Transcriptly

The Fastest Path to a $100M AI Business | Anish Acharya, a16z GP

EO14:18Added Aug 31, 2026

Anish Acharya is a General Partner at Andreessen Horowitz, investing in consumer and enterprise AI. His take on the classic Silicon Valley advice: go deep or go home. Not a hundred million free users, but 41,000 people paying $200 a month.

Watch on YouTube →
Contributed by 刘嘉琪

Transcript

Transcript format

  1. Intro

  2. 00:00I think that when the product you
  3. 00:02deliver cannot be 100x better than
  4. 00:05everything else, [music] of course
  5. 00:06distribution is what matters. And I
  6. 00:08think the lie that we have sometimes
  7. 00:09told ourselves as founders is that
  8. 00:11something that's incrementally better is
  9. 00:13100x better. You know, I think of this
  10. 00:15as silver bullets versus lead bullets.
  11. 00:17You know, one silver bullet is a
  12. 00:20dramatic improvement. Many lead bullets
  13. 00:22are many small incremental improvements.
  14. 00:2410 or 50 or even 100 small improvements.
  15. 00:2710 or 50 or 100 lead bullets never equal
  16. 00:29a silver bullet. You really need that
  17. 00:31100x value leap. Now with the models
  18. 00:33that we have access to, we are a wash in
  19. 00:36silver bullets, right? There are silver
  20. 00:38bullets everywhere. So I actually do
  21. 00:40think in this day and age with the
  22. 00:42technologies we have access to. You can
  23. 00:44win by betting by having a better
  24. 00:46product. [music] I'm Anish. I'm a
  25. 00:47general partner at Andre and Horowits. I
  26. 00:49invest out of our AI apps fund. That
  27. 00:51means consumer and enterprise. For
  28. 00:52consumer, we love to invest in companies
  29. 00:54that are weird and working. For
  30. 00:56enterprise, we love to invest in
  31. 00:57companies that are working maybe less
  32. 00:59weird, could be weird. [music] Uh, no
  33. 01:00judgment. Um, I personally am an
  34. 01:03engineer, a product person. I write a
  35. 01:04lot of code in my free time. [music]
  36. 01:06And, you know, I feel like we're living
  37. 01:07in the age of miracles. So, if you're
  38. 01:09building, I want to hear from you.

