Jul 28, 2025 · 1h 16m · 20vc
a16z GP, Martin Casado: Anthropic vs OpenAI & Why Open Source is a National Security Risk with China · 20VC with Harry Stebbings
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of 20VC, Andreessen Horowitz General Partner Martin Casado joins Harry Stebbings to analyze the economic realities of the AI landscape, focusing on open-source national security strategy, value distribution across the tech stack, and modern venture capital mechanics. He explores how AI is reshaping software engineering and debunks persistent myths around market monopolies and AI safety.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Harry holds 15.6% of the talking time here. How this is scored →
Harry as informed peer 4.0 Guest teaching 4.9 Guest disagreement 2.4 Harry pushing back 3.2
speaking balance: gold is Harry, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 40:46 Rejecting developer productivity hyperbole
Martin explicitly shuts down Harry's binary framing of 1x to 10x or 10x to 100x engineers, declaring 'I don't actually think it's that' and arguing AI only provides 2x gains for top engineers.
Hardest push from Harry ▶ 57:53 Refusing founder control over VC strategyHarry forcefully refuses Martin's framing regarding portfolio conflict management, stating directly that he will not let a portfolio founder tell him how to do his job.
Biggest teaching moment ▶ 31:11 National security and Lawrence Livermore historyMartin uses his firsthand 1999 experience with supercomputing export controls at Lawrence Livermore National Labs to educate Harry on why technological containment fails compared to open proliferation.
Harry holds his own ▶ 53:47 Demonstrating Series A vs mega-fund mechanicsHarry shows deep operational expertise in venture capital mechanics, contrasting his own Series A cost of capital and ownership constraints against Andreessen Horowitz's multi-stage flexibility.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Harry as informed peer | Guest teaching | Guest disagreement | Harry pushing back | Why |
|---|---|---|---|---|---|---|
| The AI Investment Landscape and the Zero-Sum Myth | 4 | 3 | 2 | 4 | Harry challenges the current AI hype by contrasting it with the 2021 market bubble where playing the field proved costly. Martin agrees that mark-driven investing is dangerous, but argues current behavior follows unprecedented real user and revenue growth. | |
| The Coding Frontier: Anthropic vs. OpenAI and the Threat of Monopolies | 3 | 5 | 3 | 2 | Harry asks whether developer tools built on top of Anthropic are vulnerable to platform risk. Martin reframes the premise by pointing out post-launch perception bias and explaining that models distill quickly into a competitive market. | |
| Predictability and Evolution of Large Model Providers | 3 | 4 | 1 | 1 | Harry asks Martin to predict whether an oligopoly or monopoly will emerge in AI models. Martin draws a historical analogy to AWS early dominance in cloud infrastructure to explain why Google and Microsoft will force an oligopoly. | |
| Assessing Models as Venture Investments | 5 | 5 | 3 | 4 | Harry presses Martin on whether model companies are viable venture investments given stock compensation and heavy dilution. Martin corrects the blanket generalization by categorizing model economics into specialized diffusion vs heavily subsidized frontier language models. | |
| Paradoxes in AI Stack Value and the Power of Brand Effects | 5 | 4 | 1 | 2 | Harry cites specific category winners like Lovable and 11Labs to ask about market value concentration. Martin presents his brand effects thesis, arguing household name recognition creates distribution modes in expanding markets. | |
| The Lifespan of Brand Dominance and Market Saturation | 4 | 4 | 2 | 4 | Harry interrupts to challenge whether user growth represents genuine market expansion or temporary consumer intrigue. Martin clarifies how frontier growth suppresses competing messages until market saturation occurs. | |
| Investment Strategies and OpenAI's Rational Fragmentation | 2 | 5 | 1 | 1 | Harry asks how brand dominance shapes investment choices. Martin provides a masterclass on how OpenAI made rational strategic choices to focus on core language while ceding image, video, and code niches to competitors. | |
| Geographic Balkanization and the Rise of Regional Competitors | 6 | 6 | 4 | 3 | Harry shares a real-world case study of a European medical startup facing US competition and questions low AI gross margins. Martin forcefully reframes bad gross margins as a rational business trade-off sacrificing margin for land-grab distribution. | |
