Multicoin Capital: On the Network Effects of Value Stores
This article was published by Multicoin Capital, author: Kyle Samani, Managing Partner.
This article is closely related to the piece “The Fallacy of Smart Contract Network Effects”. Bitcoin has network effects. Many evangelists of crypto technologies assert that Bitcoin's network effects are so strong that over-bitcoinization will be inevitable. But this is not entirely correct. The network effects of Bitcoin as digital gold are not as strong as many claim. Network effects are subtle and often misunderstood.
In this article, I will explore the network effects of Bitcoin as digital gold and digital cash. I will also discuss other competitive moats that do not possess network effects. For general background knowledge, I recommend reading this blog post on Medium, this slide deck from a16z, this blog post from Techstars, and this article on data network effects. If you want to read a more comprehensive long-form piece on network effects and technology platforms, I suggest looking at Platform Scale and the book written by Sangeet Choudary.
The foundation of network effects occurs as an emergent property when the value to existing users increases as more people use a product or service. There are several different types of network effects:
- Direct network effects—the increase in usage leads to a direct increase in value.
Direct network effects are effective because, as the underlying product/service gains adoption, existing users can choose to interact with more and more people. Essentially, all closed-loop communication networks exhibit this network effect, including internet-based services like Facebook and WhatsApp.
Indirect network effects—the increase in product usage spurs the production of increasingly valuable complementary products, thereby increasing the value of the original product. Operating systems (OS) are the most well-known type of product benefiting from indirect network effects. Application developers are attracted to the operating system to build on it, hoping to reach consumers; once a given operating system has built applications, it becomes more attractive to new consumers, creating a larger market for future application developers.
Two-sided network effects—the increase in usage by one group of users increases the value of complementary products to another group of users, and vice versa. Some well-known examples include eBay, Uber and Lyft, Airbnb, and Amazon's marketplace business. In these networks, consumers benefit from more choices and competition among suppliers, which drives more consumers, further attracting more suppliers.
Data network effects—a product becomes smarter as it gains more data users driven by machine learning. Data network effects occur in most modern cloud-based applications today, although their relative advantages vary greatly depending on use cases and complexity.
Quantifying the strength of network effects Measuring the precise strength of network effects is quite difficult. It is not a precise science. The challenge of this task is particularly great because the marginal value of new users in the system changes over time. For example, in recent years, many of my friends have started deleting their profiles on Facebook, but Facebook is still just as useful to me as it was three years ago. Losing 5% or even 10% of my Facebook friends has a relatively small impact on my experience on Facebook because I still have 500 friends who are still using it.
It is often said that the defensive moat of companies closely associated with network effects can be quantified using Metcalfe's Law, which states that the value of a network is proportional to the square of the number of users. To make it easier to understand, it can be explained that companies constrained by Metcalfe's Law have network effects of n^2. After Metcalfe proposed a practical definition of network value, others have pointed out the fallacies of this definition. There is currently no known network that can consistently exhibit n^2 network effects during its growth.
Moreover, the initial assumption driving n^2 is that all connections in the network have equal value. The reality is different; it is now widely believed that the network effects of most networks are closer to nlog(n) rather than n^2. While this certainly makes more sense (nothing can grow indefinitely at a square rate), even nlog(n) is still an ever-superlinear curve.
What we see in reality is that not all connections have equal value, and after reaching a certain point, the value of each marginal connection in the system begins to decline (for example, the addition of 10 million Facebook users in Asia has very little value to existing users in the U.S.). In practice, the best-case scenario for network effects resembles an S-curve rather than n^2 or n*log(n).
In practice, there are many cases that empirically demonstrate the S-curve nature of network effects. This is why Apple survived in the 90s (if Windows' network effects were indeed n^2, Apple might not have survived), why there are so many messaging applications (WhatsApp, Telegram, Facebook Messenger, Signal, etc.), why Lyft can effectively compete with Uber (as long as I can summon a car within two minutes, I don't care which app signed up how many drivers), and why so many niche e-commerce stores can compete with Amazon. Why do people always say networks are affected by n^2 or n*log(n) network effects? Because among these three curves, it is difficult to distinguish the differences in the far left part of each curve:
It is not until the right half of each curve—this part only appears when the network reaches critical mass—that substantial divergences among the three curves emerge. The n^2 curve continues to accelerate quadratically. The n*log(n) curve will also continue to accelerate upward, although at a much slower rate. On the other hand, when a network passes a saturation point, the S-curve shifts from superlinear to sublinear.