  39. Go Deep or Go Home

  40. 01:22Yeah, I've been thinking about this for
  41. 01:24some time. You know, if you look at just
  42. 01:26the broad trend in AI in terms of how
  43. 01:28many people are first [music] trying new
  44. 01:30AI products without being paid to
  45. 01:32because I think of customer acquisition
  46. 01:34cost as a form of subsidy. You know, the
  47. 01:36customer is not motivated enough to do
  48. 01:38it on their own. The company [music]
  49. 01:39really has to push them to try the new
  50. 01:42product and the magic of organic product
  51. 01:44adoption is that the customer is excited
  52. 01:46enough to just try it with no further
  53. 01:48incentive. I [music] think the first
  54. 01:49thing that we really saw was the uptake
  55. 01:51of chat GPT and midjourney and a number
  56. 01:53of other very early AI products. All of
  57. 01:55the traffic was organic which was
  58. 01:57different from what [music] we had seen
  59. 01:58in consumer product adoption for maybe
  60. 02:0010 years. Looking at the early data
  61. 02:03around willingness to pay. What we saw
  62. 02:05[music] was two interesting things. One
  63. 02:06was that a lot of people were willing to
  64. 02:08pay. so high number of people that were
  65. 02:10willing to pay and [music] the second
  66. 02:12that the AI companies quickly blew
  67. 02:14through what we thought were the
  68. 02:16ceilings on [music] ability to pay or
  69. 02:18the sort of amount that a customer would
  70. 02:19pay for a subscription. There's actually
  71. 02:21interesting reason for that. I' I'd love
  72. 02:23to give our AI companies credit and say
  73. 02:25it was foresight or experimentation, but
  74. 02:27the truth is the COGS for AI companies
  75. 02:30is non-trivial, right? It can actually
  76. 02:31be very very high, especially for
  77. 02:33[music] products like video generation.
  78. 02:35And because you had real costs in these
  79. 02:37businesses, they had to charge customers
  80. 02:39real money. And to deliver the very best
  81. 02:42product experiences and generations,
  82. 02:43they had to charge a lot of money. And
  83. 02:45what many of these AI companies found is
  84. 02:47that even as they raised prices,
  85. 02:48customers were willing to pay and in
  86. 02:50fact wanted to pay more. So that really
  87. 02:52[music] got me thinking about, hey, what
  88. 02:54is the extreme version of this? And I
  89. 02:55love exploring ideas in their extreme. I
  90. 02:57just think it's [music] a very useful
  91. 02:59way to extract the kind of core of your
  92. 03:01thinking. In the extreme of, you know,
  93. 03:04people being willing to pay high prices,
  94. 03:06there's two actual implications. [music]
  95. 03:08The first is that you can build a
  96. 03:09software company with real revenue scale
  97. 03:12with very few customers on a relative
  98. 03:14basis, [music] right? 41,000 for the
  99. 03:16$und00 million run rate at $200 a month.
  100. 03:18And the second is that software should
  101. 03:21subsume almost every part of a consumer
  102. 03:23spend over time. And increasingly
  103. 03:25[music] those dollars are going to be
  104. 03:26captured by AI and by software products.
  105. 03:29So I think it's [music] actually a very
  106. 03:31very optimistic prediction, one that
  107. 03:32we've seen come true, which is that more
  108. 03:35individuals will be able to build
  109. 03:37largecale AI companies and consumers
  110. 03:39will have more of their needs met
  111. 03:41through software.
  112. 03:43I would say in the world that we're
  113. 03:45living in, there are no marketing
  114. 03:46problems. There are only product
  115. 03:48problems. I don't [music] think products
  116. 03:50should have CAC today. And if you need
  117. 03:52significant customer acquisition costs,
  118. 03:54that means you haven't sufficiently
  119. 03:56delivered on the product. [music] The
  120. 03:57truth is that founders and companies and
  121. 04:00products were never able to deliver with
  122. 04:02the kind of ambition that they can
  123. 04:03deliver today. You can just go insanely
  124. 04:06deep. [music] And part of that is
  125. 04:08because the models can do things that
  126. 04:09they could never do before. You know, we
  127. 04:11had 40 [music] years of building models
  128. 04:13that enabled or extended the
  129. 04:15intellectual parts of our brain and the
  130. 04:17intellectual parts of our society,
  131. 04:18[music] right? But that was only a
  132. 04:20single sort of aspect of the human
  133. 04:21experience and really of our
  134. 04:23civilization and society that was
  135. 04:24addressable by technology. Now with
  136. 04:26these new subjective [music] creative
  137. 04:28computers, we can address the entire
  138. 04:31nondeterministic part of human society
  139. 04:33and the human experience. [music] That
  140. 04:35is our emotions, our relationships, our
  141. 04:37desire for self-expression, [music]
  142. 04:39the creative work that we do. So this
  143. 04:42entire part, arguably a larger part of
  144. 04:44the human experience [music] is now
  145. 04:45addressable through technology. And that
  146. 04:47simply wasn't the case before. So I
  147. 04:49think that's one important point. The
  148. 04:50second is that thanks to AI code and the
  149. 04:53like collapsing costs and difficulty of
  150. 04:55making software. It it's just way easier
  151. 04:58for a small number of people to build a
  152. 05:00lot more. [music] So this simply was not
  153. 05:02possible prior and now you have founders
  154. 05:05that can do more things at an order of
  155. 05:07magnitude lower cost. And as a result
  156. 05:09you can go deep or go home instead of
  157. 05:12going big or go home.

