Last Updated on August 16, 2026 by Taya Ziv
Quick answer: In the first quarter of 2026, the wind-down platform SimpleClosure says 2.6 times as many companies closed through its service as a year earlier, and in its 2025 annual report AI companies were nearly 16 percent of everything it wound down, one of the largest single categories. Read the vendor’s numbers with a raised eyebrow, because a shutdown company profits from shutdowns. But almost every founder in that data will tell you the same cause of death: they ran out of money. That is true the way “his heart stopped” is true. It is what happened last, not what happened. The more useful cause is that a large share of these companies were features wearing the costume of a company, and the model economics quietly stripped the costume off. If you are building right now, the question is not “how much runway do I have.” It is “if the money never ran out, would this thing actually work.”
I want to do something a little morbid today. I want to read the death certificates.
Because there is a lie we all agree to tell when a startup dies, and it is a comfortable one. The founder posts the graceful thread. The investors nod. Everyone writes the same three words: ran out of money. It is clean. It is nobody’s fault, really, just the cruel math of runway and a market that turned. And it lets everyone in the room keep believing the company was basically right and just unlucky on timing.
I have watched enough of these now to tell you those three words are almost always hiding the actual body.
The number that should stop you
Here is a number worth staring at, with a warning label attached. SimpleClosure exists to shut companies down. That is the whole business, the paperwork and the asset sales and the quiet email to the cap table. It is also a venture-backed startup that makes money when more founders wind down, and in the same April interview where it disclosed its shutdown figures it launched a marketplace to resell the assets of dead companies. So a “shutdowns are surging” headline is, for them, a sales brochure. Its data is self-reported, non-audited, and drawn from its own paying customers rather than the whole startup population. Take the direction seriously and the decimal points with salt.
With that said, the direction is worth taking seriously. In the first quarter of 2026, SimpleClosure processed 2.6 times as many closures as in the same quarter of 2025. That is a real jump, though notice the platform’s own customer base was compounding across the same window, so some of that “2.6x” is just more founders finding the service, not more startups dying. Its 2025 annual report is where the mix gets interesting. AI companies were nearly a sixth of everything it wound down, one of the biggest buckets. B2B SaaS climbed from about five percent of closures to almost eight. And the detail that changes the story: Series A closures jumped from roughly six percent of all wind-downs to about fourteen, more than double year over year.
Sit with that last one, because it is the tell. A pre-seed company dying is not news, that is just the base rate of trying things. But a Series A company dying is a company that had a product, had customers, had metrics good enough to convince professional investors to write a real check. These are not failed ideas. These are failed businesses. The correction stopped eating the dreamers and started eating the ones who looked like they made it.
One honest complication, because I promised to read the whole certificate and not just the convenient lines. SimpleClosure’s own report frames 2025 as maturation rather than collapse, and AI’s share of shutdowns actually edged down from about 17.7 percent to 15.9 percent as other sectors hit their own reality checks. So this is not a runaway AI die-off. It is a market where the companies dying are older, better funded, and further along than the ones that used to die, and where AI is a large slice of that but a shrinking one by percentage. The interesting part is not the volume. It is who is now in the room.
What the coroner actually finds
So if it is not the idea and it is not the traction, what is on the slab?
Follow the wrapper companies, because that is where the pattern is clearest. It is almost always the same. A team ships something genuinely useful on top of a big model, GPT-5 or Gemini or Claude, and it works. Users show up. The demo lands. For about two quarters it looks like magic. Then three things happen at once, and none of them are on the pitch deck.
First, the margins go underwater. You are paying the model provider by the token, and you are charging your customer a flat monthly seat, and the moment usage climbs, every new power user costs you more than they pay you. You are not scaling a business, you are subsidizing an addiction. This is the exact autopsy I wrote about the AI company that looked healthy right up until the end, where the company was growing fast when it died, and the whole story was hiding in the gross margins. Growth was never the problem. Growth was the poison.
