All Networks Are Not Created Equal: A new playbook for the AI era

Network effects underpin a disproportionately large number of mega-winners in technology. Google. Meta. Amazon. Uber. Flipkart. Swiggy. The list is long. Across all such platforms, the network creates a better experience for n+1 n+1th member directly as a consequence of adding the extra member, while simultaneously making the network more valuable for existing n members. The math behind networks is so strong that it creates an almost impenetrable level of defensibility when implemented well at scale.

If I were to categorise the different types of business moats, I'd place simple distribution and capital in the economy class, tangible data advantage in premium economy, trust and brand in business class, and network effects in the first class. Maybe the President's class, if there is one.

The rapid growth of the internet naturally lent itself as the enabler for creating some of the world's most defensible network effects businesses over the last two decades. However, the applicability of this powerful phenomenon to AI-native businesses is less apparent, at least till now.

Yes, there's the relatively well-understood concept of more users on ChatGPT (or pick your favourite AI assistant) → more feedback on the model → more complete RL loops → better model → more users. But, (A) it's an incomplete view, as the more significant piece of underlying model improvements continues to come from bespoke, non-user data collection & techniques (explains the unprecedented success of Scale AI!), and (B) there's a net new class of AI native businesses that can have inherent network effects advantages but haven't entered the zeitgeist just yet.

As the AI wave accelerates, I see many attempts at trying to "upend" incumbent networks without enough appreciation for why those incumbents became dominant in the first place.

We generally underestimate the rarity of genuine network-effect opportunities.

And overestimate how far AI alone can take us.

A relatively small number of AI-native businesses will exhibit sustainable network effects, but those that do will likely produce a generation of iconic companies.

As a student of venture history, I keep returning to one essay that has stuck with me for years: Bill Gurley's 2012 piece, "All Markets Are Not Created Equal." It's one of the clearest & most comprehensive frameworks for evaluating marketplaces, and one that has stood the test of time. Thinking of marketplaces as a sub-class of networks (e.g., Amazon is a two-sided network (of suppliers & buyers) & therefore a marketplace, while WhatsApp is a network but not necessarily a marketplace), several principles from the essay apply pretty directly to thinking about building new networks.

Today, the question is:

What is the equivalent framework for AI-native networks?

Where can AI create legitimate, compounding network advantages - not just short-term productivity gains?

Below are five conditions that can indicate whether an AI-native network becomes a category-defining business.

‍

1. The Network Makes Sense From First Principles

Before introducing AI, the underlying network must already create net new value for all members. Great networks with sustaining value will have most, if not all, of the following characteristics:

• Large total addressable market, or creation of a large market
• Untapped supply, or growth avenues for existing supply
• New forms of demand, or a superior experience for existing demand
• High liquidity and fragmentation on both sides
• Clear incentives for frequent, repeat use
• Low friction in signing up new members, or an early mover advantage in signing up high-friction members
• Transaction completion on the platform, or lock-in of the payment flow

Much of the above is now well understood in the tech industry, thanks to the remarkable success of the internet networks of our time. Also, Bill's essay remains a highly relevant read for anyone interested.

‍

2. AI Creates or Captures Data That Was Previously Impossible/Expensive to Access

A big unlock LLMs bring is the ability to capture and work with unstructured, nuanced, & high-signal data.

In some markets, it is irrelevant. Flight bookings are a good example: the data is already highly structured, uniform, and available to everyone. AI can make search friendlier, but it doesn't really benefit from differentiated data.

But in others, AI radically expands the information surface area.

Take hiring networks, for example. The future LinkedIn won't look like a grid of job titles and endorsements. It will look like a living model of a person's skills, trajectory, learning velocity, preferences, and work artefacts. Even a 5-minute phone call (with a well-designed voice AI) reveals so much more about an employee's credentials and interests than a static resume built over years.

Secondly, AI can often capture context that forms and checkboxes simply can't. Take matrimonial matchmaking, for example. Practically no user is going to respond No if directly prompted: “Do you believe in gender equality?”. An intelligent network will infer such subtler aspects through indirect conversational intelligence. Crucial data for matchmaking, which isn't practically possible to gather without AI.

