How Inconceivable Startups Happen
Why the best startups seem impossible at first, and how founders learn to see what others can’t.
Before they start companies, the best founders have already done something most people would never even attempt.
Immigration is the canonical example. For most people, moving to a new country where you know no one and may not even speak the language is impossible to imagine, let alone do. For many founders, this was simply the first of many inconceivable acts.
Doing the inconceivable compounds.
Once you do something inconceivable and nothing bad happens, the next inconceivable thing feels less scary. Founders who have done the inconceivable once are often capable of doing it again, while those who followed a more conventional path may struggle to evolve when challenges inevitably appear.
Another example of an inconceivable act is striving to become the best in the world at something. This requires an insane level of determination and self-belief. You need to become a learn-it-all — and once you’ve done that in one domain, you can often do it again in another.
Let’s consider two examples of inconceivable startups: Airbnb (a canonical example) and Cerebras (a more recent case study). Airbnb is one of the most inconceivable startups ever. The founders famously had to sell cereal boxes to pay the bills. When COVID hit and the business briefly went to zero, the Airbnb team navigated the crisis like founders who had already done the inconceivable once before. Cerebras, which is going public soon, is a more recent example. The founding pitch was completely inconceivable: a wafer-scale chip. Now, Cerebras is one of the only chip companies of its generation that is successfully riding the AI wave.
Of course, not every startup begins this way. There is no shortage of conceivable startups. “X for Y” and “X but better” are canonical formulations, and many of these companies succeed. For example, most of the great SaaS companies were formulated as “on-premise business X with TAM Y, now in the cloud.” These companies produced billions of dollars in value for founders and investors.
Who starts inconceivable companies?
A question I’ve spent a lot of time pondering is who starts inconceivable companies, and what allows them to believe that the vision is possible? Here is an incomplete list of ways that inconceivable startups can come together:
Knowing something others don’t: The most obvious case where a founder is able to start a business that seems inconceivable to others is when they know something that others don’t know. For example, when Weights & Biases was founded in 2017, it was inconceivable to most people that deep learning would be a big enough category to support a large DevOps business. The founders had previously built Crowdflower and were deeply embedded in the AI ecosystem. This allowed them to see something that the rest of the market didn’t — that deep learning was getting much more powerful, quickly. This insight allowed them to build one of the first generation-defining AI companies.
Believing that it must happen: There is a second case of founders who will their vision into existence, because they cannot bear to see a world where their startup does not exist. There are two sub-cases of this genre: One is that of inevitability (“someone is going to do it, so it might as well be me”) and the other version is that of mission-alignment (“I need to see a world where this exists, and no one else is doing it so I will”). Two examples of this type of startup are OpenAI (the former sub-genre) and Anduril (the latter). In both cases, what was conceivable to the founders was inconceivable to the outside world, for quite awhile.
Experimentation: More inconceivable ideas come from experimentation than people realize: they are discovered, rather than searched for. Before Stable Diffusion, video generation with AI was nascent and it was hard to see that it was even possible. But the RunwayML founders were obsessed with video as an artform — in fact, if they hadn’t started the company, I think they’d be artists — and they were creatively experimenting with a number of different ideas around the video space. When they landed on Stable Diffusion, they changed the industry.
One interesting element of inconceivable ideas is how quickly we forget that they were ever inconceivable. Once an idea becomes conceivable, dozens of people suddenly claim it was obvious. We literally forget what the world was like before, how precarious the line was between “before” and “after,” and how few people were truly on the critical path.
Once an idea works, we quickly rewrite history and call it inevitable (“someone would have done it,” “the time was ripe.”)
Our assimilative instinct is important, because it allows us to see ourselves as being part of the creative act, even though we weren’t. But this leads us to constantly underestimate the progenitors of inconceivable concepts (and overestimate the professional managers and copycats who come after). And as a result, we are constantly surprised (and delighted) when the “inconceivable underdog” comes back and sets things right — Apple under Steve Jobs, Cerebras coming back from the brink.
