The AI graveyard report: what kills AI tools, by the numbers (2026)

We run an AI graveyard — a standing record of the AI tools that have shut down, gone dark, or been acquired. As of July 2026 it holds 199 entries, and that number is the point of this report. Most coverage of dead AI tools is anecdotal: one famous flameout, one splashy shutdown post. We have something more useful — a structured dataset of how AI tools actually die. So we ran the numbers on all 199. This is what kills an AI tool in 2026, by cause, by year, by category, and by who is doing the acquiring.

We catalog these products for a living, which means we are also the ones who watch them end. We are not neutral observers; we maintain the dataset by hand, and we will be specific about what it shows and where its edges are. For the names behind the numbers, the AI graveyard is the running record this report is built on.

Key findings

• 199 AI tools in our graveyard as of July 2026: 91 died outright (46%) and 108 were acquired (54%). For an AI tool, getting bought is now a more common ending than dying outright.
• The most common cause of death is a quietly expired domain — 24% of all entries. Among the tools that died rather than sold, 53% simply let their domain lapse with no shutdown announcement at all.
• The death rate is accelerating. Recorded deaths and acquisitions went from 20 in 2023 to 45 in 2024 to 84 in 2025 — more than quadrupling in two years.
• NVIDIA and OpenAI are tied as the most active acquirers, each absorbing seven of the tools we tracked — driven by AI-lab "reverse acquihires" (OpenAI, Cursor, Meta) and, newly, Salesforce's agentic-CRM buying spree (Qualified, Fin, and Airkit).
• The deadliest categories by raw count are Developer Tools (30), AI infrastructure (25), and AI art and image creation (16) — the most crowded, most-hyped corners of the market.

How we built this report

The dataset is the 199 tools in our graveyard — every entry whose lifecycle we have marked deceased or acquired. Each one carries, where we could establish it, a cause of death, a date, one or more categories, and (for acquisitions) the acquirer. The snapshot for this report was taken in July 2026.

Five caveats, because the method shapes the numbers and we would rather show our work than launder it:

1. This is our catalog's graveyard, not a census of every AI startup. It covers the tools we tracked closely enough to follow to the end — skewed toward products notable enough to have been listed in the first place. It is a large, hand-maintained sample, not the whole population.
2. "Domain expired" describes how a quiet death surfaces, not always the underlying business reason. When a tool stops announcing anything and its site goes dark, the first hard evidence we get is a dead domain. That is itself a finding — most tools die without a word — but read it as "went dark," not as a root-cause diagnosis.
3. Category figures are absolute death counts, not death rates. We do not publish them as a percentage of all tools in each category, because crowded categories naturally produce more deaths. "Most deaths" is not the same as "most dangerous."
4. 2026 is a partial year, and quiet deaths surface late. We often confirm a death months after the fact, so recent-year counts are floors that will rise, not final tallies.
5. Acquisition is an ending, not a verdict. We count an acquired tool as leaving the independent market; some live on inside the acquirer, many are quietly wound down. More on that below.

Bought or buried: 54% of endings are acquisitions

The first thing the data settles is a framing we have used before: in 2026 the AI market sorts its exits into bought and buried, and acquisitions now edge out outright deaths. Of 199 endings, 108 (54%) were acquisitions and 91 (46%) were outright deaths.

That is a strikingly high acquisition share for any software category. It says the AI shakeout is not only a graveyard — it is also a clearance sale. A meaningful slice of the tools that stop existing as independent products do so because someone bought the team, the technology, or the customer list, not because the lights went out. We walked through the strategic logic of that in our piece on the 2026 AI consolidation; the dataset puts a number on it.

The catch, for anyone reading an acquisition as a happy ending: a sale and a shutdown often look identical from the user's side. The product gets absorbed, the standalone tool is sunset, and the thing you were paying for is gone either way. "Acquired" is a better outcome for founders and investors than for the people who depended on the product.

Most AI tools die quietly

Here is the finding we did not expect to be so lopsided. The single most common cause of death in the graveyard is not a shutdown announcement, a pivot, or a failed fundraise. It is a domain that simply stopped resolving — 48 of 199 entries, 24% of the whole graveyard.

Narrow it to the tools that died rather than sold, and the pattern is starker: 53% of outright deaths were silent. No farewell blog post, no "thank you to our users," no wind-down email. The company stopped paying for its own website, and that was the obituary. Announced shutdowns — the kind that make the rounds on tech Twitter — account for about 44% of real deaths (40 entries).

