Why AI Isn't a Bubble: In-depth Reflections from the Founder of a16z on Demand, Investment, and Judgment

Podcast: a16z

Compiled by: Deep Thinking Circle

If someone told you that the current AI craze is just another bubble, would you believe them? Soaring valuations, flooding funding, everyone talking about AI—it certainly seems like history is repeating itself. But after listening to a recent presentation by Ben Horowitz, my thinking completely changed. The co-founder of Andreessen Horowitz, drawing on his 16 years of experience managing a top venture capital firm and his deep understanding of the tech industry, offered a thought-provoking answer: This time it's different. Not because the technology is so cool, but because the demand is more real than ever before.

I've been pondering a question: why do some investors consistently find great companies, while most rely on luck? In this presentation, Ben revealed some perspectives I'd never considered before. He discussed how to manage a team of people smarter than you, how to make the right judgments amidst uncertainty, and why the current AI market is unlike any previous technology cycle. These insights are valuable not only for investors but also for anyone trying to make decisions in a rapidly changing environment.

The Art of Management Genius: Why GPs Are Not Ordinary Employees

Ben shared a point that really struck me: managing a GP (General Partner) is completely different from managing a company. He said, "We have a higher density of talent here than any other company, purely from an IQ perspective. If you look at people like Chris Dixon, Martin Casado, and Alex Rampel, they've all run companies, and it's incredibly difficult to gather so many highly intelligent people into a single executive team." This statement made me start rethinking what true talent management really is.

I think this point touches on a rarely discussed truth: when you're managing a group of people who are world-class experts in their respective fields, traditional management methods completely fail. Martin Casado has been one of the best architects in the web software field over the past 20 years, and is also a brilliant investor. Ben said he doesn't give Martin much direct instruction, but rather helps him understand the investment decision-making process and how the dialogue process influences the investment process.

This made me think about a deeper question: what exactly is the value of a manager in knowledge work? If the people you manage understand their field better than you do, your role is no longer to tell them what to do, but to help them maximize their value within the right framework. A key point Ben mentioned is that the biggest mistake in investing is getting too caught up in a company's weaknesses instead of focusing on what they truly excel at. This idea seems simple, but it's extremely difficult to put into practice.

I often see this in my own work: when evaluating a project or partner, it's easy to fall into a "problem-finding" mode. We list all the possible risks, all the places where things could go wrong, and then hesitate amidst these worries. But Ben emphasizes a different direction: the question you should ask is, is this team the best in the world at something? If so, then it's worth investing in. If not, even if they're good in many ways, they're not necessarily a good investment.

This shift in mindset is actually quite radical. It means abandoning the standard of "overall excellence" and instead seeking the standard of "extreme excellence in a particular area." In a world full of uncertainty, having one world-class skill is far more valuable than having ten decent skills.

The essence of judgment: knowledge plus wisdom equals correct decision-making.

When discussing decision-making, Ben offered a formula that I find very accurate: decision-making ability equals wisdom plus knowledge, and knowledge often resides with those who do the actual work, not managers. He said, "If you think about what decision-making is, what makes you good at it? It's a combination of intelligence and judgment, or rather, judgment is a combination of intelligence and knowledge. So it's about what you know, and how smart you are to translate that into sound judgment."

This made me rethink how information flows within an organization. In most companies, information goes through layers of filtering before reaching decision-makers. Each layer of filtering loses some details and adds some interpretations, so the final decision-maker often sees a highly simplified version. But Ben's approach is completely different. He spends a lot of time attending team meetings, talking directly with deal partners to understand what the accounting team is doing, what the IT team is doing, and even visiting LPs (Limited Partners) to get a detailed understanding of the situation on the ground.

He mentioned a detail that impressed me: he wanted employees to tell him immediately when they encountered a problem, even if it seemed insignificant. Because resolving these issues might only take 14 seconds, but if employees felt they "shouldn't bother him" and didn't speak up, the problem would escalate. This attention to detail isn't micromanagement, but rather ensuring one has enough knowledge to make the right judgments.

