The Biggest Companies Always Win (Until They Don't)
There’s a story about Rockefeller that most people get backwards.
In 1911, the Supreme Court broke Standard Oil into 34 separate companies. Rockefeller owned about 25% of the whole thing. He woke up a monopolist and went to bed a shareholder in dozens of “competitors” — Jersey Standard (Exxon), Socony (Mobil), and so on.
The common telling is that this was a loss. The government broke his empire. Justice served. The little guy won.
That’s not what happened.
The Rockefeller Paradox
Within a few years, Rockefeller was significantly wealthier than before the breakup. The pieces were worth more than the whole because the individual companies had to modernize and compete in the gasoline market.
But the deeper thing — the part I keep coming back to — is that Rockefeller won because he owned the infrastructure. Not just oil. The distribution networks, the refineries, the institutional knowledge. When the energy industry shifted from kerosene to internal combustion, the technology changed, the fuel changed, the applications changed. But the person who controlled distribution kept winning.
Distribution Beats Product
I’ve been thinking a lot about the difference between creating value and capturing value. They’re not the same thing, and AI is making that gap wider.
Thiel said it in Zero to One: “poor distribution — not product — is the number one cause of failure.” I don’t think that’s ever been more true than right now.
The conventional wisdom of 2023 was that AI would be the great equalizer. Solo founder with an LLM replaces a 50-person engineering team. And in some ways, that’s real. Building has never been easier.
But building was never the hard part. Scaling is. And scaling requires distribution.
The Feature Absorption Problem
We watched this play out all through the 2010s. Snapchat invented Stories, Instagram copied it. Houseparty had group video, FaceTime absorbed it. Foursquare’s check-ins became a button on Facebook.
Same pattern every time. Startup proves product-market fit. Bigger company copies the feature, ships it to their existing user base, startup slowly dies.
Now it’s happening with AI, faster.
The startup that raised a Seed to build a “legal document analyzer” or an “automated slide deck generator” — most of them are already cooked. Microsoft didn’t need the best AI. They needed Copilot shipped to 345 million paid Microsoft 365 subscribers.
As of early 2026, over 70% of the Fortune 500 has adopted Copilot. That’s thousands of specialized startups absorbed into a dropdown menu. The startups had better products. Microsoft had better distribution. It wasn’t close.
The Coordination Thing Nobody Talks About
Coase won a Nobel Prize for a pretty simple idea: firms exist because it’s cheaper to coordinate activity under one roof than to negotiate everything on the open market.
Here’s why that matters now. AI has made coordination maybe 10x cheaper. The instinct is to think that means firms get smaller — everyone becomes a solo founder. But Coase’s logic actually says the opposite. When coordination gets cheaper, the optimal firm size can increase. If it becomes easier to manage 100,000 people, the best-managed companies will just get bigger.
And that’s exactly what’s happening. Despite all the “lean startup” hype, the Magnificent Seven and the Fortune 50 keep dominating. AI didn’t shrink them. It made them more efficient at being large.
Where the Money Actually Goes
The railroad companies that built the tracks mostly went bankrupt. The fortunes went to the people who used the tracks — Rockefeller’s oil, Carnegie’s steel. Infrastructure providers captured almost none of the value they created.
Same story with the internet. Netscape and the ISPs were the pioneers. Amazon and Google captured the value. The customer relationship was what mattered.
I think AI follows the same pattern. The massive value probably doesn’t stay with the model providers — not as inference costs keep falling toward commodity pricing. It gets captured by the companies that already have the customers, the data, and the distribution. The applied AI layer, not the foundation model layer.
Build the infrastructure, someone else captures the value. Own the customer relationship, and it almost doesn’t matter what infrastructure you’re using. The pattern has been consistent for about 150 years.
What This Means If You’re Investing
If this thesis holds — that AI benefits incumbents with distribution — the most boring companies become the most interesting.
Healthcare giants with proprietary patient data and regulatory moats. No AI startup recreates decades of medical records and physician relationships overnight. Banking institutions with massive deposit bases and compliance gravity — the regulatory burden that makes banking tedious is also a moat that AI can’t route around. Logistics companies with physical footprints. Amazon’s advantage isn’t the recommendation algorithm. It’s the warehouse 10 miles from your house.
The burden of proof has shifted entirely to the disruptors. When a startup says they’ll beat an incumbent, the question isn’t “is your AI better?” Everyone has access to world-class models. The question is: “What do you have that they can’t add to a dropdown menu next Tuesday?”
If the answer is “nothing but the AI,” that’s not a company. It’s a feature waiting to be absorbed.
Rockefeller understood that railroads were a tool, not the goal. The winners of the AI era will be the companies that see it the same way — a means to scale an advantage they already had.