Apple, Amazon, Google, and Microsoft have a combined market cap of around $9T at the time of writing (March 2025). What do all of these tech behemoths have in common? They are all platforms. A platform is a generic term for a specific business model that has been made possible with the Internet. Lately I’ve been thinking about what a platform really is; what the key characteristics are that differentiate platforms from other businesses, and why they are so profitable. Interestingly, as soon as you have a decent mental model of platforms, you start to see them everywhere, kind of like when you buy a new car and you notice the same model every time you drive. But the main benefit of developing a mental model of platforms is that it gives us a generalized set of characteristics and patterns that we can use to identify opportunities for new platforms that haven’t been built yet.
Core Features
In my view, a platform is defined by three core features: marketplace creation, network effects, and leverage. My view has been influenced by two books that illustrate these features in their own way: Navalmanack by Eric Jorgenson and The Price of Tomorrow by Jeff Booth1. They both touch on these aspects repeatedly but without explicitly calling out that they belong to “platforms”. I’ve taken the liberty to synthesize and expand upon the most salient parts below.
Marketplace Creation
A platform creates a bridge between buyers and sellers. This bridge is typically a website or an app that connects producers of some thing with consumers of that thing. The platform provides technology-based infrastructure that aggregates supply from suppliers and easily disseminates that supply to consumers. Take Uber. Uber is an app that connects suppliers of rides, i.e., car owners with free time, with consumers of those rides. Uber didn’t make the market; the market for cab rides already existed. What Uber did is they made an Internet-native marketplace and in doing so aggregated the supply of rides in a single application at the scale of the Internet. Now anyone with a vehicle and some free time could earn money driving people around.
Network Effects
Network effects are another key feature of platforms. Platforms that have captured the network effect get more valuable with each additional new supplier and consumer. And this value accrues to the platform itself, the suppliers, and the consumers. I use “the” network effect intentionally, because there are usually at most a handful of dominant platforms for any particular niche. Network effects are a winner-takes-most kind of game. If network effects were audible, they would be heard as a loud sucking sound. As the network reaches critical mass, the cost of leaving the network becomes prohibitive and the cost of not joining the network becomes costly as well. If we look at the network as a dynamic system, the network effect is a positive feedback loop that creates a gravitational force and feeds on itself until it hits physical limits, e.g., the number of Internet users.
Leverage
In physics, leverage is a force multiplier. In the context of business, there are three broad forms of leverage: labor, capital, and information. Each form of leverage is either permissioned or permissionless. When you apply something with leverage, you can have outsized benefit (if it works in your favor) or outsized loss (if it goes against you). All forms of leverage are recursive. Recursion is a fundamental concept from math and computer science that is characterized by self-reference and self-propagation.
Labor is the oldest form of leverage. It was used by kings to build empires and is used today in every business that has a payroll. The problem with labor is that it is expensive. You have to deal with all the messiness that comes along with managing people. Humans can’t work all the time. They need breaks and sleep and coffee. Labor is also a permissioned form of leverage – you have to convince people to forgo all the other opportunities they may have to instead come and work for you. Recursion is expressed in labor via human reproduction.
Capital is a newer form of leverage. It can take the form of money or something in the physical world, like land. Whenever you invest in the stock market or rent out your house on AirBnB, you are using leverage in the form of capital. One advantage of capital over labor is that capital never sleeps. And in fact, if used properly, it can work for you while you sleep. However, capital is another permissioned form of leverage. Someone has to agree to give it to you. Recursion is expressed in capital via margin and compounding.
Information is the newest form of leverage. Since Gutenberg and the printing press, data has grown at an exponential rate. According to Visual Capitalist, humanity will generate approximately 181 zettabytes of information in 2025. One zettabyte is equivalent to one billion terabytes. The value of this data has grown at an exponential rate as well, especially since researchers at Google released the seminal paper Attention is All You Need in 2017. This is the paper that introduced the attention-based Transformer neural network architecture and ushered in the era of Large Language Models (LLMs). An example of this explosive growth is seen in another pure platform, Reddit. Reddit was valued at $34 per share with an implied market cap of $6.4 billion when the company went public in March of 2024. At the time of writing, around a year after IPO, the Reddit share price has increased 3.5x. Why? Because Reddit sits on a vast trove of data that can be used for training LLMs and for doing other natural language processing tasks.