Divergent vs. convergent thinking
We have talked a lot about successful companies that were originally inconceivable. What about a new company that gets started tomorrow?
One thing I look for in founders at the seed stage is the ability to think divergently. Not all startup ideas are inconceivable, but somewhere along the spectrum from “completely inconceivable” to “X but better” there is an axis that I will call “divergence.” Divergent thinking is the ability to think for yourself. Just like people who do the inconceivable are likely to do so again, people who think divergently rarely do so only once. In fact, many of the best founders are too divergent — they try to reinvent the wheel for everything (even things that don’t need reinventing!).
There is a tension that I often see between divergent thinking (essential at the beginning of a company) and convergent thinking (essential for scaling a business). You need divergent ideas to come up with something new, but you need convergent thinking to figure out how to concretely instantiate that idea in reality. Many divergent thinkers don’t make good founders because they are too divergent — six months into building their business, they get sidetracked with another good idea, and simply cannot follow through. This is intolerable to most of the people who work for them, and prevents them from attracting high-quality talent.
This tension is resolvable for founders with a strong sense of purpose and mission: For divergent thinkers with a transcendent vision, the mission becomes a constraining force. At some point (usually post-PMF), successfully instantiating their vision onto the world becomes the paramount, most important thing, and they turn their creative energy toward that. The founder reins in their divergent instincts for the sake of the goal. In crucible moments, this divergent energy reappears — it bursts out from active suppression — allowing these founders to make the right decisions when it matters most.
Fast following the bit flip
It only takes one example of the inconceivable to flip the bit.
This is why you get pockets of people that end up being really successful in highly correlated ways. For example, after the Cursor founders proved that early 20s AI-native founders were especially well-suited to take down huge incumbents (in their case, Microsoft), a flood of smart people decided to do the same thing. And it worked! There is value to being the first person to see an inconceivable thing get bit-flipped, and then fast following. This is even true at super localized levels — it’s not just that Harvard and MIT kids started taking more risk after that, but even within Horace Mann (the high school that the founders went to), there are now 5 or 6 really good companies within a 1-2 year age gap.
The localization of these dynamics is pretty incredible: Horace Mann has more good startups right now than every other New York City high school combined.
In my own career, I’ve found this algorithm to be highly effective, and I would suggest that it works well in all industries, not just technology, and not just for founders. The algorithm is to carefully watch successful people around you, try to understand the thing they are the best in the world at, and then to assimilate this into your own strategy. Usually there is something that is inconceivable to you about something they are doing — and it’s exactly that inconceivable thing that you want to unpack and copy. Naturally, this also suggests that you should go to wherever the most generative and creative people are, defined as people who are actively “bit flipping.” This is the allure of Silicon Valley.
Training your mind to see the inconceivable
The best way to train your mind to see the inconceivable is simply to do inconceivable things in your own life and see that it’s not so bad.
There is a second best way, though, and this lies in the realm of ideas. Indeed, many of the biggest “inconceivable” breakthroughs in human history lie in this realm — the advent of monotheism, agriculture, fiat currency, not to mention breakthroughs in math, physics, chemistry and biology were all at one point inconceivable.
There is a famous anecdote of Winston Churchill (from before the age of 25), where he would go back through old parliamentary speeches and debates and think about what he would have said if he was there. History gives us many chances to do this.
Charlie Munger had a deeper take on this. In Poor Charlie’s Almanac, he argues that there really are fewer than 100 big ideas in all of history (evolution, relativity, etc). He argues that we should study each of these breakthroughs across domains, to understand all of the “inconceivable” ideas that society has now thoroughly integrated, in order to better understand the world around us.
In the age of Twitter, deep reading has become a niche pursuit. But for all the AI innovation, no one has invented a better technology than the book to transport oneself into another time and place, into the mind of another. Only by seeing how others think (or thought) can we start to see our own unacknowledged biases.
