a16z Investment Review: 86% of VC Deals Flowing to AI, the Trillion-Dollar Feast is Just Beginning

Author: Z Potentials

Z Highlights

• The gap between what models can do and how they are actually used is exactly why I am so excited about the application layer.
• The median spending of the top 1% of companies on AI tools is about eight times that of the top 10%.
• For founder-led companies, the founder themselves is an asset class.

This article is compiled from the official a16z podcast "AI, Infrastructure, and the Next Investment Cycle." David George, Sarah Wang, Alex Immerman, and Santiago "Santi" Rodriguez are four partners at the a16z Growth Fund. This episode discusses the key charts from the annual report "State of Markets," focusing on the profits behind the market rise, the arms race in computing power among mega-vendors, the extremely low adoption rates at the model and application layers, and the latest observations in consumer, public software markets, IPOs, and secondary markets. The episode was released on September 30, 2026.

01 Technology is the "Everything Cycle"

David: Welcome back to the a16z podcast. I’m David George, and joining me today are my colleagues Sarah Wang, Alex Immerman, and Santiago Rodriguez. In this episode, we will go through the 25 key charts from the latest "State of Markets" report: the profits behind the market rise, the scale of AI construction, evidence of adoption rate growth, and what this cycle means for hardware, software, and the next generation of private companies. We will explain the implications behind the charts and discuss what we are seeing on the front lines of businesses. Sarah, Alex, Santi, thank you for being here.

On the same day this podcast goes live, we will also release the full "State of Markets" report. This is our growth team’s annual tradition of compiling the biggest trends in technology, AI, infrastructure, and the entire market into one document, and today we will highlight some interesting charts from it. The areas we will discuss include macro—CapEx, data centers, and accelerating demand; as well as frontline observations—models, applications, and a deep dive into several verticals. Technology is driving an investment boom across the entire economy: profit growth is supporting the market rise, and AI demand is pushing infrastructure spending far beyond previous forecasts. Mega-vendors are pouring nearly all of their recent operating cash flow into capacity building, while demand continues to exceed supply in almost every case we see. This money is flowing into chips, power, cooling, construction, and skilled labor, among others. This expansion coincides with a greater need to transform physical infrastructure, creating opportunities across various industries.

Alright, let’s dive in. Page one: technology is the "everything cycle." This may seem somewhat obvious, but the numbers on this page look quite astonishing today—high-tech equipment, software, and R&D now account for about 55% of U.S. capital expenditures, which is a staggering figure. Technology is driving the investment cycle across all areas of the economy, from software and models to power, construction, and industrial capacity. Tech stocks account for about 40% of the total market capitalization of the U.S. stock market, with eight of the top ten companies by market cap globally being U.S. tech companies. This is a grand narrative; you can see it in CapEx, in data center construction, and in the capital flowing to model companies—model companies have collectively raised over $350 billion. Looking at a longer history, because this is the most cited analogy: the scale of this construction as a percentage of GDP has already surpassed that of the railroads back in the day. We all have a sense that if we fast forward five to seven years, or even ten years, and look back at these numbers, the cumulative value will likely be more than 20 times what it is now. So, this really does feel like the next chapter of Marc Andreessen's "software is eating the world."

Alex: Over the past 15 years, software has transformed all these industries, but at that time, only a small number of people could build software—we weren’t among them. And today, look at us: each of us has several automated tasks running every night. If you extrapolate that logic for software demand, it implies a much greater need for computing power, chips, electricity, and construction. So it’s not surprising to see that most of the investment is concentrated in technology. We often get asked the same question by our audience: this all sounds great, but is it a bubble? The market has reached new highs, but the current trading multiples have actually decreased—stocks have risen about 20%, but multiples have dropped about 20%. This means that the increase is driven by fundamental profits, not by multiple expansion. The earnings multiple for the S&P 500 is below 20 times, and most of the companies that make up the S&P 500 are of very high quality. This is completely different from the internet bubble—back then, the highest market cap companies saw their stock prices soar due to multiple expansions, with many stocks trading at 100 times earnings. That is not the case now. In contrast, some storage companies—being inherently cyclical—are trading at six to seven times forward earnings. So the difference is significant.

Santi: Here’s an important addition: since the advent of GPT nearly four years ago, the market has risen by 90%, with an annualized rate of 17%. When you see four consecutive years of 17% annualized growth, the intuitive reaction is "this is bound to revert," and we are indeed in a hot period right now. But when you put it together with what was just mentioned—the market is trading at earnings multiples below 20 times, and profits are still growing at 15%—this feels indeed different from the recent year of 2021 and also different from 2000, when multiples and growth did not truly synchronize.

