The last time @GavinSBaker and I sat down to record a podcast, it was almost exactly a year ago and we started with the immediate question: is AI a bubble? Gavin's answer focused on utilization and returns (and perhaps unsurprisingly, his answer was: No). Unlike the dark fiber of 2000, there were, and still are, no idle GPUs. The largest buyers of compute were funding the buildout from some of the strongest balance sheets in the world, and their AI investment was already producing real returns. I sat down with Gavin again last week, and to chart just exactly where we are in the cycle. He has spent the summer asking operators for one quantitative measure in their business that is getting worse and has yet to find one. At the same time, public AI stocks have gone through meaningful drawdowns. Still, if you look at the way people are using AI today, there’s a good chance that demand diffusion has barely begun. AI revenue rests on fewer than 10m heavy users, against roughly 1.5b knowledge workers. At the most AI-native startups, token spend is approaching or exceeding 10% of human compensation, which suggests that there’s a long way to go before even the earliest adopters fully integrate agentic capabilities. Within Atreides, Gavin said token consumption rose 100x from March through August, and even further once the team started using products like GrokBot. As Gavin put it, once people begin approving automations, token consumption starts to feel “sort of endless.” This dynamic would be reason enough alone to believe that we’re massively undersupplied at the moment. But there are other reasons on the supply side too, namely that there are a near-unbounded number of potential winners in the space: - Frontier labs can keep winning because on the highest-value tasks, marginal improvements in intelligence are worth far more than marginal differences in price. - Open models illustrate that most work doesn’t require frontier performance, creating a much larger market for intelligence that is cheaper and customizable. - Nvidia, hyperscalers, neoclouds, and inference providers can simultaneously win because every additional token still requires physical compute, even if models become more efficient. - Enterprises can win as proprietary data, workflows, and institutional knowledge become more valuable. - Application companies can win by capturing services budgets and owning the customer relationship. This doesn’t suggest everyone will be successful, but it does mean the market itself is positive-sum and we will look back on zero-sum thinking as far too limiting. Better models create better products, better products create more users, more users create more token demand, and more demand supports continued investment in models and infrastructure. Check out the whole conversation below: @a16z

Kevin Mandia says "There's no way to make the next year pretty." He thinks the first time a model revealed itself capable of finding thousands of vulnerabilities, is when everyone finally accepted the game has changed and AI on offense is coming. One of our most sophisticated portcos recently told us they put a significant fraction of their engineers' projects on hold and enlisted them to stress-test their own walls. Kevin's new company @ArmadinSecurity is scaling up to offer this offense-as-defense to everyone, and ultimately build out the autonomous security he believes every company will need in the AI-enabled future.

When people ask whether "this" or "that" will be successful in AI, the answer is very often both. OpenAI and Anthropic together represent the fastest, largest success in the history of business. They + SpaceXAI + Cursor add more revenue per month than any hyperscaler other than Amazon. Up until now, headlines like these have been driven effectively off coding applications alone. We are very early. Thank you @jaltma for having me on!

a16z's David George, Sarah Wang, Alex Immerman, and Santiago Rodriguez on the State of Markets, from AI and infrastructure to the next investment cycles: 1. Tech is the everything cycle. High-tech equipment, software, and R&D are now about 55% of US capital spending, and the AI buildout just passed the railroads as a share of GDP. The five largest hyperscalers will spend about $780B on capex this year, heading past $1T in 2027, and demand still outstrips supply at every point in the chain. 2. It's an earnings story. Stocks are up about 20% while multiples are down about 20%, and the S&P trades under 20x earnings. The market is up 90% since ChatGPT, on 15% earnings growth. 3. AI adoption is broad and shallow. 69% of the S&P 500 has a live AI deployment, 30% report a quantified result, and just 2% have AI doing a job they'd notice if it stopped. The power users are pulling away: the top 1% of AI spenders spend eight times the top 10%. 4. The private markets are enormous. Six private companies, Anthropic, OpenAI, Databricks, Stripe, Waymo, and Revolut, add up to about $2.4T by last round, more than every IPO of the past decade combined. "The founder is the asset class. The IPO is another financing event." Plus: why a data center can lower your electricity bill, what Amazon saying no to Muse and Instacart saying yes tells you about marketplaces, and the seven areas the team is most excited about next. 0:59 Charts from State of Markets II 2:59 Tech is the everything cycle 4:56 Is it a bubble? 6:31 Hyperscaler CapEx: $1T next year 12:12 CapEx is someone else's order book 14:36 Data centers can lower your bills 17:06 OpenAI and Anthropic vs. some of the best software companies ever 18:22 Enterprise AI: 69% adopted, 2% measured 20:32 The top 1% spend 8x the top 10% on AI 24:36 Spending on AI for growth or for savings? 28:46 Only 2% of households pay for AI 31:36 Amazon blocks Muse, Instacart welcomes 36:16 Software traded growth for profit 42:56 6 startups vs. a decade of IPOs 45:46 Why employees are turning down liquidity 48:44 Seven areas they're excited about YouTube: piped.video/lr3hNhA0IfQ @DavidGeorge83 @aleximm @santiago__rdz @sarahdingwang

