Speaker · Updated
Martin Casado: latest market views
Andreessen Horowitz (a16z)
Summary
Martin Casado is a general partner at a16z who leads its infrastructure investing, and the cofounder and former CTO of Nicira, which VMware acquired in 2012, and a pioneer of software-defined networking.
Chorus has 27 views from Martin Casado across 2 episodes on Invest Like The Best and a16z Podcast since November 2, 2026. Their latest calls: bullish on Anthropic, Anysphere Inc. (Cursor), OpenAI and Snowflake.
Where they stand
Their latest bullish or bearish call on each company.
What they talk about most
- AI infrastructure & data centers · 5
- Venture capital & startups · 4
- AI: overall outlook · 3
- Cybersecurity · 2
- AI bubble & tech valuations · 1
- AI pricing & unit economics · 1
- Anthropic · 1
- Anysphere Inc. (Cursor) · 1
- Credit markets · 1
- Enterprise AI adoption & AI software · 1
- Frontier AI labs & model race · 1
- IPOs, M&A & deal activity · 1
The latest views
Newest first. Every quote is checked against the episode's transcript.
In growing markets defensibility barely matters; investors wrongly dismissed AI models as non-value-accruing.
In a growing market, companies don't compete head to head. They move into new white space, so defensibility is a second- or third-order concern. Casado says investors spent a lot of time worrying about it when it doesn't matter much in growing markets. Many said they would never invest in a model because it would never accrue value or be defensible, but as long as the large market grows, the value accrues anyway, as AI has shown.
“it doesn't really matter especially in growing markets which we invest in and we've seen this with AI. I remember early discussions on models ... investors are like, I will never invest in a model. It'll never accrue value. It's not defensible.”at 13:16
Invest Like The Best: Martin Casado on Frontier Model Economics, AI's Slowdown, and the Cursor Deal - [Invest Like the Best, EP.498] ·
AI infrastructure & data centers
Models act like ASICs, not software, and the whole infrastructure stack is at maximum disruption.
Traditional infrastructure is compute, network, storage and databases, and each is being reformed by AI. Models should be treated as infrastructure. They look more like ASICs than software because of their capital needs and competitive pressures, and they may disrupt the application stack above them. Casado says the industry is at a point of maximum disruption, though the basic components of infrastructure have not changed.
“models which I would consider infrastructure. They look to me more like ASICs than they look like traditional software.”at 18:38
Invest Like The Best: Martin Casado on Frontier Model Economics, AI's Slowdown, and the Cursor Deal - [Invest Like the Best, EP.498] ·
AI infrastructure & data centers
AI cluster networking is being reinvented, with new networking companies emerging for the first time since Arista.
Networking was sleepy and boring for about fifteen years. Now the clusters being built are very large and have rigid performance constraints, which opens room for net-new networking companies, the first since Arista. Materials used to build them are changing too, and he points to optical as an example of new activity in the area.
“Networking of course ... Like that's been so sleepy and boring for the last fifteen years and now because these clusters that we're building are so large ... we're seeing net new networking companies come out for the first time since Arista.”at 21:15
Invest Like The Best: Martin Casado on Frontier Model Economics, AI's Slowdown, and the Cursor Deal - [Invest Like the Best, EP.498] ·
Frontier models win 80%+ of dollar-weighted market and keep pricing power while demand is unsaturated.
Over three years, the best frontier models have taken 80%+ of market dollars. Being on the frontier gives tremendous pricing power, even with only an epsilon edge, because there is no demand problem. Competitive pressure means buyers will pay far more for the smartest model (he says 10x), so pricing holds. Penetration is only about 1% of enterprise budgets, so structurally nothing changes while the market grows. The open question is what happens if use cases saturate.
“these frontier models that are the best tend to get 80% plus of the market dollar weighted ... being on the frontier means you've got tremendous pricing power... There's no demand problem.”at 23:05
Invest Like The Best: Martin Casado on Frontier Model Economics, AI's Slowdown, and the Cursor Deal - [Invest Like the Best, EP.498] ·
Enterprise AI penetration is only about 1% of budgets, leaving a large runway for growth.