I believe there's a deeper insight here: in a rapidly changing environment, the quality of decision-making depends on how accurately you understand reality. If your information is secondhand, filtered, and delayed, your decisions are like driving a car with a blurry rearview mirror. Ben emphasizes that what leaders often need is not "correctness," but "clarity." Organizations need a clear direction, even if that direction isn't 100% perfect; as long as it's clear, the team can act.

The wisdom of vertical integration: Why a basketball team can't have 50 players

Regarding Andreessen Horowitz's vertical strategy, Ben shared a key insight. He recalled a conversation with the late Dave Swensen in 2009, where Dave said that investment teams shouldn't be much larger than a basketball team. A basketball team has about five starting players because conversations about investing need to truly become conversations. This analogy made me rethink the relationship between team size and efficiency.

Ben said, "I always remembered that we really didn't want any investment team to be much larger than this, so how did we maintain that? The only way was to verticalize." At the same time, software was eating the world, and they had to get bigger to cope with the market, but he didn't want the team to be bigger than a basketball team. This contradiction drove the formation of the vertical structure.

I think this insight is very profound. In many organizations, teams naturally expand as the business grows. But the problem with expansion is that genuine dialogue becomes impossible. When there are 20 people in a meeting room, discussions turn into performances, with everyone waiting for their turn to speak, rather than truly listening and thinking. By verticalizing, Andreessen Horowitz has maintained small teams for each investment firm, thus preserving high-quality dialogue and decision-making.

What's even more interesting is how they handle collaboration between vertical teams. For very close teams, such as AI infrastructure and AI apps, they have people from each team attend meetings of the other team to establish direct connections. In addition, they take all GPs out for meetings twice a year; there aren't many agendas, just opportunities for everyone to network.

Ben also mentioned a cultural perspective that impressed me. He said many people from other companies have reported that even though Andreessen Horowitz is quite large, it has less internal politics than a small company with only 10-11 people. This is a result of culture. Either you reward political behavior, and then there will be coups, infighting, mutual dislike, etc.; or you discourage political behavior, and that's how they do it.

I believe this reveals a core principle of organizational design: structure should serve the behaviors you want, not the other way around. If you want genuine collaboration and high-quality dialogue, you need to design a structure that allows these behaviors to happen naturally. Verticalization is not just a change in organizational charts; it's a well-thought-out choice to maintain the flexibility and high-quality decision-making of small teams while serving a large market.

The Art of Market Selection: Not Too Early, Not Too Late

When discussing how to choose a vertical market, Ben mentioned an interesting example: they rejected investments related to ESG (Environmental, Social, and Governance). He said, "Investing is difficult enough; there's no need to introduce criteria other than 'this thing will become a giant company and make a lot of money.'" This perspective made me rethink the key issues in investment decisions.

Ben explained that when the team came up with the concept of American Dynamism, his first question was, "Is this a marketing idea or a fund idea? I wanted to know if it was a fund idea, which is how I make money. We have investors, and we have to make money. That's a great marketing story, but we're not going to do all that. A fund's focus is narrower than marketing." Ultimately, they identified three core verticals that were indeed experiencing technological change.

I think this thought process is very valuable to learn. Often, people are drawn to a good story or concept but forget to verify whether there are real economic opportunities behind it. American Dynamism sounds cool, but what really matters is: Is there genuine technological change in this field? Are there outstanding entrepreneurs? Can it generate significant economic returns?

Ben emphasized that you can't choose a market too early or too late. It's a bit of an art. He said he was very confident that they had chosen the right market because there was a lot of very interesting activity in all those markets. But just because they were there and were Andreessen Horowitz didn't mean they would win that market. They had to constantly evolve their team and mindset to ensure they could win.