David: I completely agree. I mentioned the scale of CapEx construction in the "railroad analogy" before. Just looking at mega-vendors—Alphabet, Amazon, Meta, Microsoft, and Oracle—their CapEx for 2026 is about $780 billion, while it was $416 billion in 2025; all expectations point to exceeding $1 trillion annually starting in 2027. There’s a pattern here: each generation of computing platforms supports a far larger number of users and uses than the previous one. This round of construction can happen so quickly, with demand still exceeding supply, because the user base is driven by existing distribution channels. This trend is built on the internet, cloud computing, and mobile phones, so theoretically, it can reach billions of users immediately—this sharply contrasts with previous technology cycles.

Sarah: I think this point is worth elaborating on. We have far surpassed the "asking a question occasionally" model. Think about the rise of agents: parallel tasks, long-running tasks—you mentioned before that tasks will continue running while you sleep at night or do other things during the day. If agents can autonomously write code, perform searches, and execute these tasks, then the computing power required by these platforms will far exceed anything we have ever imagined.

David: I completely agree. This chart shows what is happening on the CapEx side: the CapEx forecasts for the five mega-vendors have been continuously and significantly revised upward in a very short time. What once seemed like a ceiling, spending that would take several quarters or even years to reach, has now become a near-term target because the demand for computing power continues to expand. Looking at this chart, I think at any given point, the instinctive reaction is "investment from here will tend to stabilize." But over the past four years, as an economy, we have consistently underestimated the strength of this trend. As Sarah mentioned, with the development of models, the utilization rate of installed capacity is continuously being maximized. So the numbers David just mentioned are the current actual estimates.

Alex: As DG puts it, we all refer to the current demand for computing power as a "model buster." A compelling example we all experienced: about a year ago, Sam Altman and Sarah Friar faced a lot of criticism for their huge commitments to computing power; everyone thought they were reckless and aggressive. And now, everyone would say that decision was extremely prescient. Even so, just two weeks ago, we saw OpenAI having to pause subscriptions for the Pro plan—a $2000 level service—due to overwhelming demand. What we are experiencing is an insatiable demand.

David: Everyone in every link of the supply chain, and everyone we’ve talked to, has been telling us different versions of the same thing: demand exceeds supply. In some segments of the data center supply chain, materials or products won’t be available until 2028. So this tension will not ease. At the same time, you can also see evidence of high-quality businesses from the demand side: Microsoft, Google, and Amazon collectively have about $1.7 trillion in cloud backlog orders. These customers are committed to rapidly accumulating, while platforms are heavily investing to service this demand. So, while free cash flow is pressured during the current construction phase, consensus forecasts show that it will rebound starting in 2028 and then grow significantly. You can look at Amazon’s latest earnings call; they explained this J-curve dynamic very well: the actual lifespan of GPUs (or TPUs) is quite long. You first have to build the data center "shell," which takes time; then you buy chips, but the economic lifespan of chips is very long and longer than any of us expected.

Sarah: Yes, I know some mega-vendors have indeed faced criticism for leveraging debt for CapEx, but similar to the dynamics of model labs, those who hesitated and were more conservative later regretted it. You recently discussed Microsoft in the podcast with Gavin, but I think this is an industry-wide issue.

Santi: Yes, another frequently discussed signal is the spot market prices for existing GPUs: any GPU that can go online for computing power can be rented out at attractive prices, and mega-vendors can achieve considerable returns. This is the key point we often discuss: every time such a wave occurs, it creates enormous user and consumer surplus. As far as we can see, consumers and users are gaining tremendous value from this—which is also why they are using it so heavily. The vendors servicing these users are making good profits. Then, going down the line to chips—the cost of acquiring chips is much higher than it was 12, 18, or 24 months ago, but you can still achieve high margins and create a lot of surplus from users.

