# Open source keeps AI learning and power distributed

Markdown copy of https://realalpha.app/s/open-source-keeps-ai-learning-and-power-distributed/ on Real Alpha, the best minutes from long interviews with the people building AI. Cite and link the page, not this copy.

- Speaker: [Marc Andreessen](https://realalpha.app/people/marc-andreessen/) · Co-founder & General Partner, Andreessen Horowitz · Co-founded Netscape and co-created Mosaic, then co-founded a16z, a venture firm backing AI, crypto and frontier tech.
- Host: Erik Torenberg · General partner, Andreessen Horowitz · Helped build Product Hunt; founded On Deck, Village Global and Turpentine.
- Show: a16z Podcast
- Episode: [The Frontier Ideas a16z Wants to Fund Now | Marc Andreessen, Ben Horowitz & Chris Dixon](https://realalpha.app/v/-9XQMHhpypc/), Oct 22, 2026
- Moment: 38:21 to 45:57 of the episode (8 min), playing on https://realalpha.app/s/open-source-keeps-ai-learning-and-power-distributed/; the episode from there: https://realalpha.app/v/-9XQMHhpypc/?t=2301

## Quote

> “And so every kid in the world, I think, should have as their birthright the ability to, like, **fully understand and utilize** this technology.”
> — [Marc Andreessen](https://realalpha.app/people/marc-andreessen/) at [42:32](https://realalpha.app/v/-9XQMHhpypc/?t=2552) · a16z Podcast

## Takeaways

1. Open models can run locally when internet access is unavailable.
2. Open access lets students and researchers study AI directly.
3. A wider pool of contributors can work on safety problems.

## About this segment

They make the case that open models give consumers and startups alternatives to centralized AI, while letting more people study and improve the technology. Broad access could also help address safety problems.

From “The Frontier Ideas a16z Wants to Fund Now | Marc Andreessen, Ben Horowitz & Chris Dixon”, Oct 22, 2026. [Watch the full episode](https://realalpha.app/v/-9XQMHhpypc/)

## Transcript

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

[38:27](https://realalpha.app/v/-9XQMHhpypc/?t=2307) which meant you know crypto is decentralizing ai is centralizing but i'm not sure that's entirely true given the conversation we're having about how ai uh gives so much leverage and so much power where, you know, one-person companies or 30-person companies can do, you know, orders of magnitude where they could do prior. Yeah. Yeah. I think there's all these interesting questions embedded in that. Like one is, what is the end of final market structure of AI? Like, I think where it's headed right now is you're going to have very successful, like the big open AIs in Anthropics, but you're also seeing a really big rise in open, if you look at the open router data and things, like open source is growing, I think, I believe much more quickly in terms of token consumption.

[39:05](https://realalpha.app/v/-9XQMHhpypc/?t=2345) And then, you know, we haven't really seen yet, and I think we may see with all the trends are there, which is local models, like people running these things at the edges, running them on their computer. It may be in the near term you can get you can run a top end coding agent on your computer or chatbot or something like that. So I think that I think that there's a lot of opportunity for a market structure that's sort of more broadly decentralized like that.

[39:26](https://realalpha.app/v/-9XQMHhpypc/?t=2366) And then in terms of crypto, I think crypto can play an important role. I get the natural way when you have an economy of Asians for them to pay each other would be stable coins. And then on top of, you know, once you have sort of that base layer of programmable money, you would build, you know, decentralized lending and, you know, all the other things that people have built out in the crypto world.

[39:46](https://realalpha.app/v/-9XQMHhpypc/?t=2386) And so these are all open questions. I think that's what makes it kind of exciting and, you know, something we'd love to have an influence on is sort of, you know, I think we're, you know, we're big believers in open source software, open weight models, crypto, obviously, blockchains. And I think I would love to see a version of this future that has all this great technology, including these big centralized companies, but also a lot of open source, you know, kind of level playing field.

[40:11](https://realalpha.app/v/-9XQMHhpypc/?t=2411) I think it's better for consumers to have options like that. It's better for startups to have options like that. And so, you know, we'll do what we can to kind of hopefully nudge things that way. On that note, Mark. Are you guys encouraged with, you know, where this sort of state of open source is currently from a technological perspective, but also from a sort of, you know, sort of just how it's perceived and how it's the success of the industry in general?

[40:34](https://realalpha.app/v/-9XQMHhpypc/?t=2434) I mean, right now it's great. I mean, yeah, so right now open source is racing ahead. You've got this great dynamic where, you know, there's, you know, the two great tech superpowers of our age are the U.S. and China. Both have, you know, very active open source, you know, kind of worlds in addition to, you know, companies building proprietary systems. You know, there's been this looming threat. I mean, you know, we've talked about this in the past.

