is now producing its own obituaries. The AI Graveyard, an independent database tracking artificial intelligence products that have ceased to exist, has reached 99 obituary entries with $15.4 billion in tracked capital. The average lifespan is 5.5 years. And the death rate is accelerating.
The numbers don't lie
The AI Graveyard published its consolidated data in June 2026: 99 companies that died outright, plus 109 that were acquired (and many discontinued shortly after). The total of 208 entries represents explosive growth — 21 entries in 2023, 46 in 2024, and 86 in 2025, more than quadrupling in two years. If this pace holds, 2026 could close with over 100 new entries, something that has never happened in the industry's history.
The ToolDirectory.ai report, based on 208 entries registered by July 2026, identifies clear patterns. The most common cause of death is not a dramatic shutdown announcement, nor a failed fundraising round. It's simply an expired domain — 24% of all graveyard entries vanished without a trace. Of products that died rather than being acquired, 51% let their domain lapse silently, with no shutdown announcement and no farewell message to users.
Who are the dead
The largest obituaries by invested capital are surprisingly familiar to anyone following the sector:
1. Argo AI (Self-Driving) — $3.6 billion. Founded with Ford and Volkswagen backing, it was the largest capital burnout in the sector. 2. Magic Leap (Consumer) — $2.6 billion. The mixed reality hype collapsed when consumers didn't buy in. 3. TuSimple (Self-Driving) — $1.6 billion. Truck autonomy that never took off. 4. Babylon Health (Health) — $1.1 billion. Telemedicine with AI diagnosis that lost its UK license. 5. Olive AI (Health) — $902 million. Hospital automation that couldn't prove consistent ROI.
But what stands out in the AI graveyard isn't the fallen giants — it's the smaller startups. Builder.ai, which reached a $1.5 billion valuation and raised $445 million, entered insolvency in May 2025. Adept AI, with $415 million raised and backing from investors including Microsoft and Qatar Investment Authority, peaked at a $1.5 billion valuation and then collapsed. These are companies that had money, teams, and attention — and still didn't survive.
The causes of death
The ToolDirectory.ai report mapped three dominant causes:
The wrapper undone by the platform. When OpenAI launches a new feature inside ChatGPT, dozens of startups doing exactly that lose their competitive advantage instantly. The classic example is Phind, an answer engine for developers that raised about $10 million in December 2025 and was swallowed when ChatGPT itself began offering similar functionality.
The enterprise adoption failure. Enterprise "agent" pilots often turned out to be disguised chatbots. When CTOs realize they paid dearly for automation that doesn't scale, the next quarter's sales pipeline fills with objections.
The infrastructure fragility. The Composio analysis, cited by AgentMarketCap, called it "the integration failure": it's not the LLM's performance that fails, it's the fragile chain of integrations around the model that prevents consistent value delivery.
What AgentMarketCap found
AgentMarketCap specifically mapped the AI agents segment — the most hyped category at the moment. Of the 14,000 new AI startups launched in 2024, 3,800 closed in 2025 (27%). Another 1,800 closed in early 2026, adding another 13 percentage points. The 24-month failure rate reaches 40% in companies with real venture capital and teams of 10 to 100 people.
Total capital lost in the agents segment exceeds $5 billion. For comparison, the global agents market is growing — from $5.25 billion in 2024 to $7.84 billion in 2025 — but the bifurcation is stark: survivors are consolidated in infrastructure or with defensible distribution, and those trying to build intermediate layers are being eliminated.
Does this mean AI is going to fail?
Absolutely not. It means AI is maturing.
The pattern has been observed in other industries. The 2000 internet bubble killed pets.com and boo.com but left Amazon and eBay. The telecommunications bubble destroyed dozens of operators, but the fiber and 5G infrastructure keeps growing. AI is no different.
What we're seeing is not a sector collapse, but the cleanup of the "hard to build, hard to defend" middle — that layer of wrappers and shallow integrations that proliferated when the barrier to entry was virtually zero. When frontier models absorb features natively, startups that only wrapped the API lose their differentiation.
The dead leave behind valuable lessons: the importance of defensible distribution, technical moats, use cases that can't be replicated with a single API call. And the acquirers — NVIDIA with eight purchases and OpenAI with seven, according to the AI Graveyard — keep buying talent and technology from those who failed.
The question that remains: of the startups still standing, how many will survive the winter of 2027?
Sources: AI Graveyard — By the numbers, ToolDirectory.ai — The AI graveyard report: what kills AI tools (2026), AgentMarketCap — The AI Agent Down-Round Signal
✓ Independent sources cross-checked and verified before publishing
