The Valuation Paradox of Compounding Growth
Here's the thing about venture capital in 2026 that keeps every LP up at night: we've never seen a growth curve like this before, and our pricing models are built for a world that no longer exists.
Chang Xu put it bluntly on stage at TechCrunch's StrictlyVC Los Angeles last week. ChatGPT went from one dollar to forty billion in revenue in six months. His portfolio company OpenArt scaled from a million to seventy million ARR in two years — cash-flow positive most of that time, with twenty people. Twenty. Not two hundred. Not two thousand.
When you have compounding accelerant growth like that, the math stops making sense by any historical standard. Are you watching the rising costs of AI inference? Xu's point wasn't that valuations are fair — it's that they're paradoxical. Price every seed deal to that terminal value and your portfolio implodes. Don't price for it and you miss the ones that actually compound.
Carter Reum laughed at the whole framing. We've seen this before, he said — cloud, the iPhone, even the automobile back in the 1920s when people genuinely feared for their livelihoods. The dynamic is the same, just steeper. I think he's partly right and partly wrong. Yes, every transformative wave creates valuation confusion. But the velocity of that confusion is what makes this cycle different.
Reum's own firm, M13, manages $2.5 billion in assets and has backed seventeen unicorns as a seed or Series A investor. He does the cocktail napkin math on every deal. Recently, they looked at an AI software business for brands and walked away because the numbers didn't check out. How big were the winners in previous cycles? Are there more brands in the world? Will they pay double or triple for software now? The answers didn't support the investment. That's discipline, not fear.
The lesson here isn't that you should price everything to the moon scenario. It's that your mental model needs two tracks running simultaneously: one for the outliers that will compound beyond anything you've modeled, and another for the portfolio where most things fail. If you only run one track, you're either overpaying or underinvesting — and in a market this fast, both mistakes are fatal.
Incumbents vs. Startups and the Agents Paradigm
This is where Reum's observation cuts deepest, and honestly, it should terrify every founder reading this.
In previous technology cycles — cloud, mobile, the internet — innovators competed with other innovators. Zuckerberg versus Evan Spiegel. Travis Kalanick versus John Zimmer. It was a fight between peers, and the better execution usually won.
In this cycle, you're competing with innovators AND the largest, most well-funded organizations the planet has ever seen. And Reum argued something uncomfortable: for the first time in history, the incumbents actually hold the advantage. The technology. The capital. The data. The talent. All of it concentrates in places like OpenAI, Anthropic, Google.
So how do you build something that doesn't get steamrolled? Xu laid out a framework that's become essential thinking for anyone raising money in AI right now: invest below the AI and above the AI.
Below the AI, you're reinventing infrastructure that was built for humans but is now being used by agents. Last year, nobody thought you'd need a new GitHub. This year, Xu can count on two hands how many strong teams are building the version control system for AI agents. Databases, deployment tools, orchestration layers — everything built for human developers needs a fundamental rethink when your primary user is an autonomous system.
Above the AI, when the infrastructure layer gets crowded, you go back to what's defensible. What has long-term differentiation? What can't be replicated by a team of engineers at a hyperscaler working off the same open-source models?
The agents paradigm changes everything about how you think about moats. When your customer is an AI that can evaluate alternatives in milliseconds, the traditional advantages of brand loyalty and switching costs start to dissolve. You need technical defensibility that's real, not perceived.
Reum's advice to founders captures this tension perfectly: you need a microscope in one eye and a telescope in the other. Execute on what's in front of you this week, but keep scanning the horizon because the board is changing constantly. Be a domino player and a chess player simultaneously, even when the market is debated as AI psychosis.
Defensibility in the Friction and Depth Markets
If you're going to build something that survives contact with hyperscalers, you need a moat. And not just any moat — one that incumbents can't simply bulldoze through with more compute and more engineers.