| Parallels in Security and the Myth of AI Safety | 4 | 7 | 4 | 4 | Harry brings up opposing VC stances on AI safety and Meta moving to closed models. Martin draws on his intelligence community and PhD background to criticize the current AI safety panic as disconnected from historical cybersecurity precedent. | |
| Open Source as a National Security Imperative Against China | 4 | 7 | 4 | 3 | Harry asks whether open source models empower hostile nation-states. Martin cites his 1999 Lawrence Livermore nuclear simulation work to argue that open technology proliferation is the proven strategy to maintain global technical leadership. | |
| Funding Scientific Research and the Academic Disruption | 6 | 5 | 4 | 8 | Harry directly challenges Martin by asking if Trump administration cuts to academic labs contradict his pro-funding stance. Martin pushes back against oversimplification, pointing out bipartisan consensus on indirect cost overhead reform dating back to Obama. | |
| The Real Business of Open Source AI | 3 | 5 | 2 | 3 | Harry questions if the market is shifting towards closed-source models. Martin outlines the core business mechanics of AI open source, where smaller models are released for brand distribution while top-tier weights remain proprietary. | |
| How AI Reclaimed the Joy of Programming | 1 | 4 | 0 | 0 | Martin reflects on how AI coding tools like Cursor stripped away framework overhead and restored the joy of programming for experienced developers. | |
| The Limits of Developer Productivity | 4 | 6 | 5 | 2 | Harry asks if AI tools make 1x engineers 10x or 10x engineers 100x. Martin explicitly rejects the binary premise, explaining that AI makes 10x engineers 2x because core trade-offs and architecture remain hard. | |
| App Defensibility vs. Infrastructure Moats | 5 | 6 | 3 | 3 | Harry quotes Fiverr CEO Misha on reduced time to copy software. Martin distinguishes between application layer code—which he calls trivial CRUD—and infrastructure, citing Aaron Levie's data on production pull request sizes. | |
| The Future of Computer Science Education and Model Overload | 3 | 4 | 1 | 1 | Harry criticizes current user friction like manually choosing between specific model versions. Martin shares a philosophical perspective on how AI can eliminate duplicate academic research and connect disparate fields. | |
| The Human Handler in the Era of Job Displacement | 4 | 5 | 2 | 3 | Harry compares current AI labor disruption concerns with historical technology panics cited by Brad Feld. Martin explains why AI is distinct because current systems inherently require a human handler to manage output variance. | |
| The Realities of Venture Capital and Fund Dynamics | 7 | 5 | 2 | 5 | Harry presses Martin on fund dynamics, price elasticity, and ownership targets. Martin details the mathematical necessity of ownership to make large VC fund return mechanics function. | |
| Resolving the Conflict of Interest Accusations | 6 | 4 | 3 | 8 | Harry confronts Martin over portfolio conflicts at Andreessen Horowitz and forcefully rejects letting portfolio founders dictate VC investment decisions. Martin defends firm practices using Chris Dixon's 'one mortal enemy' framework. | |
| Unpacking Elad Gil's Investment Strategy | 6 | 5 | 2 | 4 | Harry contrasts founder-first investing with Elad Gil's market-first strategy. Martin explains Elad's founder-market fit approach and highlights internal Andreessen metrics where missing the category winner is the sole unforgivable sin. | |
| Diligence vs. Prediction in Venture Strategy | 5 | 6 | 5 | 6 | Harry challenges whether failing to reward market foresight is flawed logic. Martin rejects the idea that VCs can predict tech adoption, comparing it to weather forecasting and prioritizing team diligence instead. | |
| The Bittersweet Reality of Selling a Company | 2 | 5 | 1 | 1 | Harry asks Martin about the emotional reality of selling Nicira. Martin shares a personal story about driving to Hollywood to leave tech, turning around on I-5, and realizing his true passion was software. | |
| Wealth Psychology and the Martine Coin | 1 | 3 | 1 | 1 | Harry asks Martin about wealth psychology and marriage. Martin describes growing up poor in Montana, using the 'Martine coin' mental model to spend money, and how family keeps high-performing founders grounded. |
Statements from this episode (59)
Casado: Open-Source AI Is Dangerous Because China Is Better At It
“I think that right now open source is most dangerous because China is better at it than we are, and as a result of that.”
Casado: AI marks the first time software development itself is disrupted
“This is really the first time like software development and software creation is, is being disrupted.”
