Build Small, Go Deep, Charge a Lot: a16z's Anish Acharya on "Narrow Startups" and the $1,000-a-Month Plan
EO Korea
Anish Acharya is a general partner at Andreessen Horowitz who invests from the firm's AI apps fund, which covers both consumer and enterprise companies. He describes himself as an engineer and product person who writes a lot of code in his free time. His argument in this session runs against a long-standing startup instinct. Founders have usually been taught to chase big markets and mass adoption. Acharya argues that AI has changed the economics, so a founder can now win by building a deeply specialized product for a relatively small number of customers and charging them a lot. He calls these companies "narrow startups." The question he wants every founder to ask is: what is the thousand-dollar-a-month version of our product?
What the Early AI Adoption Data Showed
Acharya says the idea grew out of watching how people adopted AI products. He treats customer acquisition cost (CAC) as a kind of subsidy. If a company has to spend money to get people to try a product, the customer isn't motivated enough to try it alone. Organic adoption is the reverse: customers are excited enough to try something with no incentive.
In his account, ChatGPT, Midjourney, and a number of other early AI products were adopted almost entirely organically. He says that was different from what he had seen in consumer product adoption for roughly ten years.
He also points to two findings in the early data on willingness to pay. First, a large share of people were willing to pay at all. Second, AI companies quickly blew through what investors had assumed were ceilings on subscription prices.
Acharya credits this less to foresight than to necessity. AI companies carry non-trivial cost of goods sold, and in categories like video generation those costs can be very high. Because the costs were real, these companies had to charge real money, and to deliver the best generations they had to charge a lot. What many of them found, he says, was that customers kept paying as prices rose and in some cases wanted to pay more.
Taking the Idea to Its Extreme
Acharya says he likes to push ideas to their extreme as a way of finding the core of his thinking. If people will pay high prices for software, he sees two consequences.
The first is that a software company can reach real revenue scale with relatively few customers. He does the arithmetic: 41,000 customers paying $200 a month produces a $100 million annual run rate.
The second is that software should, over time, absorb almost every part of consumer spending, and more of those dollars will be captured by AI and software products. He calls this a very optimistic prediction and says it is one "we've seen come true." More individuals will be able to build large AI companies, and consumers will have more of their needs met through software.
"There Are No Marketing Problems, Only Product Problems"
This is the line Acharya comes back to most often. He says he doesn't think products should have CAC today. In his view, needing significant acquisition spend means the product hasn't delivered enough.
He gives two reasons he thinks products can now go much deeper than before.
The models do new kinds of work. He argues that for about 40 years, computing extended the intellectual parts of the human brain and of society, which was only one aspect of human experience. What he calls "subjective creative computers" can now reach the non-deterministic side of life: emotions, relationships, the desire for self-expression, and creative work. He considers this arguably the larger part of human experience, and says it was simply not addressable by technology before.
Building software is far cheaper. AI coding and related tools are collapsing the cost and difficulty of making software. A small team can build much more, at what he describes as an order of magnitude lower cost. His summary is that founders can now "go deep or go home" instead of "go big or go home."
Defining the Narrow Startup
A narrow startup, as Acharya defines it, builds an "incredibly opinionated," deep product and charges very high prices to a relatively small number of people. He repeats the 41,000 × $200/month = $100 million run-rate calculation as the basic math.
He points to the top tiers of major AI products as precedent. He cites Google's Ultra top SKU at $250 a month, Grok at $300 a month, and OpenAI's at $200 a month, and says he believes Anthropic's is also $200 a month. According to him, consumers are buying these organically at high prices, and he says they have repeatedly received the value they expected. His slogan for the concept is "build small, go deep, and charge a lot."
How Narrow Startups Compete with the Labs and Big Tech
Acharya lays out several ways a narrow startup can defend itself against large AI labs and big tech companies.