  158. EO Partner Highlight

  159. 05:16What if your CRM kept up with the
  160. 05:19conversations that matter for you? Meet
  161. 05:21Atoio, the AI CRM for modern businesses.
  162. 05:24[music] Connect your email, calls, and
  163. 05:26product data, and AIO instantly builds
  164. 05:29your CRM with complete context. [music]
  165. 05:31Then ask anything, spot a deal at risk,
  166. 05:35draft outreach, uncover new [music]
  167. 05:37prospects, spend less time in your CRM,
  168. 05:40and more time building relationships.
  169. 05:42Whether you're running productled growth
  170. 05:44or enterprise sales, AIO [music] adapts
  171. 05:46to how your team sells and grows. And if
  172. 05:48you're building with AI agents, plug any
  173. 05:51MCP clients straight into your CRM, so
  174. 05:53[music] your agents always have the
  175. 05:55context they need to act. 9,000
  176. 05:57companies build with AIO, [music]
  177. 05:58including Granola, Modal, and Whisper
  178. 06:01Flow. Try Atoio free at aio.com/eo.

  179. Narrow Startups

  180. 06:13So narrow startups are companies that
  181. 06:15build incredibly opinionated deep
  182. 06:18products, charge very high prices for a
  183. 06:20relatively small number of people. You
  184. 06:22know, the simple math is that charging
  185. 06:2441,000 people $200 a month is a 100
  186. 06:28million run rate business. And there's a
  187. 06:30lot of precedent for this already
  188. 06:31occurring. You know, we see Google
  189. 06:33Ultra's top [music] skew is 250 a month.
  190. 06:34Grock is 300 a month. Open AAIS is 200 a
  191. 06:37month. I believe Anthropics is 200 a
  192. 06:39month. Consumers are flocking to these
  193. 06:42products organically. They're paying
  194. 06:44high [music] prices for them. And over
  195. 06:45and over again, we're seeing them
  196. 06:46delivered the value that they expected.
  197. 06:49So the whole idea behind narrow startups
  198. 06:50is build small, go deep, [music] and
  199. 06:53charge a lot. I think specialization is
  200. 06:56a new moat. I think that you have the
  201. 06:58ability to go so much deeper with a new
  202. 07:00technology and the collapsing cost of
  203. 07:02software creation for an individual
  204. 07:04customer that you can just be so much
  205. 07:06more specialized for that customer that
  206. 07:08it's hard to compete with. You know,
  207. 07:09somebody's going to have to build 3
  208. 07:11years of roadmap to have [music] a
  209. 07:13competitive product. So, it's simply
  210. 07:15differentiation taken to an extreme
  211. 07:17degree. I think that's an interesting
  212. 07:19and important form of a moat which is
  213. 07:22particularly relevant to narrow startup.
  214. 07:24I think the second is if you look at
  215. 07:25chat GPT they're trying to do a lot of
  216. 07:28things and if you think about areas in
  217. 07:31which there's a really rich software
  218. 07:32ecosystem that has to be built to really
  219. 07:35capture the value I don't know where
  220. 07:36that's going to fall on their priority
  221. 07:38list a great example is meeting
  222. 07:40recorders there's many products that now
  223. 07:42take notes for you by transcribing
  224. 07:44speech to text that is great but to
  225. 07:46fully capture the value you probably
  226. 07:48need to build a whole office suite you
  227. 07:50need [music] spreadsheets you need word
  228. 07:52processors you need a diary diary app
  229. 07:54and a notes app and you need all kinds
  230. 07:56of software. It's just not obvious to me
  231. 07:58that the labs are going to actually
  232. 07:59[music] get to that. So I do think that
  233. 08:01building a rich software ecosystem, a
  234. 08:03rich product ecosystem is another way to
  235. 08:05compete. Okay, I think the third thing
  236. 08:06is that there are many product
  237. 08:08categories like AI code where you
  238. 08:11benefit from using many models, right?
  239. 08:13It's better to be able to use anthropic
  240. 08:15and open AAI and Google's models. And if
  241. 08:18you're at OpenAI, you're never going to
  242. 08:19be able to build a product that also
  243. 08:21uses Google's models. [music] So being
  244. 08:23multimodel is a way to compete with
  245. 08:25elabs and big tech. The other important
  246. 08:27point is that when these products overd
  247. 08:30deliver for their customers and they can
  248. 08:32if you ask cursor to help you generate a
  249. 08:34feature with a model sometimes it's like
  250. 08:36wow this was even better than what I had
  251. 08:38hoped for or what I had imagined. So one
  252. 08:41the fact that these products can
  253. 08:43actually have those attributes and can
  254. 08:45overd deliver on the customer's
  255. 08:46expectations but the second is that they
  256. 08:48can charge for it. You know, sometimes
  257. 08:50the model has to think really hard to
  258. 08:53deliver that extraordinary outcome. And
  259. 08:55guess what? When it does, it's
  260. 08:56expensive. And that is the way that it
  261. 08:58should be. I think that when the product
  262. 09:00you deliver cannot be 100x better than
  263. 09:03everything else, of course, distribution
  264. 09:05is what matters. And I think the lie
  265. 09:07that we have sometimes told ourselves as
  266. 09:09founders is that something that's
  267. 09:10incrementally better is 100x better. You
  268. 09:13know, I think of this as silver bullets
  269. 09:14versus lead bullets. One silver bullet
  270. 09:17is a dramatic improvement. Many lead
  271. 09:19bullets are many small incremental
  272. 09:21improvements. Like 10 or 50 or even 100
  273. 09:24small improvements, 10 or 50 or 100 lead
  274. 09:26bullets [music] never equal a silver
  275. 09:28bullet. You really need that 100x value
  276. 09:30leap. Now with the models that we have
  277. 09:32access to, [music]
  278. 09:33we are a wash in silver bullets, right?
  279. 09:35There are silver bullets everywhere. So
  280. 09:37I actually do think in this day and age
  281. 09:39with the technologies we have access to
  282. 09:42you can win by betting by having a
  283. 09:43better product.