I think there's a key insight here: market opportunity and execution capability are two different things. Even if you find the perfect market, you'll still fail if your team, product, and strategy don't match. That's why Ben said they need to keep evolving, rather than thinking they've already won.

AI is not a bubble: Unprecedented Demand Intensity

When discussing whether AI is a bubble, Ben offered a perspective that impressed me. He said, "I get a lot of questions about an AI bubble, and I think one of the reasons people are so worried about it is is because valuations have gone up so fast. But if you look at what's happening at the bottom, customer adoption, revenue growth rates, and so on, we've never seen demand like this before. So we've never seen valuations rise like this, but we've also never seen demand rise like this."

This perspective made me rethink what a bubble is. A bubble shouldn't just be defined as a rapid price increase, but rather as a price deviating from fundamentals. If demand growth and price growth are aligned, then it's not a bubble, but rather the market's reaction to true value. Ben gave the example that even Nvidia's multiples aren't outrageous, especially when you look at its growth rate, profit scale, and so on.

I think there's a deeper insight here: many people compare AI to previous tech bubbles, such as the dot-com bubble of 2000. But they overlook a crucial difference: back then, many companies had no revenue, no clear business model, and their valuations were entirely based on future possibilities. Today's AI companies, however, have real customers, real revenue, and real growth.

Ben mentioned that, in his career, this is the largest technology market he has ever seen. Not the most promising, not the most hyped, but the largest. This judgment is based on the speed of customer adoption, revenue growth rate, and market demand strength they have observed. These are measurable, real metrics, not castles in the air.

I believe this perspective is crucial for understanding the current AI wave. Many people see it as a bubble because of high valuations, but they haven't looked deeply into what's happening on the demand side. If demand is indeed this strong, then high valuations may simply reflect the market's reasonable pricing of that demand. Of course, this doesn't mean every AI company is worth investing in, but it does mean that the AI market as a whole is not a bubble.

Limitations of the base model: Why application layer complexity is more important

In his presentation, Ben mentioned a point that I think is very important but often overlooked: Three or four years ago, people thought that large-scale foundational models would be giant brains capable of doing anything, better than anyone else. But that's not the case. He said, "Large-scale models do provide very important infrastructure, and all our companies are built on top of them to some extent. But for any specific use case, not just the long tail of the scenario, but also the fat tail of human behavior, it's ultimately something you have to model and understand very, very well."

He cited Cursor as an example. Cursor contains 13 different AI models, all of which model different aspects of programming, such as how you program, how you talk to a programmer, and so on. These models are so important that they actually released their own base model specifically for programming and coding. So they have a coding model that you can replace with Anthropic or OpenAI if you want, or you can use the OpenAI or Anthropic model with their other models.

This example gave me a profound understanding of the importance of the application layer. Many people believe that whoever possesses the largest and most powerful base model will win the entire AI market. However, the reality is that the complexity of applications is inherently very high and is not contained within the base model. Cursor's success is not solely due to its use of a good base model, but also because it understands the programmer's workflow and has built 13 specialized models to handle different scenarios.

Ben also mentioned a great article by Justine Moore from their team about why there isn't a god-like video model. This article delves into why different use cases ultimately require different models, which again differs from expectations four years ago. I think this reveals an important technological trend: the balance between versatility and specialization.

My understanding is that the foundational model provides a strong starting point, but real value creation occurs at the application layer. Just as the internet provides infrastructure, but the real value is created by the companies that build upon it. In the AI era, the foundational model is the infrastructure, but the innovation space at the application layer is enormous. This also explains why Ben believes there will be more winners, because the design space is vast, far exceeding anything we've seen in the technology field.

A New Balance of Ownership: The Magic Number of 20%

When discussing ownership, Ben mentioned an interesting statistic: they acquired 20% or higher ownership in many recent investments. While they didn't reach that level in some companies, those companies appreciated so quickly that the results were also very good. He said, "There have always been very, very special founders who, at some point, well, you know, that's reality, but for us, for a lot of core infrastructure, core applications, and so on, ownership has always been quite reasonable."