Specialization as a moat. Because the technology lets a company go much deeper for a specific customer, and building software is so much cheaper, a startup can become so specialized that it is hard to compete with. A rival, he suggests, might have to build three years of roadmap just to catch up. He calls this differentiation taken to an extreme.
Rich software ecosystems the labs may not prioritize. ChatGPT is trying to do many things, and Acharya says he doesn't know where categories requiring a large surrounding software ecosystem will fall on its priority list. His example is meeting recorders. Many products now take notes by transcribing speech to text. To fully capture the value, though, he thinks a company probably needs a whole office suite: spreadsheets, word processors, a diary app, a notes app, and more. It isn't obvious to him that the labs will build all of that.
Being multi-model. In categories like AI coding, it helps to use models from Anthropic, OpenAI, and Google together. A product built inside OpenAI, he notes, will never use Google's models. Independence across models is therefore a competitive advantage.
Over-delivering, and charging for it. He points to moments when a product like Cursor, asked to generate a feature, returns something better than the user had imagined. He argues that these products can exceed expectations and can also charge for doing so. When a model has to "think really hard" to produce an extraordinary result, it is expensive, and he says "that is the way that it should be."
Silver Bullets Versus Lead Bullets
Acharya concedes that distribution matters most when a product cannot be 100x better than the alternatives. But he says founders sometimes tell themselves a lie: that something incrementally better is 100x better.
He frames this as silver bullets versus lead bullets. A silver bullet is a dramatic improvement. Lead bullets are small incremental improvements, and he says 10, 50, or even 100 of them never add up to one silver bullet. What matters is the 100x leap in value.
His claim about the present is that the available models have left us "awash in silver bullets," with silver bullets everywhere. So with today's technology, he believes a company can win by having a better product.
Why He Doesn't Think About TAM
Acharya calls predicting total addressable market (TAM) a "fool's errand" and says it is a common source of failure for investors and founders alike.
He illustrates this with his own history. As a first-time founder, he believed a company needed a big TAM, though he admits he "wasn't even quite sure what TAM meant." It seemed important, and bigger seemed better. That led him to consider markets like healthcare and disease management, where he had neither knowledge nor energy.
His successful product came from building something he personally wanted to exist and cared about: social graphs and mobile games, built with his co-founder. When the iPhone App Store launched, he recalls, there were about 6 million iPhones in the world, which is "not much of a TAM." They built there anyway because the market seemed to be growing fast and they had a lot of energy for it, and he says the bet proved right.
Today he says he thinks about TAM very little. He focuses instead on the value delivered to the customer and the price they are willing to pay.
The Thousand-Dollar Question
This leads to what Acharya calls the most useful prompt for founders right now: what is the $1,000-a-month SKU of our product? He breaks it into follow-up questions:
- What would an extraordinarily expensive version need to do?
- Does the product do that today?
- Would people pay for it?
- Has the team tested it?
If a founder finds customers willing to pay dramatic prices, Acharya says they are probably on the right track. If a founder has a free product and has to pay people to try it, they are probably on the wrong one. He considers this a more useful signal than TAM, because customers who pay are typically getting value. He acknowledges that retention and CAC sit upstream and can be measured.
He also stresses building where you have strong intuition, so that you "feel the feelings." A founder in that position talks to customers, may be a customer, or understands their pain points well.
He lists qualitative signs of product-market fit. One is that customers have more roadmap ideas than the team does. The overriding sign is that the team simply can't keep up with everything happening. He cites Marc Andreessen's well-known description of product-market fit as the market pulling the product out of you, "often violently."
Psychological Traps for Founders
Acharya describes several traps he says he fell into himself as a founder.
Talking yourself into product-market fit. If you have to talk yourself into having it, you don't have it. He calls this incredibly important.
Hunting for metrics that justify fit. Founders can obsess over what counts as good retention or good CAC, and he says this is often unproductive. A business has physics. If you lose 90% of your customers by the end of year one, it is very hard to build something that works, even if that figure is best in class for the category. He recommends thinking about business health from first principles rather than through frameworks.