The lesson for anyone evaluating an AI tool is uncomfortable: the market's failure mode is not drama, it is disappearance. The tool you are trialing today is far more likely to vanish without notice than to send you a graceful shutdown notice. That is exactly why vendor durability has become a real input to AI buying decisions, a theme we keep returning to in our AI capex analysis.

The death rate is accelerating

The graveyard is filling faster every year. Entries by year of death or acquisition:

The count more than doubled from 2023 to 2024 and nearly doubled again in 2025, for more than a fourfold increase across two years (20 to 84). 2026 already holds 34 entries, and because quiet deaths surface on a lag, that figure will climb well past where it sits today.

This is the downstream shadow of the funding boom. The 2023–2024 vintage of AI startups — the wave launched in the year after ChatGPT — is now hitting the point where seed runway runs out and Series A either lands or doesn't. A rising death count is what the back half of a hype cycle looks like in the data, and it lines up with what we are seeing on the human side of the same shift in where AI is actually replacing jobs.

Who is buying the survivors

Acquisitions cluster, and they cluster around infrastructure and AI labs. The two most active acquirers are tied: NVIDIA and OpenAI, with seven acquisitions each. Just behind them sits a cluster at three each — Cursor's maker Anysphere, Meta, Google, and Salesforce — then a tier of repeat buyers: Databricks, Snowflake, Amazon, Thomson Reuters, and Uniphore.

NVIDIA and OpenAI sharing the top spot is not a coincidence; it is the consolidation thesis showing up in primary data. NVIDIA — which sells the picks and shovels — buys the businesses that make its hardware run better: inference optimizers, model-tuning shops, the layer between the model and the chip. OpenAI, Cursor's Anysphere, and Meta are the other engine of 2024–26 dealmaking: the "reverse acquihire," in which an AI lab hires a startup's founders and licenses its technology while the standalone product is wound down. The remaining buyers are infrastructure platforms assembling a stack (Databricks, Snowflake, Amazon) or incumbents bolting AI onto an existing suite (Salesforce, which bought Qualified, Fin, and Airkit for its Agentforce push; Thomson Reuters in legal; Uniphore in customer experience). The money flowing into these deals comes from the layers of the market that already had revenue — or the best-funded labs.

The names behind the numbers

Statistics describe the shape of the graveyard; individual tools show how each ending actually happens. Six recent entries map cleanly onto the patterns above.

Killed by a lawsuit. ROSS Intelligence was an AI legal-research engine that answered plain-English questions with cited case law. It shut down in 2021 after Thomson Reuters sued it for training on Westlaw headnotes, and a 2025 federal ruling found it had infringed Westlaw's copyrights — a landmark decision for AI training data, and a reminder that the legal layer can end a tool as fast as the market can.

Out of money, lights off. Jibo, billed as the world's first social robot, shipped at $899 in 2017 and ran out of money the next year. In March 2019 its servers were switched off and the robots delivered a farewell message and one last dance — a literal version of the quiet death that dominates this dataset.

The pivot that abandons the product. TuSimple raised over $1B to build autonomous semi-trucks, then delisted from Nasdaq in 2024 and rebranded to CreateAI to chase AI animation and gaming in China — leaving the trucking product, and its U.S. operation, behind.

Collapse, with a criminal case attached. AllHere built the Los Angeles Unified School District's "Ed" student chatbot, then filed for Chapter 7 bankruptcy within months of its 2024 launch; its founder was later charged with defrauding investors.

The enterprise customer walks. Bossa Nova Robotics put six-foot shelf-scanning robots in roughly 500 Walmart stores — until Walmart ended the contract in 2020, deciding human workers could do the job faster and cheaper, and the company wound down.

Absorbed into something bigger. VocalIQ, a Cambridge voice-AI startup, was acquired by Apple in 2015 and folded into Siri. The standalone product never reached the market — the acquisition was the ending, exactly the "bought, then sunset" pattern that now accounts for more than half the graveyard.

These are six of the 199 entries behind this report; the full roster covers the rest.