This made me think about the true meaning of ownership in venture capital. Many people believe that VCs are all about pursuing the highest possible ownership percentage, but Ben's view is more nuanced. For truly exceptional companies and founders, ownership may be diluted, but if the company grows fast enough, a 20% stake in a $10 billion company is more valuable than a 40% stake in a $1 billion company.

Ben also discussed the future of the VC industry. He said that while there are now over 3,000 VC firms, very few can truly help companies succeed. "Building a company is still very difficult. If you're just an engineer, an AI researcher, you invent something and jump into this world, it's a very competitive world. Having a financial partner who can help you build the company, is the initial valuation more important, or the partner? I think most smart entrepreneurs realize it's the partner."

I think this perspective is particularly important in the current environment. With the development of AI tools, the transition from idea to product has become much easier. This is why Andreessen Horowitz has increased its investment in the Speedrun accelerator. They want to closely monitor startups that are just beginning and not yet fully qualified for VC funding.

My view is that the rules of the game regarding ownership are changing. In the past, VCs might have been more interested in owning larger stakes, but now it's more important to find truly exceptional companies and then ensure you can win the opportunity to work with them. Even if this means taking on a smaller stake, the rewards can still be enormous if the company is great enough.

Why AI will produce more winners: The scale of new computing platforms

When asked why AI would create more winners than previous technology cycles, Ben offered a thought-provoking analogy. He said, "AI is a new computing platform. So you have to think of it as how many winners have been built on computers. That's the scale." He pointed out that if you ask how many businesses were built in the internet age, the number is actually quite large, from Meta to Netflix to Amazon to Google, and so on—these are very, very huge winners.

He believes that products in the AI field are having a greater economic impact. Therefore, he anticipates more companies will be worth over $1 billion, over $10 billion, surpassing the value of the previous era. But this represents a vast design space, a massive design space we've never seen before in the technology sector.

This perspective made me rethink the nature of AI. Many people see AI as a tool or technology, but Ben positions it as a new computing platform. This means that AI is not an application on existing computing platforms, but rather a completely new level, like personal computers and the internet. On this new platform, the possibilities are limitless.

I think this perspective is very important because it changes our understanding of the competitive landscape. If AI is just a tool, then perhaps only a few companies can master this tool and dominate the market. But if AI is a platform, then thousands of companies will build different applications on this platform, solve different problems, and serve different markets.

Ben mentioned that they had never seen such demand before. This isn't just hype; it's real customer adoption and real revenue growth. The strength of this demand indicates that AI is solving real problems and creating real value. And when a technology can create real value, the market will naturally support multiple winners because the space for value creation is large enough.

Giving people a chance: the ultimate mission of technology

Ben shared a profound insight that left a deep impression on me. He and Mark Andreessen believe that the best thing society can do for a person is to give them a chance. Give them a chance at life, a chance to contribute, a chance to do something bigger than they did themselves and make the world a better place. That's the best thing society can do.

He said, "If you look at what's good in human history, what's beneficial for humanity is when people have the opportunity to do something bigger than themselves and make a contribution. There are many systemic ideas, like if we could create a utopia or make everyone equal or something like that, but if you look at the history of communism or whatever, it ultimately does the opposite. It ultimately results in everyone having an equal opportunity but not getting it."

I believe this perspective touches upon the essence of technology investment and entrepreneurship. We're not just chasing financial returns; we're creating opportunities. Every successful company creates jobs, creates products, solves problems, and ultimately gives more people the opportunity to realize their potential. It's no coincidence that the rise of the United States coincided with the rise of free markets, capitalism, and the rule of law.