The power-user trap. It is great that power users get so much value, but if the company doesn't capture that value, each one still counts as a single dot on the growth chart. He says founders must choose one of two paths. They can build for power users and capture the value they create, which is the narrow startup approach. Or they can build for a mass market without treating a few very happy power users as a substitute for broad market fit.
"Start It Now"
Acharya's closing advice restates his central line: there are no marketing problems, only product problems. He tells founders to be insanely ambitious on product, raise prices, adjust based on what customers say, and worry less about business books and frameworks. They should build for a small number of people, charge a lot, and go deep, and he says they will more likely than not find success.
He places this on a short horizon. He calls it not a 20-, 30-, or 50-year idea but "a 3, 5, 7-year idea," and describes it as what "the abundance agenda" means. He says it is arriving now because there is both abundant capital and dramatic consumer interest in these new products. He ends with a direct recommendation: if you were ever going to start a company, start it now. In his words, this is "the best time I've seen in my entire career by a long shot."
I think that when the product you deliver cannot be a 100x better than everything else, of course distribution is what matters. And I think the lie that we have sometimes told ourselves as founders is that something that's incrementally better is a 100x better. You know, I think of this as silver bullets versus lead bullets. You know, one silver bullet is a dramatic improvement. Many lead bullets are many small incremental improvements. 10 or 50 or even 100 small improvements, 10 or 50 or 100 lead bullets never equal a silver bullet. You really need that 100x value leap. Now, with the models that we have access to, we are awash in silver bullets, right? There are silver bullets everywhere. So, I actually do think in this day and age with the technologies we have access to, you can win by betting by having a better product.
I'm Anish. I'm a general partner at Andreessen Horowitz. I invest out of our AI apps fund. That means consumer and enterprise. For consumer, we love to invest in companies that are weird and working. For enterprise, we love to invest in companies that are working, maybe less weird. Could be weird. No judgement. I personally am an engineer and product person. I write a lot of code in my free time and, you know, I feel like we're living in the age of miracles. So, if you're building, I want to hear from you.
Yeah, I've been thinking about this for some time. You know, if you look at just the broad trend in AI in terms of how many people are first trying new AI products without being paid to, because I think of customer acquisition cost as a form of subsidy. You know, the customer is not motivated enough to do it on their own. The company really has to push them to try the new product.
And the magic of organic product adoption is that the customer is excited enough to just try it with no further incentive. So, the first thing that we really saw was the uptake of ChatGPT and Midjourney and a number of other very early AI products. All of the traffic was organic, which was different from what we had seen in consumer product adoption for maybe 10 years.
Looking at the early data around willingness to pay, what we saw was two interesting things. One was that a lot of people were willing to pay. So, high number of people that were willing to pay. And the second that the AI companies quickly blew through what we thought were the ceilings on ability to pay or the sort of amount that a customer would pay for a subscription. There's actually interesting reason for that. I love to give our AI companies credit than say it was foresight or experimentation. But, the truth is the COGS for AI companies is non-trivial, right? It can actually be very, very high, especially for products like video generation. And because you had real costs in these businesses, they had to charge customers real money. And to deliver the very best product experiences and generations, they had to charge a lot of money. And what many of these AI companies found is that even as they raised prices, customers were willing to pay and in fact wanted to pay more.
So, that really got me thinking about hey, what is the extreme version of this? And I love exploring ideas in their extreme. I just think it's a very useful way to extract the kind of core of your thinking. In the extreme of, you know, people being willing to pay high prices, there's two actual applications. The first is that you can build a software company with real revenue scale with very few customers on a relative basis, right? 41,000 for the $100 million run rate at $200 a month. And the second is that software should subsume almost every part of a consumer spend over time. And increasingly those dollars are going to be captured by AI and by software products. So, I think it's actually a very, very optimistic prediction, one that we've seen come true, which is that more individuals will be able to build large-scale AI companies and consumers will have more of their needs met through software.