Of course, not all networks fit the same S-curve, and not all networks are influenced by the best-case scenario of the S-curve of network effects. As shown in the left half of the S-curve, some network effects can never achieve exponential network effects. Some networks are influenced only by log(n) network effects from the beginning, making them perpetually sublinear rather than initially exhibiting a superlinear S-curve. The most common example of log(n) network effects is trading a liquid, fungible asset.
Even if we assume that each new user increases daily liquidity, the marginal value of additional liquidity for all existing users will diminish. This holds true even in the early stages of the network. This curve is never superlinear; it always behaves as sublinear.
Let’s consider a simple example where each new user trading some fungible asset increases the daily liquidity of that asset by 0.01%. With 100 users, the daily liquidity is 1% of the asset's market value. With 1,000 users, the daily liquidity is 10% of the asset's market value. With 10,000 users, the daily liquidity is 100% of the asset's market value. With 100,000 users, the daily liquidity is 1,000% of the asset's market value (10 times the daily trading volume). If a user holds 0.1% of the asset being traded, the value provided by each marginal user in terms of liquidity becomes increasingly negligible.
Technically, the marginal value of liquidity decreases as the number of users increases, thus increasing liquidity, but in practice, the marginal liquidity gains will be very low, making it difficult to perceive for not only a given user but for all existing users. All trading of fungible assets exhibits approximately log(n) network effects, which can be intuitively represented as:
There is ample evidence to suggest that this is empirically correct. If the network effects of trading fungible assets were superlinear at any point on the curve, we would not have so many crypto exchanges. What we observe is that if an exchange has a certain level of liquidity—even if it’s just a fraction of the market leader—this is usually sufficient to keep an exchange operational and provide reasonable liquidity to market participants.
Network effects of digital gold What kind of network effects does digital gold exhibit? To answer this question, let’s look at how users utilize digital gold. The purpose of a value store like digital gold is simply to store value for future consumption. Aside from the time it takes to convert digital gold into something else, digital gold just sits there, doing nothing. It cannot benefit from the addition or reduction of new users. When a user wants to liquidate their digital gold to consume other goods or services, they need to find liquidity: someone willing to buy the digital gold. This can be achieved at exchanges that specialize in trading fungible digital gold. The utility of digital gold is a function of its liquidity.
As mentioned above, this means that Bitcoin's network effects can be approximated as log(n). Network effects of digital cash What kind of network effects does digital cash exhibit? To answer this question, let’s look at how users utilize digital cash. The purpose of digital cash is to store value and serve as a medium of exchange.
Additionally, digital cash can serve as a unit of account. Therefore, the overall utility of digital cash depends on how many merchants are willing to accept digital cash as payment for goods and services. This is similar to the direct network effects described above (telephone graph). The more people who accept digital cash payments, the more merchants can do business with existing users. All major global currencies exhibit this network effect within their respective jurisdictions. Because merchants and consumers must pay taxes in their local legal tender in each jurisdiction, they choose to receive wages (employees) and income (businesses) in their local legal tender. This creates a powerful network effect, as few people are willing to bear the balance sheet risk of holding a currency that is subject to price volatility relative to the currency they use to purchase goods/services and pay taxes.
Intuitively, this is likely influenced by S-curve network effects. The first 50% of merchants accepting digital cash payments makes it more useful than the latter 50%. Value storage vs. utility Bitcoin maximalists might argue that the above viewpoint is semantic. Specifically, they might say, "Of course, new users make Bitcoin more valuable. They buy it and hold it, which by definition makes it more valuable! The discussion about liquidity is a distraction." While this is correct in a narrow sense, it overlooks the reality of competition: what if something else becomes digital cash and achieves superlinear network effects? This is the broader point I want to express.