  284. Build for Pull, Not TAM

  285. 09:48Predicting a total but addressable
  286. 09:50market is a fool's errand. It's it's
  287. 09:52just impossible. I it's very very
  288. 09:53difficult and it's a common source of
  289. 09:55failure for investors certainly but even
  290. 09:58for founders. When I was a first-time
  291. 09:59founder I [music] had this big brain way
  292. 10:02thinking of you know products which is
  293. 10:04hey we need a big market. It has a needs
  294. 10:06to have a big TAM. I wasn't even quite
  295. 10:08sure what TAM [music] meant, but it
  296. 10:09seemed important and I know you needed a
  297. 10:10big one. A big one is better than a
  298. 10:11small one. And that's why a lot of my
  299. 10:13early thinking was in markets like
  300. 10:14healthcare and you know disease
  301. 10:17management and I just didn't know
  302. 10:19anything about those markets nor did I
  303. 10:20have energy for those markets. You know
  304. 10:22how I built a successful product was
  305. 10:24building something that I wanted to see
  306. 10:26exist and I was personally [music]
  307. 10:27passionate about which was sort of
  308. 10:29social graphs and mobile games. And
  309. 10:31that's what me and my founder built.
  310. 10:32Look, when when the iPhone app store was
  311. 10:35released, there were 6 million iPhones
  312. 10:36in the world. Like, [music] that's not
  313. 10:37much of a TAM, but we built there
  314. 10:39because it felt like it was growing
  315. 10:41quickly and we had a lot of energy for
  316. 10:42the market and we bet on, you know,
  317. 10:45perhaps not even thinking about the TAM
  318. 10:47and we were right. So, I don't think
  319. 10:49about TAM very much at all. I do think
  320. 10:51about value delivered to the customer
  321. 10:53and [music] the, you know, price they're
  322. 10:54willing to pay. Okay. So, I think the
  323. 10:56most useful prompt for a founder right
  324. 10:58now is what is the $1,000 a month skew
  325. 11:01of our [music] product, right? That is
  326. 11:02the direction we need to be thinking
  327. 11:04about like what is the extraordinarily
  328. 11:06expensive? What would the product need
  329. 11:08to do? Does [music] it do it today?
  330. 11:09Would people be willing to pay? Have we
  331. 11:11tested it? So, I think if you find
  332. 11:12customers that are be willing to pay
  333. 11:14dramatic prices for your product, you're
  334. 11:16probably on the right track. [music] You
  335. 11:17know, conversely, if you have a free
  336. 11:19product that you have to pay customers
  337. 11:20to try, you're probably on the wrong
  338. 11:22track.
  339. 11:23[music]
  340. 11:23It's a much more useful signal for
  341. 11:25builders than thinking about concepts
  342. 11:26like TAM. If people are paying for it,
  343. 11:28they're getting value typically. Of
  344. 11:29course, what is like upstream of that?
  345. 11:31[music] Things like retention and things
  346. 11:33like customer acquisition cost. So these
  347. 11:35things can be measured, but this is why
  348. 11:36it's so useful to build in an area in
  349. 11:38which you have great intuition cuz you
  350. 11:40just [music] you feel the feelings, you
  351. 11:42know it, you talk to the customer,
  352. 11:43you've perhaps you're the customer
  353. 11:45yourself or you've got great intuition
  354. 11:46around their pain [music] points. The
  355. 11:48customer has more ideas for your road
  356. 11:50map than you have. Like there are a lot