Reum's thesis on this is clean: friction as a moat. Regulated industries are where startups thrive because regulation moves slower than technology. He cited M13's near-billion-dollar exit in a company disrupting 911 call centers with AI. The hyperscalers might eventually go there, but as a few-billion-dollar outcome? They're not going to prioritize that anytime soon.
Healthcare works the same way. OpenAI isn't going to build a FDA-approved diagnostic tool next quarter. The regulatory pathway creates time — and time is everything for a startup trying to establish market position.
Xu added another dimension with his depth-versus-velocity framework. In velocity markets, fast followers are faster than ever. Execution speed is the only moat that matters. In depth markets, hard things remain hard. He gave a perfect example: M13 has a portfolio company using transgenic chickens as an alternative to manufacturing complex proteins. It's cheaper than traditional bioreactors, apparently. But chickens still take the same amount of time to hatch. Biology doesn't care about your funding round.
These depth markets are where you find genuine, structural defensibility. A fast follower can copy your code in a week. They can't copy the regulatory approvals, the clinical trial data, or the biological systems you've built into your product.
Xu also noted that some of the most interesting opportunities look like bad ideas at first glance. OpenArt started as a discovery page for generative image prompts — nobody could tell from the outside how that became a seventy-million-dollar business. The depth in that market was invisible until you were inside it.
The pattern repeats: four or five years ago, investing in anything selling to Hollywood was considered a bad idea. Then creative AI, generative images, video, and world models proved everyone wrong. Cursor was dismissed as an AI wrapper before hitting a sixty-billion-dollar exit. The consensus categories — agents for finance, agents for healthcare — will produce winners, but the real alpha is in the ideas that make you pause and think, "Huh, I don't know if that can even be a business."
Reum's rock-skipping analogy captures the cycle dynamics perfectly: the first wave is always the most obvious and crowded. Some startups are building proprietary models to compete. The second and third ripples are where it gets interesting. Those later bets are harder to get right, but fewer people are thinking about them, valuations are more reasonable, and the returns tend to be much better.
The Southern California Ripple Effect: SpaceX and Taste-Driven AI
There's a liquidity event coming that will reshape the Los Angeles venture ecosystem in ways most people aren't preparing for.
SpaceX is going public. And unlike Anthropic or OpenAI, where the liquidity flows to VCs and institutional investors, SpaceX's IPO distributes wealth widely — to employees, to early contributors, to people who actually live in Los Angeles. Reum was explicit about this distinction: never has this much money come back and been so broadly spread across a single geographic region.
Every major liquidity event generates a second wave. The previous LA cycle produced Riot Games, Tinder, Snap. This one will be a different order of magnitude.
But here's where the conversation gets genuinely interesting. Three years ago, everyone declared San Francisco dead. It turned out to be a little less dead than people expected. Reum thinks the same applies to anyone writing off Los Angeles.
The first wave of any technological cycle is technical. The talent concentrates in San Francisco, Seattle, Boston — places with deep engineering cultures and established university pipelines. But what comes after the technical wave? New business models. Creative thinking. Cultural understanding.
Xu put it most memorably: the next frontier in AI isn't more compute. It's taste. Making films that resonate emotionally. Creating content that connects with specific cultures. Building products that understand nuance in a way that raw engineering can't replicate.
"San Francisco has extraordinary technical talent," Xu said. "And that's also exactly what the models are getting very good at automating and accelerating."
Los Angeles has taste in spades. The city's advantage isn't in building the next transformer architecture — it's in understanding what resonates with billions of people across different cultures, languages, and creative traditions. The companies that will define the next phase of AI aren't just technical marvels; they're cultural artifacts.
This doesn't mean San Francisco is finished. It means the center of gravity shifts over time, and the geographic distribution of venture capital follows that shift. The SpaceX liquidity event is the catalyst that makes Los Angeles a serious contender for the next decade of venture activity.
For founders considering where to build, the implication is clear: if your product depends on cultural fluency, creative insight, or emotional resonance, Los Angeles isn't just a viable location — it's the optimal one. The technical wave built the foundation. The taste wave builds the empire.