Why 9 in 10 Startups Fail, and Why AI Startups Fail Even Faster

The Brutal Base Rate

Starting a company is one of the toughest bets in business and the odds are sobering. About 90 percent of startups end up failing. About a fifth don't survive the first year and about half die by the fifth year. Even raising money doesn't save most of them as about three-quarters of venture capital-funded startups ultimately fail despite having professional investors and millions of dollars in the bank. Success is the key.exception not the rule and everyone in the industry knows it

The Number One Killer Is Not Technology

When researchers investigate why startups die the root cause isn't bad code or weak engineering. It's building something that no one really wants. About 42 percent of startups fail because there is no real need in the market for what they created. They solve a problem that very few people have or a problem that people aren't willing to pay to solve. The second biggest killer behind nearly a third of failures is simply running out of money usually by spendingtoo quickly before finding a product that works

A Worked Example: Why the Runway Is Half What It Looks Like

Running out of money sounds like a lack of discipline. Let's do the arithmetic and it seems more like a structural trap because the usable runway is much shorter than the number the founders actually planned

Take a standard seed round. A company raises $3 million and spends $150,000 a month on salaries cloud costs and everything else. Three million divided by 150,000 is 20 months and 20 months is the number on the board

Now subtract the parts that are not yours. Raising a Series A takes time: preparing materials executing a process negotiating terms closing. Four months is a reasonable estimate and six is ​​common in a difficult market. You can't start that process with two months of cash left because investors accurately assess desperation so you need at least three months of runway when you start

Thirteen months. That's the real window in which a team has to find a problem that people will pay to solve build something that solves it and generate enough evidence for a stranger to hand them ten million dollars for it

And the burning rate is not static. Hire three people for $150,000 fully loaded and monthly spending goes from $150,000 to about $187,500. Runway drops from 20 months to 16 and the usable window from 13 months to 9. Three hires made to move faster reduce the time available to succeed by almost a third. This is precisely how a company that seems well funded runs out of money and no one in it has ever made a decision.evidently reckless

Then there's the reason the goal has been set so high. A $100 million venture fund needs to return at least three times its capital to be considered successful i.e. $300 million. If you make thirty investments and the base rate above is maintained maybe three will produce significant results so each of those three should return about $100 million. After several rounds of dilution the fund could own 10 to 15 percent upon exit meaning each winner should be worth between650 million and one billion dollars

Nothing else beats the bar. A company that could grow steadily into a profitable $30 million business is a failure within that structure so its investors will pressure it to spend faster in pursuit of an outcome that is improbable in terms of construction. So a significant part of the 90 percent failure rate is not accidental. It's the cost of a funding model that only the outliers pay and the founders who receive that money accept those odds whether they say it or not.not out loud

These are illustrative figures and each round is different. The structure is the part worth taking

Why AI Startups Are Failing Even Faster

The rise of AI has made the pattern sharper not smoother. Investors poured more than $100 billion into AI startups in 2025 and yet AI startups are failing at an even higher rate than traditional technology by some measures around 90 percent. The reasons are specific. Many AI startups are built as a thin shell on top of a core model they don't own leaving them no real moatsince a larger company can copy the feature overnight. IT costs punish and eat away at margins. And the competition is brutal because the founding model companies themselves continue to absorb the best ideas into their own products

The cost point deserves emphasis because it breaks the above arithmetic. Traditional software had near-zero marginal cost so a company that found its product market fit could increase its revenue without increasing its consumption much. Inference costs real money on each query which means that an AI product that succeeds burns faster as it grows. Success itself shortens the runway which is a genuinely new failure mode and not one around which the risk model was built

The 95 Percent Problem

There's a surprising statistic hidden behind the hype. Studies have found that about 95 percent of corporate AI pilot projects fail to generate any measurable return on investment and only about 5 percent produce a positive profit. That gap between how impressive the technology looks in a demo and how rarely it actually pays for itself in practice is exactly where most AI startups die. Building something that wows people in a demo is not whatThe same as building something that a company will continue to pay for month after month

Case Study: Quibi Had Everything Except a Customer

If the main cause of failure is building something that no one wants the strongest proof of that statement is a company that had all the other advantages in unlimited quantity

Quibi raised approximately $1.75 billion before its launch. It was founded by Jeffrey Katzenberg who had run the Disney animation studio and co-founded DreamWorks and led by Meg Whitman former CEO of eBay and Hewlett-Packard. It had the backing of every major Hollywood studio the content was about recognizable stars and a genuinely novel technical feature that allowed the video to rotate seamlessly between portrait and landscape

All the conventional explanations for a startup's failure were resolved preemptively. It didn't run out of money. It didn't have an inexperienced team. It didn't lack distribution relationships engineering talent or press attention