  357. 11:51of qualitative signals. The most
  358. 11:53overriding signal is you simply can't
  359. 11:56keep up with everything that is
  360. 11:57happening as a result. Like that's how
  361. 11:59you know you have product market fit. As
  362. 12:01Mark famously said, the market is
  363. 12:03[music] pulling the product out of you
  364. 12:04often violently. That is the experience
  365. 12:06of it. I mean I think there there's many
  366. 12:09psychological traps from being a
  367. 12:10founder. I can tell you a few of the
  368. 12:12ones that I fell [music] prey to and
  369. 12:13experienced as a founder. So one is
  370. 12:15trying to talk yourself into having
  371. 12:17product market fit. like [music] if you
  372. 12:19have to talk yourself into it, you don't
  373. 12:20have it. I think that's incredibly
  374. 12:22important. I think the second is, and
  375. 12:24perhaps a related point, you know,
  376. 12:25you're looking for metrics [music] that
  377. 12:27will justify this fact that you have
  378. 12:30market fit and you go crazy looking to
  379. 12:32calibrate on what's [music] good
  380. 12:34retention, what's good. Those are often
  381. 12:37not productive. Ultimately, a business
  382. 12:39has [music] physics and if you're losing
  383. 12:4190% of your customers at the end of year
  384. 12:43one, like [music] even if that's
  385. 12:45bestin-class for the category, it's very
  386. 12:46difficult to build something that's
  387. 12:48working. So, [music] I think thinking
  388. 12:50about the sort of business health and
  389. 12:51first principles rather than frameworks
  390. 12:53is often more productive. I think the
  391. 12:56final trap that you can often fall into
  392. 12:58is the power user trap. Power users are
  393. 13:01power users and that's great that
  394. 13:02they're getting that much value out of
  395. 13:03the product, but if you're not able to
  396. 13:05capture the value that they're getting,
  397. 13:07you know, they still only count as one
  398. 13:09dot on your growth chart. So, you really
  399. 13:11do have to either build for power users
  400. 13:13and [music] capture the value you're
  401. 13:15creating, which is the narrow startups
  402. 13:16idea, or you need to build for a mass
  403. 13:19market and not, you know, tell yourself
  404. 13:21that having some really happy power
  405. 13:23users is a substitute for having broad
  406. 13:25market fit.
  407. 13:28The most important piece of advice is
  408. 13:30that there is no marketing problems.
  409. 13:32There are only product problems. Be
  410. 13:34insanely ambitious on product. Raise
  411. 13:36prices. Adjust based on what you hear
  412. 13:38from the customer. And don't worry so
  413. 13:40much about business books, frameworks.
  414. 13:42Just build for a small number of people.
  415. 13:44Charge a lot. Go insanely deep. And you
  416. 13:47know, more likely than not, you'll find
  417. 13:48your way to success. Like this is not a
  418. 13:5020, 30, 50year idea. This is like a 3
  419. 13:52[music] 57year idea. That's what the
  420. 13:54abundance agenda means. And it's coming
  421. 13:56now because there's both abundant
  422. 13:58capital and [music] dramatic consumer
  423. 14:00interest in these new products. You
  424. 14:01know, if you were ever going to start a
  425. 14:03company, start it now. Like, there are
  426. 14:06better and worse times. [music] And this
  427. 14:07is the best time I've seen in my entire
  428. 14:09career by a long shot.