Casado: Zero-Sum AI Investing Is A Mistake; Every Layer Captures Value
“Observationally, there's only been one sin, and that one sin is zero sum thinking. We always worry about like, oh, is this defensible? Oh, will this layer get margin? Will this layer get value? And the answer has kind of been unilaterally yes. The answer has b…”
Casado: AI Models Lack Long-Term Moats Because Distillation Is Too Easy
“So historically, Models don't really keep much of an advantage because they're so easy to distill.”
Casado: The AI coding model market will become an oligopoly
“Oligopoly. This is how the cloud, well, this is how the cloud played. I think probably the best analog we have is the cloud, right?”
Casado: Early AWS Had Far More Market Dominance Than Anthropic Does Today
“AWS was like, 70 or 80% market share early on. Nobody thought they could ever catch up to them. You know, they were the massive market leaders that created the category. I mean, they had way more dominance than Anthropic has now.”
Casado: Google's Gemini 2.5 Beats Anthropic On Price-Performance For Many Uses
“Gemini 2.5 is a great model. It's a great model. And if you actually look at it, you know, on the price performance, I would say in many use cases, the one that I actually use as my standard model, it's better than Anthropic.”
Casado: Anthropic and OpenAI will remain long-term stalwarts in AI
“Both Anthropic and Opener have done a remarkable job, remarkable with brand independence and market share, and so I suspect they'll continue to be stalwarts in the industry.”
Casado Confirms Andreessen Horowitz Is an Investor in OpenAI
“We're investors in opening eye, yeah.”
Casado: Specialized Diffusion Models Like Midjourney Have Superior Business Economics
“Like, say, like 11 Labs, Midjourney, Black Forest Labs, Ideogram. These are wonderful businesses that have great economics because the models are smaller.”
Casado: Google Subsidizes Language, Code, And Video AI Models, But Not Speech
“Google subsidizes language and code and video, but not speech, right?”
Casado: Capital Invested Outside Top Frontier AI Model Leaders Is Forfeit
“And so I would say it's kind of a high stakes game where the winners really win, but like, it requires a lot of capital. To enter the game. And if you're not in one of the leaders like that, you know, capital is, is forfeit.”
Casado: Midjourney Leads AI Image Generation With Zero Institutional VC Funding
“Mid journey was the first that got above the quality bar. It's taken zero investment from institutions. It's still the market leader. And it continues to do great.”
Casado: AI leaders will maintain brand monopolies until market growth slows
“Leaders are going to have brand monopolies and brand modes and they'll be able to maintain them until things slow down.”
Casado: OpenAI Lost Its Early Leads In Code, Image, And Video Generation
“OpenAI was the first to code, right? With GitHub Copilot. I mean, they provided the weights. As far as I know, and they lost that. And they were first to image with Dali, and they lost that. And they were the first to video with Sora. And as far as I can tell,…”
Casado: AI market will see long fragmentation before market consolidation
“So I think we're going to see fragmentation for quite a while before we see consolidation.”
Casado: Regulatory and cultural balkanization is driving regional AI winners
“We do have geographic biases showing up with AI and the regulatory environments are quite balkanized. You know, there's language and cultural biases that are also balkanized. And so we're actually seeing a lot of regional players show up.”
Casado: European AI market alone is large enough for major winners
“But I promise when it comes to AI, the European market is large enough. I promise that.”
Casado: Non-generalizing AI scaling creates room for specialized models
“In this phase of model scaling, a lot of the approaches to scaling don't generalize. So if I want to be much better at like coding, I may not be so good at something else. This gives a ton of room for the application developers to build their own models that s…”
Casado: Break-Even AI App Margins Are Deliberate Board Choices For Distribution
“In my experience, most of these companies that are like, let's say break-even margins, it's like a board level specific choice to prioritize distribution, not just because this is systemically something they have to do.”
Casado: A major AI-driven cyber attack is guaranteed to happen
“I still have yet to see the dramatic new attack. It's going to come for sure, but we haven't seen it yet.”
Casado: Unlike Past Tech Cycles, AI Pioneers Are The Primary Fearmongers
“In the past, the people created the technology were kind of pro tech and the people that were like selling security solutions were like the fear mongers, right? So you'd have somebody create like the internet and they're like, this is safe and it's great for e…”
Casado: Best counter to foreign open-source software is US openness
“And any of the software that produced by a nation state that we view, you know, quasi adversarially, the way that we combat that is we also are incredibly open. And we also do a proliferation of technology.”