I would say in the world that we're living in, there are no marketing problems, there are only product problems. I don't think products should have CAC today. And if you need significant customer acquisition costs, that means you haven't sufficiently delivered on the product. The truth is that founders and companies and products were never able to deliver with the kind of ambition that they can deliver today.
You can just go insanely deep. And part of that is because the models can do things that they could never do before. You know, we had 40 years of building models that enabled or extended the intellectual parts of our brain and the intellectual parts of our society, right? But that was only a single sort of aspect of the human experience and really of our civilization society that was addressable by technology.
Now, with these new subjective creative computers, we can address the entire non-deterministic part of human society and the human experience. That is our emotions, our relationships, our desire for self-expression, the creative work that we do. So, this entire part, arguably a larger part of the human experience, is now addressable through technology. And that simply wasn't the case before. So, I think that's one important point.
The second is that thanks to AI code and the like collapsing costs and difficulty of making software, it's just way easier for a small number of people to build a lot more. So, this simply was not possible prior, and now you have founders that can do more things at an order of magnitude lower cost. And as a result, you can go deep or go home instead of going big or go home.
So, narrow startups are companies that build incredibly opinionated deep products, charge very high prices for a relatively small number of people. You know, the simple math is that charging 41,000 people $200 a month is a $100 million run rate business. And there's a lot of precedent for this already occurring. You know, we see Google Ultra's top SKU is 250 a month, Grok is 300 a month, OpenAI's is 200 a month, I believe Anthropic's is 200 a month. Consumers are flocking to these products organically. They're paying high prices for them, and over and over again we're seeing them delivered the value that they expected.
So, the whole idea behind narrow startups is build small, go deep, and charge a lot. I think specialization is a new moat. I think that you have the ability to go so much deeper with the new technology and the collapsing cost of software creation for an individual customer that you can just be so much more specialized for that customer that it's hard to compete with. You know, somebody's going to have to build 3 years of road map to have a competitive product. So, it's simply differentiation taken to an extreme degree. I think that's an interesting and important form of a moat, which is particularly relevant to narrow startup.
I think the second is if you look at ChatGPT, they're trying to do a lot of things, and if you think about areas in which there's a really rich software ecosystem that has to be built to really capture the value, I don't know where that's going to fall on their priority list. A great example is meeting recorders. There's many products that now take notes for you by transcribing speech to text. That is great, but to fully capture the value you probably need to build a whole office suite. You need spreadsheets, you need word processors, you need a diary app and a notes app, and you need all kinds of software. It's just not obvious to me that the labs are going to actually get to that. So, I do think that building a rich software ecosystem, a rich product ecosystem is another way to compete.
Okay, I think the third thing is that there are many product categories like AI code, where you benefit from using many models. Right, it's better to be able to use Anthropic and OpenAI and Google's models. And if you're at OpenAI, you're never going to be able to build a product that also uses Google's models. So, being multi-model is a way to compete with the labs and big tech.
The other important point is that when these products over deliver for their customers, and they can. If you ask Cursor to help you generate a feature with a model, sometimes it was "Wow, this was even better than what I had hoped for or what I had imagined." So, one, the fact that these products can actually have those attributes and can over-deliver on the customer's expectations, but the second is that they can charge for it. You know, sometimes the model has to think really hard to deliver that extraordinary outcome, and guess what? When it does, it's expensive, and that is the way that it should be.
I think that when the product you deliver cannot be a 100x better than everything else, of course distribution is what matters. And I think the lie that we have sometimes told ourselves as founders is that something that's incrementally better is a 100x better. You know, I think of this as silver bullets versus lead bullets. One silver bullet is a dramatic improvement. Many lead bullets are many small incremental improvements, like 10 or 50 or even 100 small improvements, 10 or 50 or 100 lead bullets never equal a silver bullet. You really need that 100x value leap. Now, with the models that we have access to, we are awash in silver bullets. That's right. There are silver bullets everywhere. So, I actually do think in this day and age with the technologies we have access to, you can win by betting by having a better product.