The deadliest categories

Where do AI tools go to die? By raw count, the graveyard is concentrated in the categories that were easiest to start a company in and hardest to defend:

Read this with the caveat from our method note: these are absolute counts, not rates, and the busiest categories will always post the most deaths. But the shape is still informative. Developer tools, image generation, and content writing — the categories where a small team could ship a wrapper on a frontier model in a weekend — are exactly where the casualties pile up. Low barriers to entry cut both ways: easy to launch, brutally hard to keep a moat once the base models absorb your feature. The AI infrastructure presence on this list is a different story — that is where the acquisitions concentrate, so a high count there reflects buyouts as much as burials.

What this means for AI tool buyers in 2026

The graveyard is, in the end, a buyer's tool. Three things it should change about how you evaluate AI products:

Assume disappearance is the base rate, not the exception. With 53% of outright deaths arriving as a silently dropped domain, "will I get a heads-up before this tool goes away" is the wrong question. Build an exit plan — data export, a fallback option — into any AI tool you come to depend on, especially the small ones.

Treat acquisition as an ending, too. A 54% acquisition share means a product you adopt has a real chance of being bought and folded into something else within a couple of years. That is not necessarily bad, but it is a change you should price in for anything mission-critical.

Weight the category. A new entrant in developer tools, image generation, or content writing is launching into the most heavily populated wing of the graveyard. The product can still be excellent — but the category's track record says durability is the thing to scrutinize hardest. Our editorial review methodology leans on exactly these signals, and the Top 100 AI Tools is our running answer to which products have cleared the bar.

Frequently asked questions

How many AI tools have shut down or been acquired? In our AI graveyard, 199 as of July 2026 — 91 that died outright and 108 that were acquired. That is the count of tools we have tracked to a confirmed end; it is a large hand-maintained sample, not a census of every AI startup, and it grows every month.

What is the most common reason AI tools fail? A quiet disappearance. The single most common cause of death in our data is an expired or dead domain — 24% of all graveyard entries, and 53% of the tools that died rather than being acquired. Most AI tools do not announce a shutdown; they simply go dark when the company stops paying for its own website.

Are more AI tools dying in 2026 than before? Yes. Recorded deaths and acquisitions rose from 20 in 2023 to 45 in 2024 to 84 in 2025 — more than quadrupling in two years — as the post-ChatGPT startup wave runs out of runway. 2026 already holds 34 confirmed entries, and that number will rise as late-confirmed deaths surface.

Who is acquiring failed or struggling AI startups? Infrastructure giants and AI labs. NVIDIA and OpenAI are tied as the most active acquirers with seven buys each, followed by Cursor's maker Anysphere, Meta, Google, and Salesforce (three each). Other repeat acquirers include Databricks, Snowflake, Amazon, Thomson Reuters, and Uniphore. Many recent deals are "reverse acquihires" — a lab hires the founders and licenses the tech while the product is sunset.

Which AI categories have the most failures? By raw count, developer tools (30), AI infrastructure (25), and AI art and image creation (16) lead the graveyard. These are the most crowded, lowest-barrier categories — easy to launch in, hard to defend once frontier models absorb the feature. Note these are absolute counts, not death rates.

Is getting acquired good or bad for users? Mixed, and often it looks the same as a shutdown from the user's seat. A sale is usually a good outcome for founders and investors, but acquired products are frequently sunset or absorbed into the buyer's platform, so the standalone tool you relied on can disappear either way. Treat an acquisition as a possible ending, not a guarantee of continuity.

How do you track which AI tools are dead? We maintain lifecycle status (active, deceased, or acquired) on every tool in our catalog, with a cause of death, a date, and — for acquisitions — the acquirer. Quiet deaths are confirmed by monitoring for dead domains and missing sites; announced shutdowns and acquisitions are logged from primary reporting. The full roster lives in the AI graveyard.

Where to go next

For the names and stories behind these numbers, the AI graveyard is the running record, and our earlier roundup of the AI tools that died or got acquired walks through the notable cases. For the forces driving the trend, our AI consolidation analysis covers who is buying and why, and the AI capex bubble covers the spending that funds — and eventually starves — these companies.

The honest version: a graveyard with 199 entries and a more-than-quadrupling death rate is not a sign that AI is failing. It is a sign that AI is maturing — that the easy-to-build, hard-to-defend middle is being cleared out, while value consolidates into infrastructure and the products with real moats. The job of a buyer in 2026 is to tell which side of that line a tool is on before you depend on it.

— The ToolDirectory.AI editorial team