The reason that information is leverage is that it doesn’t just sit there. It acts on the world. The two subsets of information that have the most action potential are code and media. Code allows anyone with the time and skill to solve problems at the scale of the Internet. Media (books, podcasts, videos, etc) independently allow amateurs and experts alike to share their knowledge with an audience that spans the globe. Code has always been a permissionless form of leverage. Some of the biggest platforms (e.g. Facebook, Apple, Amazon) and impactful technologies (e.g. Linux) were started by one or two people as software side projects. Media is interesting because it used to be permissioned, but the Internet, and specifically platforms like Spotify, Substack, and Youtube, have made it permissionless2. You don’t have to work at a big corporation in order to build a website, create a podcast, or write an article with global reach. Once created, code and media work for you 24/7, no sleep required. Recursion is expressed in code by writing code that generates code itself. Compilers and transpilers fall into this category, and as we will see later, artificial intelligence does as well. Media is recursive in the sense that one can make a video about making videos, or write a book about writing books. In another sense, media that is created influences people to create further media in response to it. This recursive cycle of media creation has existed since the printing press and has accelerated as technology has advanced.
So what forms of leverage are used by platforms today? The most important form is code. Code has an upfront fixed cost that is incurred during development, but once it is finished, the marginal cost to scale globally is extremely low. Maintenance brings this up a bit, but amortized across the potential number of users, it is basically zero. Virtually all companies today use code in some fashion, however not every company is a platform. The distinction lies in the relationship between code and labor. Both non-platforms and platforms hire labor to produce code. However, platforms take it one step further and flip this relationship around by using code to create labor. The platform’s code creates a common marketplace for suppliers and consumers, but the platform doesn’t hire suppliers nor consumers. It outsources this “work” to the public. For example, YouTube is the marketplace for permissionless video, but it doesn’t have millions of people on its payroll to create videos. It outsources video creation to creators around the world. YouTube used code to create a virtually infinite amount of leverage in the form of labor to create videos. Spotify doesn’t keep musicians on payroll. Rover doesn’t hire dog sitters. Uber doesn’t hire drivers or buy cars. All platforms follow this pattern, by definition.
In addition to the platform itself using leverage, the marketplace that it creates also implies certain forms of leverage usable by its suppliers. And this supplier leverage is directly related to the income distribution of its supplier network. For example, the primary form of leverage as a supplier of YouTube videos is media. As discussed above, media is a permissionless, low-cost, highly scalable form of leverage. A single video can be watched millions or billions of times. When views are monetized as they are on YouTube, creator income scales with the number of views. YouTube is driven algorithmically to maximize the number of views, which results in creator income distribution following a power law. In other words, a small number of creators on YouTube capture the majority of the income.
To get an idea of the actual numbers, I asked Perplexity Deep Research to research platform income distribution, including YouTube. The following is an excerpt from the report (with my additions in [bold]):
For example, Millie, a full-time YouTuber, earned $110,000 in 2024 from 7.4 million views, translating to an RPM [revenue-per-mille, or revenue-per-1000 views, a metric used by Youtube] of approximately $14.86 [3]. However, earnings are highly volatile: a creator with 100,000 subscribers reported monthly income fluctuating between $1,000 and $6,000, while another with 1 million subscribers earned over $50,000 in a single month [1][3]. Nano-influencers, such as Jen Lauren, may earn as little as $213 monthly, illustrating the platform’s long-tail distribution [1].
Let’s take Uber as another example. The marketplace for Uber uses two permissioned forms of leverage: capital (cars) and labor (time spent driving). As a result, the supplier income is closer to a uniform distribution. The reason is the leverage available to Uber drivers doesn’t scale. There is only so much time in the day to drive people around. And a person can only drive one car at a time. Here is Perplexity again:
Unlike YouTube’s skewed distribution, Uber driver earnings cluster around median ranges due to standardized fare calculations and hourly caps. Drivers cannot scale income passively; each ride requires real-time labor. Promotions and surge pricing temporarily boost earnings but lack the compounding potential of viral YouTube content. For example, drivers in high-demand urban areas may earn $20–$30 hourly before expenses, whereas suburban drivers average $15–$20 [5]. The platform’s reliance on upfront pricing since 2022 has marginally increased pay for short rides but introduced volatility as algorithms adjust to market conditions [5].
The distinction here is important. All platforms use leverage in the form of code to create marketplaces for a set of suppliers. This code scales globally for the platform itself. However the leverage available in the marketplaces they create varies depending on the platform, and the value of participating in these marketplaces as a supplier varies accordingly. Platforms that create marketplaces with permissioned forms of leverage, such as Uber and AirBnb, scale less well for the suppliers in those marketplaces. On the other hand, for platforms with code and media based marketplaces, such as cloud providers (e.g. Amazon Web Services, Microsoft Azure, etc.), Spotify, Substack, and Youtube, scale is only limited by the number of Internet users. Platforms have become so dominant because of their clever use of the recursive property of leverage. They use leverage to create more leverage.