02 A New Era of Atoms

David: For the capex of super-large manufacturers, we should view it as someone else's order book. These historically profitable platforms are continuously investing cash into AI infrastructure. As mentioned earlier, this temporarily depresses their free cash flow, but it is a huge boon for chip orders, electricity, and construction—this is also why this broader wave of technology has turned into an industrial boom. We are spending more and more time looking at the companies servicing this industrial boom, along various links in the data center supply chain. Interestingly, traditionally our world has been more about monetizing existing IP—build once, sell infinitely; but these businesses have to manage much more complex affairs: financing, managing supplier relationships, forecasting capacity, and predicting demand. We see some teams excel at this, while others struggle or even pause. This is slightly different from the expertise we usually spend our time on. This brings us into "the new era of atoms." This page shows big numbers: by 2040, global infrastructure investment demand is estimated to be $90 trillion. But importantly, this far exceeds AI and data centers, also including electricity, water, roads, and transportation. We can see this everywhere in our portfolio: Anduril's massive manufacturing facility, equivalent to 87 football fields; Waabi is aggressively expanding their transportation hubs; and SpaceX's multi-billion dollar investment in Louisiana. Betting on the future today, from our perspective, is different from the past. As Santi said, this means building factories, expanding infrastructure, and increasing skilled jobs across the country.

Santi: When it comes to data centers, investing in physical infrastructure is very different from investing in software. Elon proposed the idea of "factory as product" many years ago. Many of the excellent founders we support come from Tesla or SpaceX, which is not surprising because they know that, at the execution level, factories will ultimately become a huge competitive advantage in the scaling process. We often hear many myths about data centers: they will deplete America's water resources; wealthy people do not want to live nearby; and our favorite—my electricity bill is going to skyrocket. But what seems counterintuitive is that data centers can actually help lower your electricity bill. A recent study in the U.S. shows that for every 10% increase in data center capacity, residential electricity prices drop by 40 basis points. You can think of it this way: the power grid is a shared fixed cost space—utility poles, cables, substations. A large and stable customer like a data center can spread these costs over more units of electricity. Simply put, more demand on a shared system is a good thing.

David: Yes, I completely agree with you. Investing in shared infrastructure can also improve the economic benefits for the households and businesses connected to it.

Sarah: This point has not received enough media attention. Dina Powell recently mentioned Meta's Louisiana project in a podcast, where they collaborated with the community to lower electricity prices. There are already real examples today; this is not just talk.

03 Model Layer and Application Layer: Significant Revenue, Extremely Low Adoption Rate

Sarah: I completely agree. Now let's talk about the trends in the model layer and application layer. On one hand, as we have forecasted, AI is creating significant revenue and savings; on the other hand, adoption rates are still in the very early stages. Based on this, the trend is very clear: costs are plummeting. We have already forecasted intelligent agents, and now intelligent agents are becoming economically viable for a broader range of work. And as we speak, the landscape is still changing: new innovations like Jev from labs like TypeSafe are pushing this to extremes. You will see that a two-order-of-magnitude cost difference will really affect the number of use cases. This is somewhat like the classic Jevons paradox, so the name is quite fitting. Overall, this is a great landscape for both the model layer and application layer. These improvements are creating real opportunities for software companies in both growth and profitability. It’s not to say that everyone will win, but it is not surprising that AI is attracting venture capital into an increasingly broad range of industries. We are also seeing more companies achieving massive scale in the private market. This is not a new chart, but it is one of my favorites: the combined revenue scale of OpenAI and Anthropic. In the past few years, we have shown this chart at every GP internal meeting, updating it every month because it becomes outdated too quickly. The key point you can see in the numbers: the annualized revenue of OpenAI and Anthropic has risen to an extraordinary level. We like to compare it with the greatest software companies in history—on the right, you can see that even based on the net new revenue estimates of the best software companies in history, the net new revenue of leading labs has far exceeded that. This is not just about scale, but an astonishing combination of speed.

Santi: But we are still very early. The interesting thing about AI migration compared to other platform migrations is that we are not only early in the proportion of the population using AI, but also early in the share of wallet that those users spend on AI. This is different from 2008—back then we modeled based on the "proportion of people who own iPhones"; now we need to consider how many people will use AI products and to what extent they will use them.

Sarah: Yes, the point of "still very early" is very important, and there is a set of metrics to support this. Firstly, the actual deployment rate of S&P 500 companies is 69%—if you are like us, "AI addicts," this is probably close to your expectations. But when you look at the quantifiable impact—this might be a good indicator of their deployment progress—it is 30%. Then to the ultimate metric, a time-tracked metric, it is actually only 2%. So we believe AI is delivering results, but there is still a huge space for deepening use within organizations and tracking these results over the long term. Moving from these scattered deployments to truly deep impacts and recurring workflows, frankly, is the next stage. I can say that for most companies we have communicated with, their engagement with AI still primarily revolves around Microsoft Copilot, which itself indicates how far they have to go.