Figures cite about 3% of enterprises, but measured by budget, penetration is about 1% or less. Casado concludes there is a lot of room to grow given how useful the technology is. He treats this headroom as the reason pricing power and frontier leadership can persist while the market keeps growing.
“They say 3% of the enterprise, but actually from budget wise, it's like 1% or less. Right? And so there's just a lot to grow given how useful the technology is.”at 23:43
Invest Like The Best: Martin Casado on Frontier Model Economics, AI's Slowdown, and the Cursor Deal - [Invest Like the Best, EP.498] ·
A frontier model's useful lifespan is only three to six months, so labs must ship successors fast.
Casado says models are not like software, which has a long tail of retention. If a lab that put out the state-of-the-art model slips even three months, for regulatory, technical or other reasons, it could lose the entire market. He bases the three-to-six-month estimate on his experience and on the model graphs he sees in pitches. Models therefore need a rapid release cadence, and the financial models of these companies don't account for this risk well.
“I would say that lifespan of a model is three to six months. So that means you you have to release the next model.”at 24:55
Invest Like The Best: Martin Casado on Frontier Model Economics, AI's Slowdown, and the Cursor Deal - [Invest Like the Best, EP.498] ·
Value could crash back into open-source models if use cases saturate or the market slows.
If the market saturates in pockets, or slows for another reason, value could shift to open source. He says this would be a shift never seen before, because software has always had a long tail of retention while frontier model lifespans are very short. He presents it as a real possibility, not a forecast, and says nobody knows the answer.
“I think there's a very real world where the value comes crashing back down into open source and we'll see a shift that we've never seen because software's always had a long tail of retention.”at 25:04
Invest Like The Best: Martin Casado on Frontier Model Economics, AI's Slowdown, and the Cursor Deal - [Invest Like the Best, EP.498] ·
Enterprise AI adoption & AI software
Coding is about 95% automated, so token growth will flatten, but teams won't downgrade to cheaper models.
Casado separates two claims. First, the coding market is near saturation: teams are not using more tokens because they have automated about 95% of their work, and he thinks they are close to that ceiling. Second, that does not mean teams will switch to a cheaper model. In competitive equilibria, companies would always pay more for the smartest model to ship faster and better. He expects growth to broaden into general knowledge work outside code.
“statement number one is, I am not using more tokens. I have enough tokens. I've automated 95% already so I can't automate anymore. And I think that we're getting pretty close to that.”at 25:52
Invest Like The Best: Martin Casado on Frontier Model Economics, AI's Slowdown, and the Cursor Deal - [Invest Like the Best, EP.498] ·
Anything doable on a computer will be done by AI; physical-world scaling is still unproven.
Anything digital can be solved with enough capital, because any simulated task can get an RL environment (3D with Blender, general computer use, code) and be trained on. So AI will be able to do anything done with a computer, perhaps not yet. Money can now be pointed at problems that engineering used to block. He is less sure about the physical world: whether robotics, science and medicine follow the same scaling curves and distributions, which he calls an unlock still to come.
“anything you can simulate on a computer, you can create RL environments for ... Anything you do with a computer, AI will be able to do. ... What I don't know is, does the same thing apply to the physical universe?”at 31:13
Invest Like The Best: Martin Casado on Frontier Model Economics, AI's Slowdown, and the Cursor Deal - [Invest Like the Best, EP.498] ·
Today's AI-era breach wave looks less severe than early-internet worms that caused billions in damage.
Casado lived through the early internet, which he says was much worse. Worms caused billions of dollars of economic damage and took out critical infrastructure. Windows 95 machines connected to the internet could be infected before the installer finished, so there was no time to patch. He treats the current wave of breaches as less severe than that period, which is the same pattern that happens with any new technology.
“I lived through the internet and it was way worse than this. Like by this time with the internet, we had worms that had caused billions of dollars of economic damage. We'd taken out critical infrastructure.”at 32:42
Invest Like The Best: Martin Casado on Frontier Model Economics, AI's Slowdown, and the Cursor Deal - [Invest Like the Best, EP.498] ·
AI-found software vulnerabilities are a healthy sign; software will end up more secure as AI also secures it.