On its own, the sublinear network effects of liquidity can distract people. But if something else becomes digital cash with superlinear network effects while Bitcoin remains digital gold with sublinear network effects, then Bitcoin will be surpassed. In the crypto space, this framework is often described as the value storage (SoV) vs. utility debate. The SoV perspective is based on reflexivity: the more people hold it, the more valuable it becomes, thus encouraging more people to hold it.
Of course, reflexivity is subject to fluctuations. This leads to excessive volatility, and thus instability, completely undermining the purpose of value storage. When prices are consistently rising, it is easy for people to believe in the SoV hypothesis. But when prices decline, the potential value of utility creates an organic price floor. People may easily forget this fact—cryptocurrencies are still in their early stages. There are 7 billion people on Earth. Fewer than 50 million people own cryptocurrencies. Our global saturation is less than 1%.
In an open-source software world where all functions can be replicated, the key to winning is to achieve network effects as quickly as possible. This is why the potential strength of network effects is so important. Whether these benefits can successfully compound will create a massive difference in the terminal value of networks with tens of millions or even hundreds of millions of users. Other moats Network effects are just one type of competitive moat. There are many types of moats. Other moats supported by Bitcoin maximalists include "brand recognition" and third-party ecosystem integration, such as exchanges, ATMs, other financial products, hardware, and mobile wallets.
To explore the power of these moats, I will compare Bitcoin with Ethereum. This is not to say that Ethereum is likely to replace Bitcoin. I just want to use Ethereum as an example to illustrate what a competitive network can achieve in less than three years after launch. Brand recognition is indeed a moat.
Bitcoin is the leader in cryptocurrency. However, suggesting that Bitcoin's brand is somehow insurmountable is a grave mistake. No brand is insurmountable. For an open, permissionless brand like Bitcoin, there is no good way to measure its brand value, but we can use Google Trends as a rough measuring tool.
Bitcoin is blue, Ethereum is red. At its peak, Bitcoin's search frequency was about 11 times that of Ethereum. Today, the gap is 8 times. Given how volatile this space is and how rapidly it evolves, this lead could disappear in a few years. What about third-party integration?
In this regard, Ethereum is almost on par with Bitcoin: Exchanges— all major exchanges support fiat-to-Bitcoin and fiat-to-Ethereum trading pairs. Hardware wallets— all major hardware wallets support both Bitcoin and Ethereum. ATMs— to my knowledge, all crypto ATMs support both tokens. Mobile wallets— both Bitcoin and Ethereum have a plethora of mobile wallets available for iOS and Android. Other financial products— Bitcoin has a head start in futures on the Chicago Mercantile Exchange (CME), Chicago Board Options Exchange (CBOE), and Nasdaq.
But considering Ethereum's trajectory, it seems quite reasonable to predict that it will achieve parity within 24 months. Again, my point is not that Ethereum will necessarily surpass Bitcoin, but that the scope of third-party integration is not insurmountable. Conclusion Network effects and competitive moats are often misunderstood.
Contrary to common perception, no network exhibits n^2 network effects; in fact, many networks exhibit log(n) network effects, particularly evident in the trading of fungible assets. As digital gold, Bitcoin will be subject to permanent sublinear log(n) network effects, while as digital cash, Bitcoin can achieve superlinear network effects as the adoption rate of cryptocurrencies grows from 1% of the global population to 50%.
By definition, a cryptocurrency that becomes the dominant value store must exhibit superlinear network effects during its growth. Furthermore, other types of competitive moats, such as brand and broader ecosystem integration, do not show increasing scale benefits that can easily overcome competitive networks with superlinear network effects.
We have ample evidence to prove this is correct. The battle to become a super winner in the crypto space has only just begun. At the far left 1% of all network effect curves, the differences are not apparent. It is easy for people to think that network effects have already begun to take effect before they actually occur. Thanks to Chris Dixon and Matt Huang for their feedback on this article.