A significant side-effect of this recursive property is that existing forms of leverage combine to create entirely new forms. Labor and capital creating computer hardware and code is one example of this. Code is a relatively new type of information leverage that has only existed for about 80 years. A more recent example is artificial intelligence, which was created by labor, capital, and code. One could be pedantic and claim that AI is just another piece of code. Technically that is true, but in my view AI is more than just a new software program. It is a new type of information leverage that is going to have a bigger impact on human progress than the Internet itself. Bigger than the Internet? Yes. The reason is that the Internet didn’t have the Internet before it. AI does. The frontier LLMs are trained on the open Internet itself, and they present the information in a concise way that is a 100x improvement over a maze of Google searches. Another reason is that AI is a new permissionless form that can intelligently create the two forms of permissionless leverage we’ve already discussed: code and media. What used to require significant labor and time can now be done in seconds by a machine. This is the recursive property manifesting an entirely new form of leverage. The next logical step in this recursive process is AI creating more and better AI. I’ve personally already used LLMs to help me create machine learning algorithms in my work. Currently this involves some human in the loop, because LLMs are currently limited in what they can do. However, even this is changing rapidly. Anthropic, one of the AI startups developing the frontier LLM Claude, published an open specification called Model Context Protocol (MCP) in 2024. MCP allows LLMs to do things beyond giving simple text or image responses, such as interact with APIs, browse the web, take screenshots, and create files. You can think of MCP as the USB of the AI world, enabling LLMs to connect and interact with their environment.
My question is, what’s next? What new form of leverage will labor, capital, code, and AI create? The obvious answer is robots. Humanoid robots, drones, and full self-driving cars are basically here today. Elon Musk and Tesla are creating a future where normal people don’t buy cars and where single family homes don’t have garages. Instead, if you need to go somewhere, you will hail a robotaxi that you pay a monthly subscription for. In the next 5 years, full self-driving will be so much safer than humans that entire business models will have to be re-imagined (car insurance) or may not exist at all. That is the horizon we can see. What lies beyond that is up to our imagination. It will likely involve using new leverage to replace old, or to make the old forms work more efficiently, like AI is doing to labor today.
Identifying Platforms
Now that we are familiar with the key characteristics of platforms, we can start to identify the ones that exist today. In the following table, I’ve broken down each platform according to the 1) marketplace they’ve created, 2) the primary form of leverage used by the platform (“Platform Leverage”) and 3) the primary leverage available to suppliers in the platform’s marketplace (“Marketplace Leverage”). I’ve also included the leverage dependencies, which are the forms of leverage that are inputs into the primary form of leverage, i.e., the leverage used to create the primary form. You can think of the primary form of leverage as the “newest” form of leverage that is used by either the platform or a supplier in the platform’s marketplace. I’ve tried to think of as many platforms as I can. If you can think of more let me know and I will add it to the list.
| Platform | Marketplace | Platform Leverage | Marketplace Leverage |
| YouTube | Video | code (dep: labor, capital) | media (dep: labor) |
| Substack | Writing | code (dep: labor, capital) | media (dep: labor) |
| Spotify | Music, Podcasts | code (dep: labor, capital) | media (dep: labor, capital) |
| Rover | Pet care | code (dep: labor, capital) | labor |
| Uber, Lyft | Transportation | code (dep: labor, capital) | labor (dep: capital) |
| AirBnB | Housing | code (dep: labor, capital) | capital (dep: labor) |
| Etsy | Handmade goods and art | code (dep: labor, capital) | labor |
| Amazon, eBay | Books, physical goods | code (dep: labor, capital) | labor and capital |
| Facebook, X, Instagram, Reddit, Google, etc. | Information | code (dep: labor, capital) | media (dep: labor) |
| Coursera, edX, etc. | Education | code (dep: labor, capital) | media (dep: labor) |
| Business networking | code (dep: labor, capital) | media (dep: labor) | |
| Betterhelp | Therapy | code (dep: labor, capital) | labor |
| Github, Gitlab | Code | code (dep: labor, capital) | code (dep: labor, capital) |
| Apple, Android App Stores | Code | code (dep: labor, capital) | code (dep: labor, capital) |
| AWS, Azure, GCP | Compute, Network, Storage | code (dep: labor, capital) | code (dep: labor, capital) |
One callout for the last line: the reason that cloud providers are so damn profitable is that they’ve positioned themselves upstream of this first wave of platform innovation. Every platform (and most other non-platform businesses too) today uses code that runs in the cloud in some capacity. Cloud providers created a marketplace featuring compute, storage, and networking at extremely low cost that dramatically lowered the costs of creating a software-based startup. This has led to the proliferation of cloud-native platforms that we know today.
You can see from the table that platform leverage follows a uniform pattern, with code being the primary form. However there is significant deviation in the marketplace leverage. All of the platforms themselves enjoy the scale made possible with the scale of the Internet. That is not true of each of the marketplaces they create. You can’t really scale dog walking or Uber driving.