Casado compares the AI security shift to earlier technology shifts such as the internet, which brought a few years of mayhem and breaches before the industry adapted. He argues that defence has historically won when the good guys can outspend the bad guys, and that they will likely do so again. The vulnerabilities now surfacing are good because they show someone using models to find and fix weaknesses. His conclusion is that the net impact on software will be much more secure because AI is also securing it.
“These vulnerabilities that you see showing up, this is good. That means that somebody used the model to make things safer by pointing out the vulnerabilities. ... But I think the net impact at software will be much more secure because AI is also securing it.”at 33:34
Invest Like The Best: Martin Casado on Frontier Model Economics, AI's Slowdown, and the Cursor Deal - [Invest Like the Best, EP.498] ·
SnowflakeBullish
Snowflake, growing 30-40% with good economics, is not at risk of AI disruption.
Casado uses Snowflake as his example of a data infrastructure business that is hard to replicate. Building the product is not enough, because the long tail of features and performance tuning needed to operate such systems is very hard. He points to Snowflake's 30-40% growth and good economics and says he does not see AI as a disruption risk to it. The only competitor he names is Databricks, which he says is a pre-AI concern rather than an AI one.
“Like I don't think that they're at any risk of being disrupted by AI.”at 36:21
Invest Like The Best: Martin Casado on Frontier Model Economics, AI's Slowdown, and the Cursor Deal - [Invest Like the Best, EP.498] ·
Profitable software firms growing 40-60% are good businesses that AI will not disrupt; more software needs more infrastructure.
Casado describes profitable software companies growing 40-60% at scale, which he says are very good businesses. He argues that AI makes software easier to create, but as an infrastructure investor he says more software means more infrastructure is needed. He notes that vibe-coding has not yet replaced databases, and that building the long tail of features and performance work is hard. He concludes that AI is not going to disrupt these companies and that the market simply does not know how to value them.
“My view is like I don't think AI is gonna disrupt these. I think they are good businesses.”at 36:27
Invest Like The Best: Martin Casado on Frontier Model Economics, AI's Slowdown, and the Cursor Deal - [Invest Like the Best, EP.498] ·
AI lets capital substitute for engineering, an industrial-revolution-scale shift unlike past software tech.
Casado argues that engineering has always been a sublinear bottleneck, since adding people slows projects down, as the mythical man-month shows. Ten years ago giving a founding team a billion dollars would not have worked. With AI, 20 people can productively use a billion dollars. He does not know whether that money turns into the right returns, but he thinks the ability to deploy capital productively is new. He calls it an industrial-revolution-level shift.
“the big disruption of AI is the fact that you can throw capital at problems. In the history of tech, you haven't been able to do that because it was always an engineering problem which just doesn't scale.”at 37:41
Invest Like The Best: Martin Casado on Frontier Model Economics, AI's Slowdown, and the Cursor Deal - [Invest Like the Best, EP.498] ·
OpenAIBullish
OpenAI has raised huge capital and is a mechanism for turning money into growth.
Casado says OpenAI and Anthropic have probably raised more money in aggregate than the entire downstream ecosystem combined. He argues that they are mechanisms for turning money into growth, which is why capital is flowing to them. He also notes that OpenAI's original thesis was that solving intelligence would make the business model easy.
“if you look at Anthropic and OpenAI, they've probably raised more money in aggregate than the entire downstream ecosystem combined by a factor of whatever. They are these mechanisms of turning money into growth.”at 39:15
Invest Like The Best: Martin Casado on Frontier Model Economics, AI's Slowdown, and the Cursor Deal - [Invest Like the Best, EP.498] ·
AnthropicBullish
Anthropic has raised huge capital and is a mechanism for turning money into growth.
Casado says Anthropic and OpenAI have probably raised more money in aggregate than the entire downstream ecosystem combined. He argues this is what makes them powerful: they are mechanisms for converting capital into growth. He sees the frontier labs as the clearest place where extra capital produces value, which is why he thinks capital markets need to understand this shift.
“if you look at Anthropic and OpenAI, they've probably raised more money in aggregate than the entire downstream ecosystem combined by a factor of whatever. They are these mechanisms of turning money into growth.”at 39:15
Invest Like The Best: Martin Casado on Frontier Model Economics, AI's Slowdown, and the Cursor Deal - [Invest Like the Best, EP.498] ·
Debt is running out, so non-investment-grade AI cloud firms will struggle to borrow and need customer prepayments.