Casado: US government must heavily fund open-source AI development
“I think we should be funding this stuff like crazy. I think we should get the national labs involved. We should get academia involved. You know, we should make this a national priority, just like China does, and we should just, you know, a full-throated endors…”
Casado Predicts Tech Ecosystem Will See Less Open-Source AI
“I do think it's quite likely that we're gonna see less open source.”
Casado: The AI Playbook Is Open-Sourcing Small Models While Keeping Advanced Closed
“The standard model of open sourcing AI is you open source the smaller model and you keep the more capable model closed source. And it's a way that you get distribution and ran brand recognition, but you don't actually erode your business.”
Casado: Open-Sourcing AI Models Does Not Enable Full Replication By Third Parties
“Unlike actual software open source, just because you release your model doesn't mean somebody can replicate it. Like to replicate it, you'd have to like recreate the data pipeline and the training pipeline.”
Casado: Open Source Captures Higher Market Value Share in AI Than Software's Historic 20%
“Historically, open source has only been about 20% of the total market value. I would say it's much higher than that for AI.”
Casado: I consistently underestimated how fast AI coding models would advance
“I mean, the one for me that I've just consistently got wrong is just how fast these coding models advance. And this is probably just sunk cost fallacy my entire life. I've just been this nerdy program and programming since the nineties. I mean, it's like, it's…”
Casado: By 2015, 90% of software engineering was platform management, not coding
“By, I would say, like, 2015 or so, you know, writing with something is, like, you'd have to, like, fucking, like, download, like, fifty million packages, and, like, to run it, you gotta run some stupid dev server, and to, like, actually have anybody else use i…”
Casado: AI Coding Models Are Bringing Veteran Developer-Executives Back To Programming
“I know a bunch of very strong developers that have been developing for a very long Time that have basically stopped. They're like running companies now or whatever. And they're all back to programming at night.”
Casado: AI Coding Assistants Only Double The Productivity Of 10x Software Engineers
“I think they make 10 X engineers two X.”
Casado: Every Single Company I Work With Uses The Cursor AI Assistant
“I would say every company I work with uses cursor, right?”
Casado: Vertical SaaS Tech Is Simple CRUD Where The Team Doesn't Matter
“Every time I look at vertical SAS, I'm like, why do we even care about the technical team? It's fucking crud, man. It's like, crud is like create, you know, read, update, delete. It's like, they all do the same thing. They all just kind of look like a web app.…”
Casado: AI won't speed up infrastructure dev, but will cut bugs
“So for infrastructure, I think it's quite unlikely that AI will really help in like speed that up because it comes down to something that the developer has to decide on, has to articulate the trade-offs. But I do think it could really help with the development…”
Casado: Average production code pull request is only two lines
“It's two. It's two. Yeah. It's very, very small. It's actually two, but let's say it's 12, right?”
Casado: AI is eliminating routine middle-tier software development
“And so in many ways, I would say, you know, the AI is getting rid of the middle, right? Like, so very new computer science, like models, they don't know how to do just because nobody's done it before. And that's kind of pushing the state of the art. And then i…”
Casado: Programmers will eventually stop worrying about frameworks and languages
“I just think hopefully we'll just stop worrying about frameworks altogether and maybe even languages, maybe even a, like a proto language evolves and we can just focus on, on, on logic and fundamental trade-offs.”
Casado: AI will liberate scientific research by bridging cross-disciplinary literature
“And so in a way, I think AI has the ability to pull out of this mass craziness, this mass ineffectiveness, which a it's very good at telling you if you've done it before, right? You know, it's very good at that. It actually knows all the literature, knows all …”
Casado: All Monetized AI Use Cases Currently Require A Human Handler
“One thing that's very unique about AI is that it actually requires today a human handler. I mean, they're just so unpredictable, you know, I mean, most of the use cases that we know, all the monetized use cases have a human on the other side of it, right?”
Casado: Andreessen Horowitz Walks Away From Deals Over Ownership, Not Price
“Price? No. Ownership? Yes.”