Introducing our State of Markets pt 2, along with a companion podcast where we unpack the data and discuss what comes next. Tech is the everything cycle. Supply: putting the buildout in context, just passed railroads as % of GDP. The wisdom of Elon is real: the factory (or the datacenter!) is the product. Demand: diffusion is so, so early. Median AI vendor spending in the top 1% of companies is 8x that of the top 10%. Only about 30% of S&P 500 companies report a quantified AI impact, which means there’s a substantial opportunity in connecting models to a company's data and workflows. Diffusion into companies is one of the main themes of the next 5 years. We’re entering the agent work period. Only a few million users today, but applicable to billions of internet users with massive surplus created. META/GOOG monetize US users at $200+ per year today. Agent opportunity is much higher. Mega-trends the next 5 years: Consumer agents, Robotics, Autonomy, AI x bio, Personal health, Diffusion into enterprise, New era of American Dynamism. Much more in our SoM report here - a16z.news/p/state-of-markets… @a16z @sarahdingwang @aleximm @santiago__rdz

🇺🇸🇺🇸

Big and public used to be the same milestone. Tech IPOs at the 75th percentile enter around $3.5 billion. The ten companies here cleared it by 12-500x in private rounds. Great piece from @jkhamehl on the trillions accruing in private markets and the venture fund allocations that are still sized for the old world.

Flock is making cities safer. Safety doesn’t have to be a privilege. @flocksafety

Our fifth growth fund is now $8.5b. The founder is the asset in late stage, and we back the best in the world. At the same time, the best founders face global complexity earlier in their lives than ever before. So we are building out an enhanced Growth Platform for these exceptional founders. Biggest product cycles of our lives. Never been higher stakes, and never been more excited. @a16z

New $1.1b fund for physical AI. We are witnessing a reinvention of compute infrastructure at historically unprecedented magnitude and breadth. It’s time to build the Machine Age. @a16z

I had a great time talking with @GrantLaFontaine, who co-founded @Whatnot with @loganhead13 in 2019. What began in collectibles is now the largest live-commerce platform in North America and Europe. Whatnot reported more than $8B in 2025 GMV and says H1 2026 exceeded that total. @a16z partnered with Whatnot at the Series A. In this chat we reached all the way back to the company’s origin story (beginning with Grant selling Pokemon cards as a kid) and the larger market opportunities that lie ahead. Some takeaways: -Traditional e-commerce owes much of its efficiency to obscuring the merchant. On Whatnot, the seller has a face, a brand, and customers who return for them. Per Grant, it can be an advantage to lack experience in an established category: instead of copying what exists, you focus on the experience users want. When Whatnot was getting started, lots of teams were trying to emulate the live shopping marketplaces growing popular in China. Instead of doing this, Grant and Logan just tried to make the best place to sell collectibles. -Whatnot feels more like entering a store or mall than querying a product catalog. The buyer may know the kind of thing she wants, but the exact purchase emerges from browsing and entertainment. Live formats make expertise valuable. A seafood distributor can show what came off the dock, explain why it is in season, answer questions, and reach buyers nationwide. -Today, people spend ~95 minutes a day on Whatnot, and on a given day more than 80% of them don't buy anything. "It's because it's fun. If you simplify Whatnot down, the buyer value prop is a fun experience to shop, and there's not a whole lot of fun experiences shopping online." -When it comes to using AI, the company wants to empower their sellers. Whatnot uses LLMs to infer listing metadata and provide business insights while preserving the seller relationship that brings customers back. A fun conversation about marketplaces creating new demand, new sellers, and entirely new businesses.

Congratulations to @OpenRouter and @stripe! This is a massive win-win. Here is @gaybrick talking about the movement between tokens and dollars. Also, Late Stage Venture is about Late Stage Founders 🤝 "Deft Helmsmanship" @a16z

Stripe’s strategy is to “win all the startups and then win them again.” I had a great time talking with @gaybrick. Will has held a broad seat at @stripe for more than a decade, joining as CFO in 2015, and today leading Technology and Business. This makes him the right person to explain how Stripe went from a simple payments API to a platform with ~25-30 banner products and 288 launches at its most recent Sessions event. Here’s what their strategy looks like: -Companies like DoorDash and Instacart started on Stripe when they were tiny and pulled the company upmarket as they expanded. -The fastest startups act as canaries for new product opportunities. Cursor surfaced a burgeoning problem of AI free-trial and token abuse; so Stripe built a token detection pipeline with the team in a weekend, and the resulting network now helps companies on Stripe’s platform block thousands of fraudulent trials daily. -Stripe keeps expanding the number of needs it serves inside each company, from Billing, Radar, and Tax to Managed Payments, Treasury, and stablecoin infrastructure. There is a large and timely lesson from Stripe which I strongly agree with: using AI solely to reduce cost is essentially going short your own company. Instead, companies should be using AI (and the tailwinds it creates) to ship more, serve users better, and create their next product line. And as Stripe continues to invest in agentic commerce, the company can move from processing a transaction to orchestrating the entire economic loop around it. The line between tokens and dollars is already blurring, and Stripe’s next mandate is to make moving between them as seamless and safe as moving between dollars and euros. A fun conversation about what it takes to build the next Stripe inside Stripe. @a16z

Welcome to the SpaceXAI team! Cursor will help us move faster on the path to the world's most useful AI, starting with software engineering and expanding into knowledge work.