It launched on April 6 2020 and announced it would close in October ceasing operations in December approximately six months later. The assets were sold to Roku in early 2021 for a fraction of the proceeds

The premise was that people wanted top-quality expensively produced ten-minute episodes designed for viewing on mobile devices during commutes and other brief stretches of the day. Almost every part of that sentence turned out to be an assumption rather than an observation. People were watching short videos on phones in enormous numbers on free social algorithmically powered shareable platforms. Quibi's product was none of those things and it launched into a lockdown where no one had any commute

What makes it the ultimate case study is that the bug was readable before launch and no one could see it because the company had spent so much money that the question of whether anyone wanted the product had become impossible to ask internally. With $1.75 billion committed there is no cheap experiment left to run and no version of the answer on which anyone can act

Money is a solution to the second most common cause of startup death. Not only is it useless against the first it actively hides it

Where These Statistics Mislead

I have quoted many round numbers in this article. Most of them are more unstable than they seem and anyone who uses them should know how to do it

No one agrees on what failure means. Does a company that is acquired for less than it raised count? A team hired by a larger company while the product is closed? A company that never grows but pays its founders well for a decade? The 90 percent figure includes them all together and several of them are outcomes that a reasonable person would take

42 percent comes from the founders explaining themselves afterwards. The no market need statistic is largely drawn from autopsies written by people describing why their own company died. This is a valuable source and is a self-report compiled after knowing the outcome which is exactly the condition under which people construct orderly narratives. No market need is also the most flattering answer available as it blames the world rather than execution

The 95 percent pilot figure is a widely circulated study with a high bar. A measurable return on investment within a pilot period is a high standard for any new technology and most enterprise software would not pass the same test in its first year of implementation. The number is worth knowing and is not the indictment as it is often presented

An unparalleled failure rate means nothing. Ninety percent sounds catastrophic until you wonder what an alternative deployment of the same capital and five years would have produced. Business returns are generated from a small number of huge outcomes so the expected outcome of the model is a high failure rate rather than evidence that it doesn't work

My view is that the direction of each of these statistics is correct that their accuracy is fabricated and that the underlying lesson survives them all: demand is the binding constraint and almost no one tests it early enough

What the Survivors Do Differently

The startups that create it tend to share some unglamorous habits. They obsess over a real painful problem that customers will pay to solve rather than falling in love with their own technology. They manage cash carefully and treat their lead as the most precious resource they have. They find something defensible a network proprietary data a relationship or a workflow that a larger competitor can't simply copy. And they stay close enough to customers to change direction quickly when the first idea fails.works which normally doesn't happen

How I Would Actually Judge a Startup

I'm more of a student than an investor but I read a lot of company descriptions and have settled on a short order of questions that covers most of the presentations

I start with what the customer was doing before. If the honest answer is nothing because the problem was tolerable then I'm looking at a product that should create a habit as well as serve them and that's a much harder job than it seems. If the answer is that they were paying someone else or paying an employee to do it manually then clearly there is a market and the issue becomes competitive rather than existential

Second I calculate the usable lead from the example above rather than accepting the title. Cash divided by burn minus the fundraising process minus a cushion. That number tells me how many attempts the team actually makes

Third for anything in the form of AI I ask what is left if the underlying model vendor ships the feature natively next quarter. If the answer is nothing the company is a feature with a fundraiser attached. If the answer involves proprietary data an integration that no one wants to rebuild or a regulated workflow there is something to defend

Fourth I look for evidence that the demand question was tested early and cheaply. Quibi is my constant reminder that spending is the enemy of learning here because a company that has raised enough money can postpone the only question that matters until it is too expensive to answer

Fifth I try to notice when a founder describes the technology instead of the customer. The relationship between those two guys in any tone is the most reliable sign I've found and it costs nothing to observe

That's how I read them. It's not investment advice and I have no money in any of this

The Bottom Line

Failure is the default outcome of a startup and no amount of funding or publicity changes the underlying math. About 42 percent die because no one needed what they built about a third because they ran out of money and about three-quarters of venture-backed companies fail despite professional investors

Clue arithmetic explains more of that than the founders admit. A $3 million seed at $150,000 a month reads as 20 months and is actually about 13 once you subtract a four-month raise and a three-month cushion and three additional hires bring it down to nine. Meanwhile a $100 million fund needs a handful of winners to reach a value between $650 million and $1 billion.dollars so the model itself demands the change that makes failure likely

Quibi is the case that settles the ranking of causes. It had $1.75 billion Katzenberg and Whitman all the Hollywood studios and it disappeared in six months because none of it answers whether anyone wanted the product. The companies that beat the forecasts are rarely the ones with the flashiest technology. They are the ones that found a problem worth solving saved enough cash to survive the search and built something difficult to copy. In a year when money is flooding into AI thatold discipline matters more not less