Casado says he is not a debt investor but reports what he sees on the ground: debt is getting scarce. He asks whether a large pension fund would lend to a random neocloud or buy T-bills. He expects investment-grade companies to be favoured, so scale wins again and smaller non-IG players will find debt harder to raise. He argues that if their business model depends on debt, they will need to get customers to prepay, which means the availability of cash and capital ultimately flows through the whole system.
“I'm not a debt guy, but here's what I'm seeing on the ground. We're running out of available debt. And if you're the Norway pension fund, are you gonna give debt to Random Neo Cloud or buy T bills?”at 39:27
Invest Like The Best: Martin Casado on Frontier Model Economics, AI's Slowdown, and the Cursor Deal - [Invest Like the Best, EP.498] ·
AI infrastructure & data centers
Power, chips and materials, not capital or debt, are the real constraints on AI compute.
Casado says capital and debt are not the binding limits on AI compute. He says the real bottleneck right now is GPU supply, along with power and materials. He argues that this is why scale providers will capture most of the market. Because the constraint is physical rather than financial, adding more money does not solve it quickly.
“The actual real limitation right now is [GPU] supply. That's it. It's not capital. It's not debt. ... The real constraint right now is power. It is chips. It is materials.”at 41:43
Invest Like The Best: Martin Casado on Frontier Model Economics, AI's Slowdown, and the Cursor Deal - [Invest Like the Best, EP.498] ·
AI infrastructure & data centers
About 80% of AI compute will go to scale providers, with the remaining 20% fragmented into niche neocloud operators.
Given that power, chips and materials are the binding constraints, Casado expects about 80% of the AI compute market to go to the large scale providers. The other roughly 20% will sit in hard-to-reach niche places. He gives the example of a GPU marketplace whose GPUs were in converted chicken sheds. He argues that the market is large enough to support a wide, fragmented set of small players in that niche, because it is not worth the big players' time to serve them.
“With that sort of a limitation, 80% of these things are gonna go to the scale providers. 20% are just gonna be kinda hard to reach places, nichey type places.”at 42:38
Invest Like The Best: Martin Casado on Frontier Model Economics, AI's Slowdown, and the Cursor Deal - [Invest Like the Best, EP.498] ·
AI infrastructure & data centers
Data center permitting pressure could push AI buildout offshore or into space if US construction stays blocked.
Casado cites around 500 proposed bills to stop data center construction in the US, including in Texas. He says he has heard from many top data center operators that it is very hard to break ground right now. If that regulatory pressure continues, he expects the buildout to move either international or into space. He sees orbital compute as a regulatory escape route as much as a physics one.
“If that regulatory pressure continues and you actually can't break ground and I've heard from many of the top data center people like it's very tough to break ground right now, then either you go international or you go into space.”at 43:53
Invest Like The Best: Martin Casado on Frontier Model Economics, AI's Slowdown, and the Cursor Deal - [Invest Like the Best, EP.498] ·
Private markets are very efficient and price deals correctly, so VCs cannot find cheap deals.
Casado argues that as a capitalist he has to believe markets are efficient, and that private markets figure out price quickly. He says many people think they can find non-consensus deals in private markets, but competition for good deals is intense. He says he pays the market price and worries less about negotiating a discount. His focus is on lowering risk through diligence, not on finding a better price. He suggests that the best-performing parts of his portfolio were competitive deals.
“I think private markets are very efficient and private markets figure out the price. And so I just don't try and bet against the market.”at 44:39
Invest Like The Best: Martin Casado on Frontier Model Economics, AI's Slowdown, and the Cursor Deal - [Invest Like the Best, EP.498] ·
Calling AI a bubble is unfounded; market prices are efficient and reflect real value, so bubble talk is overblown.
Casado says he does not see much value in bubble talk, and that the market price is the market price. He argues that if everyone is calling something a bubble, it probably is not one, because investors keep buying when they think there is still upside. He ties this to his belief that markets are efficient and that tech creates value. He says investors who call bubbles while investing in private markets are inconsistent. He does concede that valuations can be too high for good companies to earn good returns.