What we find inconceivable often reveals our blind spots.
Investing in the inconceivable
Venture capital itself was once inconceivable. When it first emerged in the 1960s, everyone went to work for big companies. No one started their own company. The idea that the “traitorous eight” would leave their jobs in 1957 to do a startup was preposterous. Now, sixty or so years later, venture capital is abundant.
The last decade of venture capital has seen the industry assume a mammoth scale. A new inconceivable frontier has been broken: the frontier of the $1Tn company. The Magnificent 7 are all Silicon Valley companies. Amazon, Microsoft, Meta, Google, Apple, Tesla and Nvidia now represent one third of the S&P 500 and all were once “startups.” With the rising size of outcomes, the dollars at play have risen too: Now, it is not uncommon to see private companies raising $1B+ rounds.
But capital access for truly inconceivable ideas has not increased by nearly this amount. Investors claim they want the next big thing. In practice, they usually fund things that already make sense. Almost all of the growth in the capital markets has accrued to the universe of conceivable ideas, because of two deeply entrenched commandments in technology investing: 1) “The big get bigger in technology,” and, 2) “Technology markets are bigger than people think.”
No matter how much evidence accumulates that inconceivable startups led by tenacious entrepreneurs make wonderful investments, truly inconceivable ideas still scare most investors. Investors are psychological creatures with their own biases. The fear of being wrong is hard-wired into the human psyche: as long as humans continue to seek pleasure and avoid pain, they will struggle to invest in crazy visionaries. It will always be easier to invest capital behind the “sure thing.”
As long as humans fear being wrong, there will always be returns in backing the inconceivable.
If you have an inconceivable idea, please email me at dcahn@sequoiacap.com. I’d love to chat about it.
When I hear the word "inconceivable," the first thing I think of is the movie, "Princess Bride." Inigo Montoya tells Vizzini "You keep using that word. I do not think it means what you think it means." Well, that's how it works with innovators. They don't know the meaning of the word "inconceivable," because innovation involves conceiving what was formerly inconceivable.
At Reliath AI we get told all of the time that our model's inability to hallucinate is "inconceivable." What makes it "conceivable" is that we don't rely on LLMs, which must hallucinate.
Creative and critical thinking both start with the same question, "How could things be different?" It is that question, more than genius, that makes innovation possible. When it does happen, it usually opens as many doors as it closes. Users are able to do new things that were once "inconceivable."
The scaling hypothesis that was once widely thought to eventually lead to general intelligence has now been more widely recognized as "inconceivable." Token guessing simply does not have the representational power to provide the desired intelligence, only to mimic it in limited circumstances. New architectures will be needed. Somehow those new architectures need to jump from "inconceivable" to "conceivable." Recognizing them in their early days may be a challenge, but failure to identify new architectures will inevitably lead to yet another AI winter.
David Cahn — there's a structural problem hiding inside your divergent-to-convergent transition.
You write: "Once an idea works, we quickly rewrite history and call it inevitable. We literally forget what the world was like before."
That forgetting isn't just psychological. It's an infrastructure failure.
The founder's original reasoning — why THIS approach when everyone said it was crazy, what alternatives were rejected, what assumptions had to hold — is the most valuable reasoning the company will ever produce. It has no system of record. It lives in the founder's head.
This matters most at your transition point: when divergent thinking gives way to convergent execution. The people hired to scale never had access to the original divergent reasoning. They inherit the WHAT without the WHY.
So when a crucible moment arrives — as Airbnb faced during COVID — the company depends entirely on the founder still being in the room.
I've been building infrastructure for this: Judgment Capital Management. The reasoning behind consequential decisions is the most underleveraged asset in every scaling company, and it requires its own system of record.
Your essay describes why founders see what others can't. The question I'd add: what preserves that seeing after the founder is no longer the one looking?