Casado: Early-stage deals must return 20% to 50% of fund on median outcomes
“For early stage investments. You know, you kind of need to understand what the median outcome is, and you have to be able to size the median outcome in a way that at least returns, say, a fifth of the fund or half of the fund.”
Casado: Multi-stage VC firms must offer full-stack funds to avoid being squeezed out
“The market is competitive. And everybody's scrambling for deals. And if you don't have the different funds or products to offer, then often that's kind of where people are going to squeeze you out or get alpha, et cetera. And so I think that for the game that …”
a16z frequently passes on startup investments due to portfolio conflicts
“In fact, I mean, we routine, I would say the number one reasons we don't, that's not true. One of the top reasons we don't invest in companies is because of conflicts.”
Casado: Founders Can Block a16z From Investing In Exactly One Mortal Enemy
“Listen, you have one mortal enemy, and you choose whoever that mortal enemy is, and whoever it is, I'm with you, we're gonna go kill that mortal enemy together, but you get one. You don't get an arbitrary number of mortal enemies.”
Casado: Elad Gil Is Perhaps The Best In Venture At Founder-Market Fit
“Elad is very, very focused on the founder. I think the one thing I would say, Is he's very good with founder market fit, maybe the best in the industry.”
Casado: The only real sin in venture capital is missing the winner
“The only sin in investing and I've sinned so much. The only sin investing is, is, is, is missing the winner. Like there's no, it's fine to like invest in a category that doesn't work. It's fine to lose money all. But like, if you choose the wrong company, like…”
Casado: Data Streaming Turned Out To Be A Subset Of Batch Analytics
“The entire streaming market has been very, very tough. Like the data streaming market. It's just turned out to be a subset of the analytics batch market.”
Casado: ClickHouse May Be The Only Data Streaming Breakout Since Confluent
“Click houses, Aaron Katz is doing phenomenal with, and I'm not an investor, but he's doing phenomenal, but that may be the one breakout since Confluent.”
Casado: Company diligence is far more reliable than predicting market adoption
“And so, I actually don't believe You can predict the future of technology adoption. It's a very tough thing, right? I mean, you don't know what a big company is going to do, can wipe out an entire market. You don't know what innovation will wipe out entire mar…”
Casado: AGI won't eliminate enterprise SaaS because humans are already AGIs
“I mean, I would say humans are AGI and we still invest in enterprise SaaS.”
Stebbings: Whatever Sam Altman and Microsoft decide is AGI will define AGI
“I think to be honest, Sam Altman, that's the definition of what AGI is. So whatever him and Microsoft decide is AGI will be AGI.”
Casado: Cursor founder Michael Truel has impeccable intuition and great product taste
“He knows what he wants. He's got an intuition that's impeccable and he listens incredibly well and gathers information. And that's a very, very potent combination. And then of course he's incredibly smart and he's got great product taste.”
Casado: Founders shouldn't execute post-exit plans created under peak stress
“Don't use those dreams that you concocted when you were like really in the pressure cooker, like not sleeping, your relationships are falling apart, that whole thing. Like that's not the thing, that steady state you're going to want to do.”
Casado uses 'Martine Coin' mental model to rationalize spending after creating wealth
“So I actually had to adopt a lot of these mechanisms where, like, I'll make a Martine coin and it's worth this much money.”
Casado: Male Founders With Stable Family Relationships Perform Much Better At Work
“I have found That men in particular that have stable relationships just do a much better job In work. They're just much more stable. I think the best founders I have tend to be, like, have families and etc.”
a16z GP Martin Casado works 80 to 100 hours per week
“I probably work all in. 80 to a hundred hours a week. I've been doing it for 10 years.”
Casado: Traditional VC Partnership Models Kill Decision Velocity And Adaptability
“And I think it's very, I mean, it's kind of a historical quirk that VC was created around a partnership model. Like that's the same thing you'd use for a dentist office or a law firm. And I think it's, there's Positives in that there's a bunch of different age…”
Shorts cut from this episode
▶ AI Startups Should Build Their Own Models · 20VC with Harry (@0:24)
▶ Brand Effects in AI · 20VC with Harry Stebbings (@0:19)
▶ How a16z Avoid Investing in Competition · 20VC with Harry St (@58:11)
▶ Distribution Over Margins · 20VC with Harry Stebbings (@22:35)
▶ The Biggest Sin in AI Investing… · 20VC with Harry Stebbings (@0:00)