Alex: Yes, the gap between what models can do and how they are actually used is exactly what excites me about the application layer. Sarah, your conversation with Ali Ghodsi and Martin from Databricks was fantastic; he mentioned that AI can understand the world very well, but knows almost nothing about your company. We are seeing this situation in the application layer companies in our portfolio—they are bridging this gap. There was an example in the past few months that impressed me: Revolut has a very mature engineering team, but they still need to collaborate with 11 Labs to integrate 11 Labs' top-notch voice model into secure connections for customer accounts and banking processes, so that when I call as a customer, my questions can be resolved smoothly, efficiently, and securely. Today's opportunities largely lie in harnessing these capabilities to build reliable services.

04 Heavy Users Are Pulling Ahead

Sarah: Besides the fact that corporate adoption rates are still early, another significant trend we want to illustrate with data is that heavy users are pulling ahead comprehensively. You can think of AI spending rising generally among companies, but those heavy users we mentioned are spending much more. We looked at some data from YipitData: the median spending of the top 1% of companies on AI tools is about eight times that of the top 10%.

Impressively, the top 1% is almost equivalent to the total of the top 2% to 10%.

Even within our own portfolio companies—you could say they are almost entirely AI-native—the spending of top users is about $7,500 to $9,000 per month. Meanwhile, the median user—also among these most AI-native companies—spends closer to the subscription price, around $200, possibly adding another subscription, totaling $200 to $400. So if you look at the spending gap between median and top users, it exceeds 20 times.

David: Yes, one question we discuss with our portfolio companies is: how much adoption do they really have, and how do they measure it? For example, if you look at the proportion of money they spend on AI tools relative to employee costs: for relatively forward-looking large Fortune 500 companies, it is about 1% today.

And for the most forward-leaning AI-native companies in our portfolio, it can be as high as 10%. So all of this is just about how early the diffusion is and how deep it will spread. We have already seen another example within the portfolio: many companies were very excited when they first procured Cursor or Cognition, but the reality is that this is just the beginning of the journey. From procurement and trialing these tools to full adoption, there is still a long runway.

05 Quantifiable Cases: Cost Side and Revenue Side

Santi: Sarah, you mentioned earlier that we are starting to see quantifiable case studies. We have named a few publicly traded companies in the report that have indeed disclosed these metrics and seen substantial improvements from AI. On the cost side: Chime disclosed that over the past four years, they have reduced service costs by more than 10% each year, compounded, meaning service costs have nearly decreased by 50%—this is supported by our portfolio company Decagon. On the revenue side, I think Shopify's example is interesting: they launched an AI assistant (sidekick) to help merchants onboard faster. As a result, the proportion of customers reaching five orders within 15 days of onboarding increased by 8%. Shopify tracks this metric because once merchants reach five orders, they tend to stay and continue using Shopify. So this is also a meaningful tailwind for their business.

Sarah: Absolutely. Speaking of the advantages brought by reduced service costs: you have more room to compete—better pricing, possibly broader services, and the ability to reinvest in growth. These benefits not only reach customers but also improve profit margins, which then reinvest back into the business. Another very fascinating phenomenon is that, of course, companies vary, and we cannot generalize, but incumbents are also doing quite well in this round. ServiceNow is a good example; they disclosed that their annual contract value (ACV) for AI exceeds $1 billion, and agentic deployments have grown ninefold. So you can see this across the entire tech stack, from incumbents to new companies.

David: We often discuss an interesting question—our most forward-looking portfolio companies, like Stripe, have talked to us about where they are actually spending incremental AI investments. Is it on building new products for customers to drive higher revenues? Or is it for optimization and efficiency improvements on the cost side? This serves as our litmus test: where do the founders and CEOs of these companies see the biggest opportunities? The upside potential for driving revenue growth is boundless, while the opportunities on the cost side...

David: Improvements on the cost side—of course, you can reinvest that back, but that will always just be a potential opportunity that will always exist. If you believe that the best use of capital today is to optimize the cost structure, what does that mean for your new revenue opportunities?

Santi: Even within cost optimization, there are different flavors. One is reducing service costs, which allows you to reinvest and become a better company. I think six months ago, our entire industry was really focused on "systems of record." If your engineers are busy rebuilding a system of record just to save a few thousand or tens of thousands of dollars, I think those resources could be better used.

Sarah: Yes, if you aggregate these, for application-based businesses, the relevant unit is actually the cost of "completing customer tasks." If you can decouple this—charging for completing tasks for customers—you can use better routing to make products more economical—this is amazing because now you’re not using the most expensive model at every step, but tasks are still being completed. I think we will see the best application companies showing more improvements in profit margins.