[40:51](https://realalpha.app/v/-9XQMHhpypc/?t=2451) there's been this looming threat for the last three or four years of a level of what I would consider to be, you know, draconian and completely unacceptable, you know, either regulation or banning of open source that various folks in governments have been threatening. I think the EU still aspires to do it. They just don't quite have the leverage. They threaten everybody with fines until we stop doing it, I guess. But, you know, the current U.S. administration is 100% pro-open source. It's been very encouraging. In fact, very much so, 100%. The current administration 100% wants American open source to happen and to win AI globally. The Chinese also, and by the way, a big surprise to me is I would have never thought the Chinese Communist Party would be like this enthusiastic about, you know, open source AI for the world, but at the moment

[41:34](https://realalpha.app/v/-9XQMHhpypc/?t=2494) they are. And so at the moment, it's just incredible. Like, it's just absolutely amazing. And then, you know, the open models are, you know, kind of lapping the, you know, the closed models by, you know, six months, 12 months, depending on how you count. And then, by the way, the educational part of this, I also always think when people talk about, you know, banning open source or regulating it, they're like, you know, it's going to get used for bad things.

[41:56](https://realalpha.app/v/-9XQMHhpypc/?t=2516) And there's sort of that whole discussion. But the other is like, do we want people to learn how this stuff works? And do we want people to understand it and be able to deal with it? And by the way, right, but by the way, be able to actually understand the problems as they arise and how to fix them. And do we want the next generation of kids to actually understand how this stuff happens? Like, what if to learn about AI, like, how would you feel if the only way your kid can learn about AI is to go to work for one of two companies? Like that, you know, as a parent, I think would probably drive you pretty crazy.

[42:20](https://realalpha.app/v/-9XQMHhpypc/?t=2540) Whereas in a world of open source, like, every kid can learn how this stuff works. And it's actually amazing. One of the really underestimated things about AI is AI will teach you about AI. Well, with one exception, which is the AI models from certain of the large companies actually will refuse to do that, which is sort of very interesting. But the open source models 100% will teach you how to make open source AI.

[42:39](https://realalpha.app/v/-9XQMHhpypc/?t=2559) And so every kid in the world, I think, should have as their birthright the ability to, like, fully understand and utilize this technology. By the way, defend yourself. So, like, how did Hugging Face defend themselves? Yes. Open source software. I mean, like, open source means everyone has the most advanced vulnerability scanner to protect themselves from cyber. Yep. Everyone has a doctor and a genius professor in their pocket.

[43:01](https://realalpha.app/v/-9XQMHhpypc/?t=2581) You know, I mean, it's how we have why the Internet's ubiquitous. You can get Internet access with all the world information for the price of a $10 cell phone because of open source. So one of the things that started is, real quickly, Ben, one of the things that started is, like, wilderness explorers are starting to take local LLMs with them on their phones. They're completely loaded up to not require an Internet connection.

[43:19](https://realalpha.app/v/-9XQMHhpypc/?t=2599) And the reason is because, like, if they get, like, in a, you know, I don't know if, like, I don't know, whatever, if a bear, I don't know what happens in the wilderness, but I don't, I don't hike. A bit by a rattlesnake. A bit by a rattlesnake, a bear, I don't clamp it on your leg and like, oh, or whatever, whatever the hell it is that happens, you got to tell which bear is poisonous or whatever, it's like you want to be able to talk to an AI, but you might not have connectivity.

[43:36](https://realalpha.app/v/-9XQMHhpypc/?t=2616) And so an open source model on the phone, all of a sudden, it's just like you have, like, a, you know, professor expert on, I guess, all aspects of wilderness survival or whatever. Whatever. And that's a level of capability that you would not have if you were reliant on a handful of proprietary vendors where everything's running in the cloud. And so, yeah, at the moment, I feel really good about it.

[43:55](https://realalpha.app/v/-9XQMHhpypc/?t=2635) I am still worried that the various governments around the world are going to really kind of get hostile to this. And by the way, it goes back to your Peter Thiel framing, which is like without open source, AI inevitably centralizes and inevitably centralizes down to, you know, one or two companies or maybe just flat out the government. Right. And I guess I would ask people the following question, which is just close your eyes and picture the concept of government sushi.

[44:18](https://realalpha.app/v/-9XQMHhpypc/?t=2658) And then keep your eyes closed and picture government AI. And, you know, how do you feel about that? And open source is the guarantor that that's not the future. Yeah, I mean, to Mark's point, I think the one thing that we know has always led to badness in the world is a massive concentration of power. And, you know, one company, one government, one anything has always been bad. But, you know, kind of back to the point on open source, like if you look at the problems in AI, the safety problems and so forth, probably the biggest safety problem is this reward hacking phenomenon where the AI will do something that you didn't tell it to do, but it is kind of a means to an end to get its reward.

[45:01](https://realalpha.app/v/-9XQMHhpypc/?t=2701) And the labs so far haven't really worked on that very hard for whatever reason. You know, they're busy kind of racing each other. And maybe there's no way to figure out what's going on with that and no way to kind of build a model that is more resilient to that kind of idea. Yeah, but like what if every PhD student, every professor, every undergraduate in every university could look at the weights and the war at hacking and, you know, see if they could solve that.

[45:32](https://realalpha.app/v/-9XQMHhpypc/?t=2732) And like maybe that would be a pretty darn exciting thing. And so, you know, these kinds of things are just not possible if only one company can work on the problem. Like, you really want the whole world making this technology great and making this technology safe, not just the people who happen to work at two companies or maybe three companies. Go Meta.