Or can you?
Platforms of the Future
What if you had a personal Rosie robot in your house, like straight out of the Jetsons? It could put away your groceries, or perhaps walk your Rover client’s dog while you put away the groceries? That moment is closer than you may think:
What will the platforms of today look like with AI and robots integrated into our life? AI is causing the labor cost of producing code and media to trend to zero. It has never been easier to start a Youtube channel or create a software application. I’ve easily seen at least a 10x increase in my own software development efficiency using AI. Robots have already begun to replace labor in certain circumstances. Amazon “employs” about 750,000 robots in fulfillment centers around the world, and I expect that number to only increase in the coming years. And instead of driving people around yourself, imagine having a robotaxi that you rent out for rides. What once depended on labor can now be done using robots and capital while you spend time doing something else. That “and capital” is important, because it places a limit on the scalability of the “Ride” marketplace due to its permissioned nature, at least initially. However the recursive cycle of innovation we are witnessing is massively deflationary, so over time the capital required should tend to zero.
The general pattern at play here is newer forms of leverage are used to increase the efficiency of older forms, or replace them outright. For platforms specifically, this results in lower cost and in some cases may shift the primary form of leverage available in the marketplace to entirely new forms. New forms of leverage beget new problems as well. And these problems can in turn be solved by new platforms. One example is PromptBase. LLMs work best when the user submits a prompt that primes the associations within the neural net to the objective at hand and provides detailed instructions of what they want. PromptBase is a platform that created a marketplace for LLM prompts. Anyone can sell access to useful prompts they have made, and users can search for prompts that solve problems they have.
But what about platforms that don’t exist yet? How do we identify opportunities for new ones? All the existing platforms solved some problem that was centered around two sides of a market. So you have to start with a problem or a need that people or businesses have. And once you identify a problem, instead of trying to solve it all by yourself, you ask if you could outsource it to suppliers in some marketplace.
For example, people will always want more free time, so looking for ways to give them that time would be one path to pursue. One example I’ve thought of is home cooked meals. People with busy schedules may get tired of eating out or think DoorDash is too expensive or just want a healthy home cooked meal. What if they could hop on an app and schedule a local chef (the supplier) to come to their house and cook a meal for them on their busiest nights? This would be a platform where the suppliers are people who enjoy cooking and have time to do so, and the consumers would be people that value home cooked meals but don’t necessarily have the energy or time for them. In theory, the platform itself could scale globally through the use of code, but the primary marketplace leverage would be labor, until we have robotic chefs of course.
An “AI-native” example would be a platform for AI agents. AI agents solve two key limitations of LLMs. The first is that each LLM has a “knowledge cutoff”, which is a date in the past when their training stopped. Any information that is created after this date is inaccessible to the LLM. The second limitation is that LLMs are limited to reading and writing information in a particular “mode”, such as generating text or images. LLMs themselves can’t use a browser, for example. AI agents solve these problems. Agents enable LLMs to access real-time information. If you want to know the weather forecast for today, you can use an agent to get that information from an API somewhere, then feed that information back into the LLM to present to the user. The MCP protocol mentioned above allows for anyone to create an MCP server (i.e., an agent) for carrying out particular tasks, such as looking up the weather forecast. An AI agent marketplace could be created based on this protocol that connects creators of AI agents with the people and businesses that need the agents to solve their problems. For example, a “Microsoft Excel agent” could be written for doing particular spreadsheet manipulation tasks. This agent would be written once and sold in a marketplace. Users that need to perform spreadsheet tasks would search for that agent in the marketplace and pay a small fee to use it. Businesses could develop their own AI agents to answer customer queries depending on real-time information that would otherwise be found “manually” (e.g. through a Google maps search) in a pre-AI agent world. Businesses in various professions with specialized knowledge, such as medicine, law, and engineering could monetize that knowledge by creating an AI agent that the market could use when needed. In this example the platform uses code to scale, while the marketplace utilizes AI as the primary form of leverage to its suppliers.
Hopefully by now you can see the opportunity that platforms present, both to its creators, as well as the suppliers in the marketplaces they create. Of the three core platform features, marketplace creation, network effects, and leverage, it is the creative, recursive use of leverage that yields the greatest impact on the value of the platform. Now that you have a solid understanding of what a platform is and how they work, you can start to think of ideas for new platforms. And with the impressive capabilities of AI out there today, there has never been a better time to bring an idea to life. If you do think of something, let me know! We could be co-founders :).
Thanks for reading,
Connor
- I recommend them both ↩︎
- The transition of media from permissioned to permissionless and its impact on society is examined in detail in The Revolt of the Public and the Crisis of Authority in the New Millennium by Martin Gurri. Highly recommend ↩︎