Ben points out that if you look at human history, wealth, life expectancy, and the size of the global population have all grown remarkably in the last 250 years. The United States has played a crucial role in this. And today, the United States remains the country and system where people are most likely to have a real chance at life. In order for the United States to maintain its importance in the world, it must win economically, it must win technologically, and it must win militarily, which means it must win technologically.

Their job is to help the nation win technologically. This is important not only to them and to the nation, but also to humanity. I think this perspective elevates investment to a higher level. It's not just about making money, but about participating in the grand narrative of human progress.

Ben gave a concrete example of how this philosophy drives action. He and Jen recently went to Mexico, largely because a junior member of the team said, “What we’re doing is very important. We need to help this alliance. We need to help protect the border. We need to help our own defense manufacturing. We must help solve the energy problem. I’m going to get this meeting.” And then they got the meeting.

I believe this reveals a profound truth: if you want to change the world, you must believe you can change the world. This is not arrogance, but a necessary belief. Without this belief, you won't take those seemingly impossible actions. And it is these actions that ultimately create real change.

The Return of M&A: AI Forces Everyone to Rethink

Ben offered an interesting perspective when discussing the M&A (Mergers and Acquisitions) market. He said, "AI is such a disruptive phenomenon that every company, every existing business, is threatened by AI. Therefore, many ways to address this threat are to acquire the DNA of the future. So I think there will be a lot of M&A because I think people need to restructure the way they work to survive."

This perspective made me rethink the impact of AI on existing companies. Many people focus on AI startups, but Ben points out that existing large companies also face immense pressure. If they cannot adapt to AI quickly, they may be replaced by emerging AI-native companies. And one of the fastest ways to adapt is to acquire companies that have already mastered AI technology and the mindset.

I think this explains why the tech M&A market is reopening. In the past few years, large acquisitions have been relatively few due to regulatory and other reasons. But the emergence of AI has changed the game. Existing companies can no longer learn and adapt slowly; they need to acquire capabilities quickly, and acquisitions are the most direct way.

What does this mean for startups? I think it creates a new exit path. For companies that have built strong AI capabilities but may not become independent giants, being acquired could be a good outcome. For large companies, it's also a survival strategy.

My view is that this increase in M&A activity is actually healthy. It indicates that the market is functioning effectively and resources are flowing to where value is most created. At the same time, it provides entrepreneurs with more options; not everyone wants or needs to build a multi-billion dollar independent company.

My thoughts: Judgment is the scarce resource.

After listening to Ben's sharing, my biggest takeaway is that in this era of information overload and rapid change, what is truly scarce is not information, not capital, not even technology, but judgment. Judgment is a combination of knowledge and wisdom, the ability to make the right choices amidst uncertainty.

Ben's approach to managing Andreessen Horowitz has been very inspiring. He doesn't manage by establishing detailed rules and processes, but by cultivating sound judgment. He spends a lot of time understanding the details, not for micromanagement, but to ensure he has enough knowledge to make good judgments. He focuses on the GPs' performance at the moment of investment decision-making, rather than waiting 10 years for the results, because 10 years is too long.

Regarding the debate about whether AI is a bubble, I now have a clearer perspective. The definition of a bubble shouldn't solely focus on price, but rather on the relationship between price and value. If demand is real, growth is real, and value creation is real, then high valuations may simply be the market's reaction to this reality. Of course, this doesn't mean every company is worth that price, but as a whole, the AI market reflects genuine technological change and business opportunities.

I've also come to a deeper understanding of why the foundational model isn't everything. The value of technology ultimately lies in how it's applied, how it solves real-world problems, and how it serves users. The foundational model provides possibilities, but innovation at the application layer will determine who truly wins the market. This is why there will be multiple winners, because the design space for applications is enormous.

Finally, Ben's philosophy of "giving people a chance" made me rethink the meaning of technology and investment. We're not just chasing returns; we're creating opportunities, helping people realize their potential, and driving human progress. This sense of mission isn't an empty slogan, but a real force driving action. You only take world-changing actions when you believe you can change the world.