06 When Costs Plummet, New Markets Open Up

Alex: I completely agree. Sarah mentioned that ServiceNow has seen massive agentic usage. Agents are already here, executing tasks. Tasks require multiple steps, and multiple steps mean multiple model calls—this explains the 14-fold increase in agent token usage on OpenRouter. However, these steps are becoming increasingly cheap. We see cached applications—systems can reuse background information instead of processing from scratch every time. I see Hebbia taking full advantage of this: their financial chat workload has seen costs drop by ten times. The same budget can now support more work.

Sarah: Yes, I think your broader point is: lower costs allow agents to now check for errors and retry, suddenly opening up massive tasks—previously, reasoning or tool calls could be prohibitively expensive. This is especially important in scenarios where reliability is the key factor in deciding whether to use agents. Also, we talked about TypeSafe before; imagine when costs drop by an order of magnitude or two, the number of use cases opened up is actually unimaginable.

David: Plus, the significant improvement in latency. Absolutely.

Alex: Yes, so companies are now thinking more seriously about how to optimize latency or performance against costs at scale. Take Databricks as an example; they use routing—to select the appropriate model for each specific task—to improve performance and costs. Their intelligent routers perform better: solving more problems at a cost 35% lower than the strongest single model. We now see many other approaches like fine-tuning: Harvey and Elise AI are examples. Elise fine-tuned a smaller model, reducing costs by 60% and latency significantly, making some real-time use cases in audio scenarios feasible. These engineering improvements make AI more useful and cheaper.

07 Consumer Side: 2% Subscription Penetration and the "Smile Curve"

David: We mainly talked about usage within enterprises, but the consumer side is also very important—after all, that’s where we’ve seen some of the biggest returns and spent a lot of time. We believe the consumer side is still very early. A recent survey showed that only slightly more than 2% of American households are actually paying for AI subscriptions. Maybe subscriptions are not the best way to monetize for most households, but so far, that’s what it is. One interesting point about subscription revenue is that it has one of the best retention rates we’ve seen on the consumer side. Alex likes to talk about the "smile curve"—it’s really rare for a retention curve to not only flatten but actually rise as products improve and people come back. But that’s exactly what we see with AI, which perfectly illustrates the value that subscriptions can deliver to payers.

Sarah: Speaking of subscriptions, what shocks me about these numbers—especially looking at the bottom right graph—is how small they are. Think about Amazon Prime, which has over 200 million households. Netflix is also a good example—70 million households. These numbers made me double-check; I thought, "Is this a mistake?" So I completely agree, this is just the beginning.

Alex: Yes. Our partner Josh Elman has a saying I like, which might be a bit counterintuitive: consumer AI is something I use in my daily life, and it doesn’t necessarily need to be reimbursed. So in his view, it’s not surprising that 97% of households are using AI but haven’t paid for it yet. You’re right, Santi, I do like the smile retention curve—we love to see users coming back more frequently. Another common feature of large consumer platforms is time spent: active users on Facebook, Instagram, TikTok, and Snap spend 30 to 60 minutes a day. And today’s best AI assistants—many of which are emerging—want to become my go-to applications. I hope they can be extremely persistent, always online, and super proactive; if that’s the case, they might not be the places I spend the most time in the future. That’s right, when agents work in the background and we no longer measure engagement using external data like social media screen time, our work becomes harder, but it’s important. I remember when the latest model was released in December, even in our work, we would initiate a deep research task or agent task and then drive into town—this wouldn’t show up in traditional screen time metrics, nor is it like the engagement tracked by past generations of consumer products.

Santi: Yes, I completely agree. It’s early, but the consumer side is truly a brand new and exciting moment.

Alex: Muse and Instinct are undoubtedly new players, but they are growing rapidly. Together with ChatGPT, they are increasingly taking away my queries from traditional search.