“I just don't see so much hand wringing about bubbles or this or that or the other thing. The market price is the market price.”at 47:30
Invest Like The Best: Martin Casado on Frontier Model Economics, AI's Slowdown, and the Cursor Deal - [Invest Like the Best, EP.498] ·
More money in private markets expands the investable TAM and creates value rather than oversaturating them.
Casado disagrees with the view that venture is not an asset class, that only about 20 startups matter, and that too much money means overfunding. He notes that private-market money has grown enormously over the last twenty years, and value has kept pace, with massive private companies and very large exits. He argues that more private money means more opportunity and more companies choosing to stay private, so private markets have an unlimited TAM rather than a limited one.
“So I do believe money in private markets creates TAM. It creates the investable TAM. I think it creates value.”at 50:00
Invest Like The Best: Martin Casado on Frontier Model Economics, AI's Slowdown, and the Cursor Deal - [Invest Like the Best, EP.498] ·
Anysphere Inc. (Cursor)Bullish
Cursor's combination with SpaceX made sense: SpaceX had compute, Cursor had data and a model-building team.
Casado says Cursor's model-building plans needed far more compute than most companies can access, which drove strategic discussions. Cursor had by far the most users and data, and a team that had successfully built models and a strong code practice. SpaceX had the compute. He describes the resulting alignment as straightforward, and says the raise running at the same time helped set the price.
“SpaceX had the compute and Cursor did have the data and had a team that had very successfully built models and built this practice and code was very important. So there was just a lot of alignment.”at 56:11
Invest Like The Best: Martin Casado on Frontier Model Economics, AI's Slowdown, and the Cursor Deal - [Invest Like the Best, EP.498] ·
Strategic M&A works on trust and good faith; leaking deals or playing games destabilises them.
Casado argues that strategic deals are high-stakes enterprise sales, not auctions, because buyers are buying an option on a large market. The company has to show the base case and the inductive step. He says the most sophisticated counterparties will see through games, so investors who leak deals or bring in competing bidders are wrong. He recalls deals that fell apart because boards pushed founders to leak. He expects more strategic exits as tech timelines compress, so people need to understand this process.
“If it's strategic deal for a lot of money, I think you need to view it very professionally. I think you need to develop a trust relationship. I think you need to do it in good faith. You should not leak these deals”at 58:11
Invest Like The Best: Martin Casado on Frontier Model Economics, AI's Slowdown, and the Cursor Deal - [Invest Like the Best, EP.498] ·
Late-stage investors entering early-stage startups is a mismatch; VC should separate private-market roles.
Casado says the boundary between startups and public-market-style investing has blurred, because the largest private companies now rival public companies from twenty years ago. As a result, late-stage investors have been putting money into early-stage companies. He cites 2021, when PE investors invested in Series P companies and sat on boards with him, and says they tried to reason about companies through burn rate, which did not fit. He argues the industry needs more self-awareness and a return to traditional productive VC for early stages.
“I think maybe it just actually makes sense to start delineating a bit more on the private side. Or if you have later stage people investing in earlier stage companies, maybe a little bit more self awareness.”at 59:42
Invest Like The Best: Martin Casado on Frontier Model Economics, AI's Slowdown, and the Cursor Deal - [Invest Like the Best, EP.498] ·
Robotics & autonomous vehicles
Generative-AI-style data scaling may not carry to robots; some tutors may stay needed.
Casado contrasts generative AI, where training on one data distribution yields outputs people want, with robots, where the economically valuable tasks may be high-dimensional and require data slightly outside the training distribution. He argues it is not clear this carries over to robots. He suggests that even if 90% of tasks can be solved, the last 10% may need human tele-operators for a while, because collecting enough data for that tail would take too many tutors over time.
“It's not clear that this is actually the case with these robots, right? You could train it on a distribution of data, but then actually what you need is just slightly out of that distribution.”at 13:03
a16z Podcast: The Robots Already Working in American Factories | Tutor Intelligence ·
Shows
- Invest Like The Best · 1 episode
- a16z Podcast · 1 episode
Ask your AI what Martin Casado thinks
Chorus gives Claude, ChatGPT and other assistants every view Martin Casado has shared on the shows it covers, next to everyone else's. Ask where they stand, what changed, and who disagrees.
Chorus reports what speakers said. It isn't investment advice.