‍

3. Matching/Curation Leverages Reasoning and Explainability

Discovery, matchmaking, or curation on most traditional networks relies on filters, rules, and manual discovery.

AI-native networks have the potential to leverage model reasoning and contextual intelligence for not just faster but better curation. This difference is subtle but profound.

For example, Boardy AI doesn't provide investors with a lengthy list of startups categorised by tags. Such information is readily available on web portals anyway. Boardy's voice AI instead tries to understand an investor's thesis and patterns of founding teams/markets/products they are most interested in to surface only the most relevant startups.

Secondly, a well-functioning AI-native network should also respect a key truth:

Users should understand why they're seeing what they're seeing.

If explanations are opaque, trust erodes.

If the system is a black box, liquidity stalls.

The most valuable AI networks will sit at the intersection of:

Complex reasoning + transparent decisioning + user trust.

And they will eventually own not just matching, but reviews, insight graphs, and the whole context layer across the network.

‍

4. Automating grunt work in communicating with the network unlocks faster adoption

Network effects accelerate when the platform performs tasks that users dislike.

Logistics teams currently call multiple suppliers, compare quotes, negotiate manually, and search for capacity. Companies like HappyRobot have already started showing how AI enables logistics operators to manage supplier communications at a far larger scale and with greater efficiency than was ever possible manually.

In the consulting world, analysts spend 30 to 70% of their bandwidth on collecting primary research data from subject matter experts and relevant stakeholders. Bridgetown Research lets consulting teams run dozens of concurrent expert calls via automated AI voice. It frees up consultant bandwidth to uncover new insights and go deep manually with only the most relevant experts on the most important subtopics. At the same time, experts benefit because they can now share their insights outside work hours at their own convenience.

Finding such automation-led wedges is catalytic for network growth, eventually leading to the benefits of network effects at scale.

‍

5. The network continuously learns user preferences

Preference learning can create long-term moats for AI-native networks.

Most markets where transactions are primarily about "getting the job done” are already taken and hard to disrupt with AI. Take Uber or Rapido, for example. Finding the best & cheapest commute option at a given time has more to do with supply-side liquidity and efficient algorithms than subjective preferences of users.

However, suppose there's a market that awards a premium for understanding the granular needs, tastes, constraints, risk profile, and decision patterns of the demand side. In that case, the opportunity for an AI-native disruptor becomes clearer.

Preference mapping is the kind of defensibility that compounds quietly and relentlessly.

Over time, the system knows:

• Which suppliers can you trust
• Which employee traits set them up for success at your org
• What type of insights are your consulting clients looking for
• Which logistics routes & price points have worked for you
• Which type of matches actually end up in second dates
• Which product recommendations do you consistently like

And so on.

Such signals can't be easily exported or recreated, as they often sit as vectorised representations within the network's system. They emerge only from long-term engagement and are usable only within the respective network's environment.

‍

Some Markets Naturally Lend Themselves to AI-Native Network Effects

Once you apply the above five filters, the map becomes clear.

Some markets provide a strong fit for AI-driven networks:

• Hiring: high signal, high variance, high complexity
• Dating: deep preferences, reasoning-based matching
• Market research: latent knowledge, asynchronous scalability
• Logistics & B2B procurement: massive coordination overhead, bespoke demand side needs, fragmented supply side offerings

In these markets, AI not only improves efficiency but also reshapes liquidity, trust, matching, and data ownership. It’s definitely possible to create large networks even if only 5/5 conditions aren’t met, but I believe they will still at least partially meet most conditions.

At the same time, many other markets will likely never generate strong network effects for AI native upstarts.

If the underlying data is largely structured, the matching rules are relatively straightforward, automation drives only marginal efficiency, or the switching costs are low, the network will stall before reaching escape velocity.

The opportunity lies in seeing the difference early.

‍

Closing Thought

AI doesn't change the laws of marketplace physics. It amplifies them and creates non-obvious yet high-quality moats.

It accelerates liquidity where matching is complex.

It unlocks data advantages where the real world is messy.

And in the right markets, it sets the stage for the next decade's $ 10 billion+ companies.

I've highlighted a few promising categories above just as examples. There are several more opportunities hiding in plain sight, waiting for founders who understand both the long arc of creating network effects and the new leverage AI provides. We live in exciting times indeed!