This is a very dynamic moment, so if you are an existing consumer discovery platform or an existing marketplace, you really need to consider your next steps now. I would ask a few questions. First, how much incremental demand can I get from AI agents? How many additional orders can they bring? Second, in my business, what proportion of the profit pool comes from ownership of customer relationships and discovery? There was a lot of news last week: Amazon told Muse "no, thank you," but Instacart said "yes, please." Think about Amazon (and AWS) and Instacart: their advertising revenue exceeds all their operating profits. So advertising revenue and ownership of customer relationships are extremely important. But for Amazon, how much incremental new orders or new customers can access Muse? Not that much. And looking at Instacart: the online penetration rate for fresh produce is still relatively early, and there are many orders to compete for. The optimistic possibility is that there will be more orders coming from these applications. In the past, I had to click through various ways to place the next order; if I hand this over to an agent, suddenly there are no clicks, but hopefully more GMV. Trips I once wanted to book but didn’t will get booked; dinner orders we want to place tonight will get placed. So I optimistically believe there will be more GMV, compensating for some of the lost advertising revenue.

David: However, there’s an open question: how will it be compensated? Take Amazon as an example; it has over $70 billion in advertising business, which is a very high-margin stream of monetization, entirely based on consumers visiting the site and clicking on ads. What happens if they stop doing that? Today, Meta and Google are famously better at advertising than any internet platform, earning over $200 per user in the U.S. and developed markets. Meta is basically an entertainment application; let’s see what happens, it might be relatively safe. On the Google side—interestingly, when all this happened two years ago, the entire narrative around Google was, "Oh my, what will happen to Google’s search business?" The result has been very resilient. Part of the reason is that their monetization rate for high-margin keywords is relatively safe—those keywords are more like "I need to buy insurance" or "find me a hotel in this city."

David: Or something similar—you can directly monetize, but AI can’t do those things for you yet. If that changes, it could be a very different dynamic.

David: Yes, I want to add that last week’s narrative had a bit of a "zero-sum" flavor, very negative—"Look, these marketplaces will be affected." But I do believe the positive-sum perspective is: first, that $70 billion spent on Amazon advertising may just be channeled elsewhere—perhaps directly spent on agents in different forms, or finding better ways to reach; second, with friction reduced, we may see more consumption—people may buy more, which is a positive flywheel driving economic growth. I feel like we’re a bit trapped in the narrative of "this is detrimental to the profit pool," but in reality, it may just be good. ... Yes, both could be true at the same time.

David: The profit pool may change, but overall it’s incremental to the economy. I think if we look back three years from now, if there hasn’t been a meaningful efficiency improvement at the macro level—that would push consumer spending in advertising higher—we would all be disappointed. But it could also be that the profit pool disappears in some places and then gets redistributed.

Alex: The immediate reaction is that the net market value of the entire ecosystem is positive. Meta’s gains far exceed the losses of the marketplaces.

08 Divergence in the Public Software Market

Sarah: Yes, returning to our discussion about software, it may be shifting more towards the public market. The most prominent trend in these charts is that the composition has clearly shifted towards slower-growing, more profitable companies. About 75% of the public software sample is profitable, while only 30% are growing at 20% or more.

David: That’s an astonishing number: only 30% of public software companies are growing at over 20%—and not just 20%, but 30%.

Alex: If you draw the line at 30%, we’ve discussed that there are absolutely fewer than five companies. And in the private market, every company we see and spend time with is growing far beyond 30%.

David: And honestly, that’s a bit obvious: you go talk to IT managers, CIOs—see their growth and the money they’re spending on AI; the easiest source of that money is not spending incremental funds on new SaaS software projects.

David: We have invested in several SaaS software companies, including those in the public market. Our conversations with their founders are very focused: how can I leverage my existing distribution—often very strong, with customers locked in, as you mentioned to Santi—to apply genuinely interesting new AI products and push revenue growth even higher? I believe that to dispel the greatest fears surrounding the "SaaS apocalypse," we need to see a period where software companies continue to record numbers like 98% gross retention—this has always been a key point that makes them attractive as investment targets; continue to drive the efficiency improvements we’ve discussed; but most importantly, accelerate revenue growth. This shows that they are defensively secure and that their aggressive investments will actually improve their business.

Sarah: I find the insights you shared in that blog about "two paths" a few months ago to be very interesting as a rule of thumb. Perhaps you could share a bit more about the specific percentage details.

David: Yes, the more I write and talk with founders, the more I feel that almost everyone we’ve spoken to says, "Yes, I am trying to push revenue growth higher." This is clearly the main path for the companies we invest in and support. It’s very different from a year ago—when the fear was that "everyone would rewrite their software systems with vibe coding," and that was clearly not happening in the market. But there is a lot of pressure to deliver accelerated revenue growth. So we’ve said: aiming for over 10% acceleration is a high number, but with magical products and the budget available for AI, it seems feasible. We are looking forward to it. We have several companies that are already close, and we will see results in the next 12 to 18 months.

