AI and Network Effects
And, why B2B Distribution is the AI Goldmine
Hi everyone - Merry Christmas and Happy Hanukkah!!
I was listening to a recent conversation with Microsoft CEO Satya Nadella on the BG2 podcast with Bill Gurley, the iconic investor formerly with Benchmark, and Brad Gerstner, another innovative investor and founder of Altimeter.
I highly recommend giving the whole 90 minute interview a listen.
As Applico Capital invests in AI companies, we’re actively shaping, challenging and rethinking our investment theses in this fast-moving space. A recent question we asked ourselves:
How can an AI startup become a winner take all platform?
Given that we literally wrote the book on platform business models, it’s been fun rereading parts of our book, Modern Monopolies, which is surprisingly still relevant to this thought exercise. Here’s our topics for today’s newsletter:
Where are Network Effects in the AI Stack
How can AI Capture Network Effects?
Critical Mass in Network Effects
Why is B2B Distribution the AI Goldmine?
Examples
Consumer - Producer AI Role Pairings
Where are Network Effects in the AI Stack
In Satya’s interview he makes a couple comments about network effects in different parts of the AI technology stack:
“In the hyperscale side and infrastructure, absolutely not will there be a winner take all… there will be multiple winners.”
“Network effects will always be at the app layer, the software layer.”
Listen to Satya talk about the AI tech stack and winner take all dynamics starting at 19:42.
Highlighted in red is the part of the stack that Satya thinks you could see AI capture winner take all dynamics. The “app server” correlates to the yellow “ML Ops” in the graphic above – Satya also thinks there won’t be a winner take all dynamic in this part of the stack. All the hyperscalers (Google, Amazon, Microsoft) will have their own versions of the infrastructure and app server part of the stack – as will other AI infrastructure startups.
An application server for AI is a platform or framework that facilitates the deployment, management, and operation of artificial intelligence (AI) models and applications. It acts as the middleware between the AI models and the end-users, providing the necessary infrastructure to integrate, scale, and serve AI functionalities efficiently. Microsoft Fabric, for example, is Microsoft’s app server.
The application layer, as Satya talks about it, is where the AI and humans intersect. It’s where the consumers and businesses interact with the AI. OpenAI, for example, has an application called ChatGPT that has been wildly successful with consumers. But, OpenAI’s foundational models are also key AI infrastructure that enables the rest of the ecosystem to build on top of.
How Can AI Capture Network Effects?
The platform business model facilitates the exchange of value between two parties, typically a consumer and a producer.
In Agentic AI, where there are different AI bots performing different tasks – imagine if both parties to a transaction are using AI bots. And, those AI bots begin to communicate with one another – to help facilitate that transaction.
Now, imagine that those AI bots, on both sides of the transaction, are being provided by the same AI company. That AI company is now a platform business.
As a gating criteria, the industry needs to be fragmented and large enough for a winner take all dynamic to be possible. Industries where one side of the equation is more consolidated, with fewer participants, are going to be harder to successfully penetrate.
Critical Mass in AI Network Effects
Critical mass is the point in which the network effects start working for you, rather than against you.
When the AI platform is talking to itself because it’s penetrated enough consumers and producers, the AI can perform its tasks much more quickly and effectively. Until that point, the AI still needs to interact with humans! When interacting with humans, the AI will be slower and the accuracy won’t be 100%.
Or, the AI is interacting with other AI agents. Satya also talks about this in the same interview starting at 28:00– when asked about the future of AI agents by Bill Gurley. In the enterprise, Satya expects there to be connectors that enable different AI systems and apps to interact and share information with one another.
Why is B2B Distribution so important for AI platform businesses?
B2B distribution is the classical middleman – sitting between suppliers, business customers and logistics providers. As AI companies roll-out solutions to their supply chain ICP (ideal customer profile), B2B distribution – at $8 trillion in size – will probabilistically be on one side of that transaction.
Think of the distributor as both a buyer and seller of products, a logistics company and a financier all compiled into one entity - the versatility of the distributor business model creates a variety of transaction types between distribution and other parties in the supply chain. If an AI company is providing a solution to the supplier of a distributor, what’s the reciprocal solution for the buyer (the distributor)? If an AI company is providing a solution to the customer of a distributor, what’s the reciprocal solution for the seller (the distributor)?
The solution for consumers and producers is ultimately different, but the underlying technology and foundational AI is hopefully similar and can share synergies between both personas.
What are some examples?
We’re seeing companies that assist the customers of B2B distributors with procurement start to expand into offering solutions for B2B distributors to build proposals.
For example, Choco is providing digital commerce technology in food service to restaurants and distributors. They have embraced the AI vision in the past couple years and have gone all-in on the subject. They have a closed loop – where they are providing AI to the food service distributor that can automate order taking, adjusting delivery schedules and other common tasks requested by their restaurant customers. And, Choco provides AI to the restaurant to help them automate their ordering with distributors. We could see Choco naturally expand to help the restaurant automate their account payable reconciliation using AI. If the restaurant is procuring through Choco’s AI and the distributor vendors are using Choco’s AI to receive the orders – one would think that Choco’s AI would also be in an advantageous position to assist the restaurant with back-office AP reconciliation and the distributor with their AR workflows.
This is one example of many where – assuming Choco can reach critical mass – you could start to see the network effect create lock-in in Choco’s favor. It’ll be easier, faster, and more secure if Choco’s AI bots are talking to themselves across the network of restaurants and distributors.
Consumer - Producer Pairings in Distribution:
Consumer role on the left. Producer role on the right.