Second, the wedge disappears. The clever thing you built in a weekend, the prompt chain, the little UI on top of the API, becomes a feature the model provider ships natively in the next release. You did not get out-competed by a rival. You got absorbed by the platform you were standing on. This is the one part of the SimpleClosure report I would actually hang weight on, because it is a qualitative read anyone can check against the release notes: it found that AI wrappers and application-layer tools built quickly on commoditized models, without deep defensive moats, are facing the sharpest correction of any category, and that when infrastructure and developer-tool companies do fail, they have raised roughly twice the capital of their wrapper counterparts. Translation: the thin layer dies cheaper and faster.
Third, nobody is locked in. Your customers can leave on a Tuesday because there is no data, no workflow, no switching cost holding them. Back in February, Google Cloud’s Darren Mowry, who runs Google’s global startup org, said the quiet part out loud: wrapping “very thin intellectual property around Gemini or GPT-5” means a startup’s “check engine light” is on, and “the industry doesn’t have a lot of patience for that anymore.” He was not being cruel, and he was not writing anyone off. He named the survivors too, the wrappers with a real moat or a genuine vertical, which is exactly the escape hatch the rest of this piece is about. He was reading the same chart everyone read and choosing to say it into a microphone.
None of those three is “we ran out of money.” Money running out is just the sound the building makes when all three land at once.
Why this is the comfortable lie
Here is why the death certificate matters, and why the lie on it is dangerous.
When you write “ran out of runway,” you file the death under finance. The lesson becomes “raise more, raise sooner, keep a bigger buffer.” And so the next founder does exactly that, closes a fat round, and buys themselves eighteen more months to run the same doomed experiment at higher cost. The fat seed round does not save this company. It just funds a longer, more expensive version of the same funeral. I have said before that a bigger round is a bigger promise, not a bigger cushion, and this is the mechanism underneath that.
And the money is absolutely there. North American startups raised a record $392 billion in the first half of 2026, with AI taking roughly 80 percent of the second quarter’s dollars. Record funding and a rising death count are not a contradiction. They are the same story told from both ends. Capital is not the scarce thing right now, and a company that dies in the middle of the most generous funding market in history did not die of thirst.
Because if you write the honest cause, the death gets filed under product, and the lesson flips completely. It is not “you needed more money.” It is “you needed a reason to exist that a platform release could not delete.” Those are opposite instructions. One tells you to fundraise harder. The other tells you to go find something you own, a proprietary dataset, a workflow customers can’t rip out, a wedge into a market too annoying for a big model to bother with. Plenty of the companies in that graveyard raised fine. What they never had was the second thing.
This is also why “seventy percent of AI startups are building nothing” is not an insult, it is a diagnosis. I made that case when Google’s own numbers showed most AI startups had no defensible product underneath the demo, and the 2026 shutdown data does not contradict it so much as sit next to it: the companies without a moat are the ones getting corrected hardest, exactly as you would expect.
What to do if you’re the one still breathing
I am not writing this to dance on anyone’s grave. I am writing it because if you are building right now, you are close enough to these companies to inherit their exact ending, and the good news is the failure is legible. You can see it coming.
Run three checks this week. First, the honest-margin test: if you charged what you charge today and grew your usage ten times, would you make money or lose it faster? If the answer is “lose it faster,” you do not have a pricing problem, you have a business-model problem wearing a pricing costume, and no round fixes it. Second, the platform-release test: write down the single feature the underlying model provider would have to ship to make you pointless, and then ask how many quarters until they ship it. If the honest answer is “one,” you are not a company yet, you are a countdown. Third, the leave-on-Tuesday test: if your best customer wanted to churn tomorrow morning, what would they lose? If the answer is “nothing they’d miss,” you have usage, not lock-in, and usage evaporates the second the novelty does.
And underneath all three is the oldest test, the one none of the AI money changed. Did you ever actually prove someone would pay for this before you built the whole thing. Because the work of validating that real people will pay real money for the specific thing you’re making is boring, and it does not demo well, and it is the only thing on this list that has never once shown up on a death certificate. The companies that do it do not always win. But they almost never die of this.
The graveyard is filling up with teams that were smart, fast, and funded. Smart, fast, and funded is not a moat. It never was. It just took a market this generous to make the difference impossible to hide anymore.
Read the death certificates. Then make sure yours would have to lie about you too.
