I was part of a team building an internet marketplace. The vision was to aggregate a fragmented, underserved market that didn't have a centralized place for buyers and sellers. As we built the platform and approached launch, we realized the model itself was structurally undermined — not by a competitor, but by a shift in the conditions that made marketplace defensible in the first place. This essay is a formal articulation of that thesis.
Money out of nowhere
In 2019, Bill Gurley (Benchmark partner, early Uber investor) published Money Out of Nowhere: How Internet Marketplaces Unlock Economic Wealth. He argued that internet marketplaces don't just move money around — they create it. By connecting buyers and sellers who couldn't find each other, they unlock economic value that didn't exist before.
And he was right. Many of the success stories over the past two decades launched with laughably small TAMs, then grew far beyond them. Today, they look like dominant businesses, operating in healthy markets with strong defensibility.
But that durability rests on a single assumption: that the marketplace model remains the most efficient way of connecting buyers and sellers.
As AI advances, this assumption breaks. The need for marketplaces that don't own supply will no longer exist. There will be a more efficient model — and the money created out of nowhere will flow back to the sellers.
Not all network effects are equal
The long term risk profile of a marketplace can be seen through its network effects.
The most defensible marketplaces are "heavy marketplaces" with operational network effects. Competitive advantage compounds at the unit economics level. As Uber grows, rides are not just easier to find; they're quicker and cheaper. These companies are better understood as operational platforms with marketplace functionality, rather than pure-play marketplaces.
Most consumer-facing marketplaces, however, are "light marketplaces" with informational network effects. Competitive advantage compounds at the customer experience level. As Upwork grows, freelancers that match your needs are easier to find. But the hourly rate or the quality of the service itself does not change.
Light marketplaces are not pure aggregators, but their moat is being an aggregated point of demand. Heavy marketplaces can deepen their moat indefinitely by investing in operational infrastructure. Light marketplaces have no equivalent path, and therefore are at high risk of disruption.
Category | Heavy marketplace | Light marketplace |
|---|---|---|
Network effect | Operational | Informational |
Disruption risk | Low | High |
Advantage | Unit economics | Customer experience |
Moat | Physical logistics | Demand aggregation |
Examples | Amazon, Uber, Doordash, Opentable, etc. | Booking.com, Expedia, Upwork, Taskrabbit, Fiverr, Thumbtack, Airbnb, Etsy, Vrbo, Classpass, etc. |
Discovery and booking are co-dependent
The core function of a light marketplace is to control both discovery and booking. Fundamentally, this means:
Make it easy for buyers and sellers to find each other. Aggregate supply, standardize offerings, structure search, and capture demand.
Make it easy for buyers and sellers to connect with each other. Enable seamless booking, payments, while creating trust and taking on relevant liability.
The two layers are co-dependent. This bundle is the value proposition.
Strip away discovery, and the marketplace loses its pricing power. A 10–20% take rate is only justifiable if they are an irreplaceable source of customer generation.
Strip away booking, and the marketplace exits the payment flow entirely. There is no mechanism to collect its take rate.
If the two layers get de-coupled, the business is structurally broken.
Sellers go where revenue opportunity is highest
Sellers are not loyal to any platform (nor should they be). They are individual businesses looking to maximize bookings. When free listing is the standard, supply is fluid, and will go wherever demand is.
Airbnb launched in 2008. Because supply was essentially non-existent, they had to go out and create a market. A decade later, Booking.com entered the same market. Today, both companies have the same number of listings (roughly 8 million globally). There is likely significant overlap, and that is the point.
When competing marketplaces have similar supply, they are competing for who builds the better brand and execute on demand capture.
Every distribution network becomes a marketplace
The discovery layer is being de-coupled by LLMs. We have seen ChatGPT integrating with Etsy, Booking.com, and others. But the bigger story is: LLMs are now the source of demand, and they represent a step change in discovery.
By step change, I mean that an LLM can handle nuance across verticals, standardize any data, pull from multiple sources, and investigate multiple avenues. To a model, there is no meaningful difference between finding a gardener, yoga instructor, hotel, or weekend cabin. Users are conversing with LLMs rather than searching. As this becomes the primary interface for discovery, the value of aggregation through vertical-specific marketplaces diminish.
Every source of demand will incorporate LLMs more deeply. Obviously, frontier models are the ultimate network. But social media platforms are also clear winners because of their unique and longitudinal trust graph. Both will now compete directly with light marketplaces, rather than being adjacent threats. As sellers continue optimizing for multiple online channels, the information asymmetry will disappear. It took a couple. What took Booking.com years to match Airbnb’s supply will take frontier models and social platforms a fraction of that time.
The new paradigm
The atomic unit of the old paradigm was a marketplace listing — the seller’s Vrbo profile, Fiverr gig, or Thumbtack page. The atomic unit of the new paradigm is an agentic business profile: a portable, machine-readable record of services, calendar, pricing, credentials, and preferences. Any buyer-side AI agent can discover it, check availability, and complete the transaction directly. No platform intermediary required.
The infrastructure already exists. Stripe has launched its Agentic Commerce Protocol. Google and Shopify have announced a Universal Commerce Protocol. These moves signal that payment and coordination are core, not the marketplace and discovery layer built on top of them.
In this context, the sellers who were once platform tenants become the infrastructure owners. I discuss with more detail in Unbundling the Marketplace Stack. But the bottom line is simple. There is a clear path to a world where sellers are in a position of strength. Marketplaces lose their leverage, and sellers are able to maximize bookings while keeping more of their revenue.
Money into the hands of sellers
For two decades, internet marketplaces were the most efficient models for connecting buyers and sellers.
But increasingly, they will become the inefficiency provider. The high take rate becomes unjustifiable, and represent an arbitrage to attack.
As LLMs dismantle the information asymmetry that marketplaces had, sellers will increasingly look to these platforms as a source of customer generation. There will be digital infrastructure that gets created to solve nuances in the booking layer.
We will see an unbundling of the marketplace business model. The money out of nowhere becomes money into sellers' hands -- at higher margins, through infrastructure and relationships they control.