Predicting total addressable market is a fool's errand. It's just impossible. It's very, very difficult, and it's a common source of failure for investors, certainly, but even for founders. When I was a first-time founder, I had this big-brain way thinking of, you know, products, which is, "Hey, we need a big market. It needs to have a big TAM." I wasn't even quite sure what TAM meant, but it seemed important, and I know you needed a big one. A big one is better than a small one. And that's why a lot of my early thinking was in markets like healthcare and, you know, in disease management, and I just didn't know anything about those markets, nor did I have energy for those markets. You know, how I built a successful product was building something that I wanted to see exist, and I was personally passionate about, which was sort of social graphs and mobile games, and that's what me and my founder built. Like, when the iPhone App Store was released, there were 6 million iPhones in the world. Like, that's not much of a TAM. But, we built there because it felt like it was growing quickly, and we had a lot of energy for the market, and we bet on, you know, perhaps not even thinking about the TAM, and we were right.
So, I don't think about TAM very much at all. I do think about value delivered to the customer and the, you know, price they're willing to pay. So, I think the most useful prompt for a founder right now is what is the thousand dollar a month SKU of our product. Right? That is the direction we need to be thinking about. Like, what is the extraordinarily expensive? What would the product need to do? Does it do it today? Would people be willing to pay? Have we tested it? So, I think if you find customers that would be willing to pay dramatic prices for your product, you're probably on the right track. You know, conversely, if you have a free product that you have to pay customers to try, you're probably on the wrong track. It's a much more useful signal for builders than thinking about concepts like TAM.
If people are paying for it, they're getting value, typically. Of course, the what is like upstream of that, things like retention, things like customer acquisition cost. So, these things can be measured, but this is why it's so useful to build in an area in which you have great intuition, because you just feel the feelings. You know it. You talk to the customer. Perhaps you're the customer yourself, or you have great intuition around their pain points. The customer has more ideas for your road map than you have. Like, there are a lot of qualitative signals. The most overriding signal is you simply can't keep up with everything that is happening as a result. Like, that's how you know your product market fit, as Marc famously said, the market is pulling the product out of you, often violently. That is the experience of it.
I mean, I think there are many psychological traps from being a founder. I can tell you a few of the ones that I fell prey to and experienced as a founder. So, one is trying to talk yourself into having product market fit. Like, if you have to talk yourself into it, you don't have it. I think that's incredibly important. I think the second is, and perhaps a related point, you know, you're looking for metrics that will justify this fact that you have market fit and you go crazy looking to calibrate on what's good retention, what's good CAC. Those are often not productive. Ultimately, a business has physics and if you're losing 90% of your customers at the end of year one, like even if that's best in class for the category, it's very difficult to build something that's working. So, then thinking about the sort of business health and first principles rather than frameworks is often more productive.
I think the final trap that you can often fall into is the power user trap. The power users are power users and that's great that they're getting that much value out of the product, but if you're not able to capture the value that they're getting, you know, they still only count as one dot on your growth chart. So, you really do have to either build for power users and capture the value you're creating, which is the narrow startups idea, or you need to build for a mass market and not, you know, tell yourself that having some really happy power users is a substitute for having broad market fit.
The most important piece of advice is that there is no marketing problems, there are only product problems. Be insanely ambitious on product, raise prices, adjust based on what you hear from the customer, and don't worry so much about business books, frameworks, just build for a small number of people, charge a lot, go insanely deep, and you know, more likely than not you'll find your way to success.
Like this is not a 20, 30, 50-year idea, this is like a 3, 5, 7-year idea. That's what the abundance agenda means and it's coming now because there's both abundant capital and dramatic consumer interest in these new products. You know, if you were ever going to start a company, start it now. Like if there are better and worse times and this is the best time I've seen in my entire career by a long shot.
Article published · Updated