Sarah: Absolutely. As with everything, you can’t lump every company into the same "software" bucket and talk about multiple declines and growth declines. In fact, software has become highly differentiated: look at this chart, cybersecurity and observability really stand out, and vertical software overall is supported much better than horizontal applications. The framework we use to think about "what is holding up, or even doing well, and what is in a longer-term decline" is to really think about how AI changes customer demand for this product. The cybersecurity risks brought by AI have been fully publicized; but as more software and agents create new security and monitoring needs, this actually creates greater demand for incumbents in these markets—you can see the annotation for CrowdStrike on the right. Conversely, applications are facing varying degrees of workflow changes: one of our best CEOs, Ali Ghodsi, likes to talk about the "guillotine of AI" and what will be cut first. You can also think from the other side: what does AI actually need? Putting these two dynamics together, I think explains most of what we see in the public market.

Alex: Yes, I think this aligns very well with what we see in the private market. Just like you hinted at in the security space, agents are accessing more and more systems and taking more actions—events from OpenAI and Hugging Face have been eye-opening for many. So security is paramount. Also, look at vertical AI companies: they are the fastest-growing companies in our private market, Harvey, Abridge, Elise, growing faster than any predecessors in their respective industries. This reflects that customers realize: the key is not just the model—as we’ve discussed before—but how you orchestrate it, how you build applications around it, and the workflow requirements relevant to the vertical.

Santi: We spend a lot of time looking at the public market, not to invest, but to price exit multiples, so we track it very closely. If I summarize this year: the first few months were the "SaaS apocalypse." We wrote a blog about what types of companies would perform well. Fast forward to today, the software index has actually returned to where it was at the beginning of the year, but it has really differentiated into what are considered AI losers and winners. The public market sometimes oversimplifies things. But according to your point, AI winners are not just those who can improve cost structures, but those who are genuinely accelerating and capturing this new money; if you haven’t captured it, you might miss this wave.

Sarah: Conversely, we want to showcase some operational data from private companies to see what customers are actually doing. In Stripe’s SaaS customer data, there is actually a "counter-narrative" finding: growth accelerated into 2026 for both young and mature companies.

David: Stripe calls it "RenaSaaS."

Santi: Oh, "Renaissance."

David: I actually quite like the term "RenaSaaS."

Sarah: Yes, I like "RenaSaaS" too.

09 Between Public and Private: $2.4 Trillion in Private Champions and IPOs

Santi: We talked a bit about the public market, and Sarah mentioned the acceleration of SaaS in the private market. We are often asked: will companies stay private longer? Look at the top six companies ranked by their latest round valuations today: Anthropic, OpenAI, Databricks, Stripe, Waymo, Revolut, totaling about $2.4 trillion. This exceeds the total market capitalization of IPOs over the past decade (excluding SpaceX), which is about $1.7 trillion. So the activity level in our market alone—once again, six companies, $2.4 trillion—is almost as large as the Russell 2000 index, which is a huge public market index with hundreds of public fund managers spending most of their time. Therefore, spending time on these late-stage champions is becoming increasingly exciting; they can continue to invest in growth rather than maximizing short-term profits. But David, could you talk about how we advise companies on IPOs and when they should stay in the private market...

David: Yes, you see, regarding IPOs, especially for founder-led companies, one thing we’ve been discussing and I’ve written about is that at this point, the founders themselves are an asset class. The reason we bet on them, and part of why some companies benefit from staying private, is that they can often place larger bets in the private market, and these bets have longer return periods. Zuck and Elon are obviously exceptions in the public market. But you can see how this manifests in the numbers for companies like Databricks and Stripe: huge new bets in new product areas, followed by accelerated revenue growth. You can also do this in a public company, but it will be under stricter scrutiny—obviously, Meta’s stock price fell below $100 at its lowest point, when people were extremely skeptical about their investments in AR/VR. For us, the way we talk with founders is: an IPO is just another financing event. What do you gain from becoming a public company compared to being private? Clearly, you can achieve significant scale in the private market. Some companies are already operating like public companies: highly focused on efficiency, with basically every metric being tracked. Others may benefit from publicly disclosing finances to win customer trust at both the enterprise and consumer levels—like Navan after going public, we saw it accelerate again, partly because some larger enterprises trust them more because they are a public company.

David: Yes, of course, there are examples like this: capital needs can become very large over time, and the capital pool in the private market is very deep, capable of serving many such companies, but at some point, capital needs may become so large that the private market cannot accommodate them.

