The Pattern Nobody Wants to Name
I've been tracking infrastructure-to-application value transfer for longer than most of the founders currently raising at $100 million post-money valuations.
Here's what the data actually says, and why it should keep you up at night if you're building infrastructure.
PC era: infrastructure market caps peaked in 2000. Twenty-six years later, in nominal dollar terms, they haven't surpassed that peak. Not even close.
Web era: new infrastructure entrants produced $400 billion in market cap. Application companies? $3.1 trillion. That's 88% of all new value created.
Mobile era: infrastructure pulled in roughly $700 billion. Applications generated $3.7 trillion — Netflix, Spotify, Meta, Uber, Airbnb. The same pattern. Different decade.
Chi-Hua Chien doesn't need to predict the future to know where AI value will flow. He just needs to look at the last three technology cycles and apply Occam's razor.
Infrastructure gets commoditized. Applications capture the value. Always has. Always will, until someone can prove otherwise with data that doesn't come from a pitch deck.
And right now, we're watching the commoditization accelerate in real time. Google just announced their subscription AI product is dropping from $7.99 a month to $4.99 while doubling storage. That's not competition. That's a structural advantage being deployed by someone with vertical integration and distribution that no startup can replicate.
When Google starts price-warring for the average consumer, you don't need a crystal ball. You need a calculator.
The Trust Gap That Kills Super Apps
Facebook tried everything.
Facebook Credits launched in 2009. Facebook Pay came later. Then there was Libra, which somehow managed to attract regulatory attention before it even launched.
None of it worked. Not because the technology was wrong. Not because the user experience was bad.
Because Americans don't trust their social feed with their money. And honestly? They shouldn't.
Chien puts it differently than most VCs would. He doesn't call it a "trust issue." He calls it a fundamental mismatch in how people allocate attention and risk.
Social entertainment has high time, low monetization per interaction. You scroll for hours. You tap a few things. The business model runs on attention.
Financial services is the complete inverse. High monetization, low time. You open your banking app, you transact, you close it. But you demand extreme confidence in security and reliability.
Bridging that psychological gap isn't a product problem. It's a cultural one. And no amount of engineering can fix what isn't broken in the code.
This matters because every AI company currently trying to be your assistant, your banker, and your social network is walking into the same wall Facebook bled against for fifteen years. The difference is they don't have Chien's track record of watching these patterns play out before they become obvious.
He found The Facebook when it was six people at Harvard. He knows how these things start. He also knows how they fail when they try to be everything.
Personalization Is the Real Moat
Everyone's talking about personalization like it's a feature. It's not a feature. It's the entire business.
Chien's portfolio tells the story better than any thesis document could. Triumph, Ritten, Flow GPT — these are entertainment companies doing $100 million to $600 million in ARR very quickly. Great margins. Fast growth.
Here's what's interesting: customers don't describe these as AI applications. They describe them as entertainment applications. The AI is invisible. It's just making the experience more customizable, more personal, better over time through feedback loops that compound.
That's the pattern. The technology disappears into the experience. What remains is something that knows you better than you know yourself and adjusts in real time.
Midi Health plays the same game in a completely different category. Women's health, specifically perimenopause and hormone replacement therapy, has a provider shortage that no amount of medical school expansion can fix quickly. There simply aren't enough clinicians trained in this space.
Midi uses AI to scale expertise, not replace it. An AI assistant helps a single clinician triage hundreds of patients daily. It flags symptoms. Pulls lab results. Reminds the doctor when a patient hasn't had a bone density scan in eighteen months.
The result? They're treating hundreds of thousands of patients who otherwise couldn't be reached. Cost-effectively. In a market that was supply-constrained for decades.
This is what happens when you apply personalization at scale to a category where human expertise is the bottleneck. You don't get a chatbot. You get a system that makes rare expertise accessible to people who need it.
And that's worth $400 million in ARR because people will pay for access to care that was previously unavailable.
The Local AI Inflection Point
Two years ago, the gap between what you could run locally on your phone and what frontier models in the cloud could do was eighteen to twenty-four months.
Today, that gap is six months.
Next year? Chien expects it to be three months. Maybe less.
This isn't a prediction. It's an extrapolation of a trend that's been accelerating for five years. The hardware is getting cheaper. The models are getting smaller without getting dumber. The latency is dropping.
What does this actually mean for consumers? Your phone can now do what GPT-4 could do last year. And it does it without asking for your data. Without logging you in. Without selling your attention to the highest bidder.
That's personal. That's private. That's yours.
But here's what most people miss: the technology is only half the story. The use cases aren't well-defined yet. When the iPhone launched in 2007, everyone thought it was going to be web applications ported to mobile. It took years for entrepreneurs to figure out what was actually possible.
We're in that same moment now. The capability exists. The distribution exists. What doesn't exist yet are the applications that feel inevitable in hindsight.
Chien's point is subtle but important: AI makes two things possible at scale. First, it lets you process large amounts of context and make sense of it. Second, it enables personalization down to the individual, cost-effectively, with feedback loops that improve over time.
Those aren't features. They're foundations. And the companies that build on them will look obvious in five years. They just don't look obvious today.
The Counter-Reaction Nobody's Pricing In
Here's what happens when you saturate a population with digital content: they start craving the one thing that's in short supply.
Real human contact. Real-world experiences. Physical presence.
Chien's portfolio bets on this counter-reaction more aggressively than most VCs would admit. Not because it's trendy, but because the data supports it.
Fever started in London and Madrid with candlelight concerts. Quirky events. Small scale. Now they're selling tickets to the Bridgerton Experience and operating at a scale that makes them essentially the Live Nation of Europe. They're not selling entertainment. They're selling belonging.
Bump, based in Paris from the original founders of Zenly (acquired by Snap), built an interface that lets people interact in the physical world, catalyzed by digital information. You don't use it to date. You don't use it to hook up. You use it to find someone to walk with.
And when you do walk? The app gets out of the way. You don't talk. You just walk. And that's enough.
This is the pattern Chien sees playing out across his entire portfolio. AI as enabling technology — knowing where you go, who you hang out with, where you spend time — can extrapolate relevant interests that make real-world experiences more useful and more personal.
We've watched the world go digital for fifteen years. Now we're watching it swing back. Not to the screen. To the sidewalk. To the coffee shop. To the quiet moments that don't need an algorithm to validate them.
The companies building for that reality? They're not selling AI. They're selling presence. And presence is the one thing no model can generate.
What This Means for How We Actually Invest
I've spent enough years in this business to know that the companies that win aren't the ones with the best technology. They're the ones with the best understanding of human behavior.
Chien thinks like a cultural anthropologist, not a technologist. That's not a compliment he's earned through branding. It's earned through twenty years of watching consumer trends play out before they become consensus.
The pattern is clear: infrastructure commoditizes. Applications capture value. Personalization creates moats. Trust gaps define categories. And the counter-reaction to digital saturation is always physical connection.
None of this is new. It's just been happening in slower motion than the current cycle allows people to appreciate.
What's different now is the speed. The commoditization is accelerating. Google dropping prices signals that infrastructure players are already competing on cost, not capability. The local AI gap shrinking means the distribution advantage of cloud models is evaporating faster than anyone expected.
The companies that will win are the ones building applications that feel personal, not pervasive. That solve real constraints — provider shortages, trust gaps, attention fragmentation — instead of creating new ones.
That's what Goodwater is betting on. Not AI. Not models. Not chips.
Human applications. Built with AI as the engine, not the message.
And if history is any indicator — and it always is — that's where the money actually lives.