Santi: That’s the public market. In the private market, another area we spend a lot of time on is the secondary market. There are two dynamics here. First, we see more and more companies doing tenders for employees—if they choose to stay private, allowing employees to gain liquidity outside the public market. An interesting data point here is that Carta shows the participation rate in tenders is only 58%—employees choose not to take liquidity because they have great confidence in their business performance. This is in stark contrast to the narrative that sometimes "employees are cashing out"—what we see is employees choosing not to cash out because they believe so much in their company.

David: Imagine if that were 100%, what would happen to housing prices in San Francisco?

David: That would be tough. Indeed. We clearly participate in and lead these tender offers often. We believe they are a good thing. They serve two purposes. First, for employees and potential employees, it can sometimes be difficult to compete with the liquidity appeal of the public market—RSUs entering your account every quarter on a net tax basis. So this is a competitive weapon for the private market when competing for talent or retaining employees against public companies. On the other hand: we advocate for more frequent resets of valuations. This keeps your private stock price fresh for several reasons: first, it makes it easier to discuss this with employees and potential employees; second, if you want to do M&A—we have many companies that have done this—you have fresh currency to use.

Santi: I think the discount in the secondary market has been discussed a lot because it’s been the dynamic over the past five years: as we come out of 2021, 22, 23, and 24, the median discount to the latest round price in the secondary market is quite meaningful because some rounds may have been priced too high or outdated. But what we are actually seeing today is that when there are transactions in the secondary market, the discount to the latest round price is basically zero. This just shows that the financing prices for companies are fresh, and there are always new investors willing to buy at the same price—which hasn’t been the case in the past year or two.

10 Everything is AI: Next Opportunities

David: Yes, good point. This interesting slide is called "Everything is AI Computing." In this 2026 snapshot, AI-related companies account for 86% of VC transaction activity in the U.S., up from 65% in 2025. Clearly, AI has become the dominant destination for venture capital.

Alex: Yes, but under the AI label, the opportunities have actually broadened significantly: enterprise applications, consumer applications, as well as services, semiconductors, and power.

Alex: And defense. A lot of attention is focused on OpenAI, xAI, and Anthropic, but beyond that, our opportunity set has never been as deep and broad as it is now.

David: Finally, we think it would be interesting to talk about some areas we are very excited about right now. Clearly, we are touching on some consumer work that will be done by agents—you can call it long-running agents, autonomous agents, heavily used consumer. If you think from this perspective: consumer agents can do all the tasks you don’t want to do, giving you things you wouldn’t spend time or money on—this is very appealing, and I think its spread will be rapid.

Robotics is an area we have spent a lot of time and attention on, and we happen to believe it may even be bigger than LLMs—but it may arrive three to five years earlier. So over the next five years, this will be a huge area of investment and excitement.

Autonomous driving is here, and driverless technology is feasible. This is an incredibly exciting time. Looking at the automotive and transportation industry: it is one of the largest industries in the world, yet we may not talk about it much because we are spending so much time discussing AI. If we look at the mileage of Uber and Lyft, it only accounts for about 1% of the total mileage in the United States. I believe that with an autonomous driving network that is 14 times safer than human drivers, this number will at least increase by an order of magnitude in the coming years. Additionally, with 17 million new cars sold in the U.S. each year, I believe that in the next decade, they will all be autonomous.

David: AI multiplied by biology: this is an incredibly exciting field, and people in laboratories are clearly talking about it—new drug discovery, addressing some of the most debilitating diseases in the world. We all have great hopes for this promise, and we will see a lot of progress in the next decade.

David: Personal health is another area that I am very excited about— as a health maximalist, I have always felt that there is no good place to put all your information about yourself and get highly personalized advice.

David: Finally, entering enterprises and the diffusion beyond coding: this is what you just talked about, Sarah—the actual diffusion of technology in enterprises is still so early, and I think this will be super, super exciting.

David: Lastly, there is a field that is not in the AI bucket, which we call American Dynamism—the re-arming of the entire "American Dynamism" tech stack. We have been investing in this area for a long time and are major investors. It still accounts for less than 5% of total defense spending. Newer suppliers like Anduril, Saronic, and Castelion are expected to grow significantly with changing demand.

David: There are so many exciting fields. Clearly, we are very active in AI, and we are optimistic about its impact on the overall economy—building scale is huge, but we believe it will be a massive efficiency boost in the U.S. It has been a pleasure recording this show with you all. Thank you.