The Rich Are Getting Richer: Supporting a Global Economic System for Billions of People in the Digital Age
A common critique of Proof of Stake consensus algorithms is that “the rich get richer”, which roughly translates as “economies of scale favor wealthy incumbents over less resourced ones”. We can easily see that this is the case by looking at any industry that operates at scale: over time, those companies that are able to operate more efficiently and at a larger scale are able to use their influence and size to drive towards monopoly of the industry. We can even see this even at the level of nation states, where larger and more well-resourced nations attract greater political influence, make larger trade deals, and drive international affairs directly in their favor and at the expense of poor nations.
In the context of cryptocurrency consensus algorithm however, it is easy to see how this would be the case. Larger participants are able to put in more capital into their operations, increasing their overall share of the blockchain’s security system, and drive more rewards to themselves. The cycle repeats, and they are able to reinvest those earnings into larger operations, more reward earning potential, ad infinitum. These are continuous, zero sum games because the rewards offered are fixed, but the participants are potentially infinite (in a publicly-accessible and unlimited participation cryptocurrency security scheme) and the timeline is never-ending. If a competitor is able to effectively leverage the economies of scale of their operation, they gain market share at the expense of their competitors, and over time can exploit that differential to build an insurmountable lead. It’s easy to see how the rich are getting richer in these systems, and why that’s dangerous.
But enough about Proof of Work, I’m here to talk about the differences of economies of scale between Proof of Work and Proof of Stake systems.
Proof of Stake Economic Security
The idea behind Proof of Stake is that individuals take collateral they own, typically in the native currency of the cryptocurrency being secured, and put that collateral in escrow of the cryptocurrency protocol itself. In return, they gain access to the responsibilities of proposing new blocks and participating in the consensus of the final ordering of those blocks in the cryptocurrency’s history (which we call the “blockchain”). They are rewarded for these activities and thus are incentivized to participate by earning block rewards for publishing blocks (typically in the native currency itself). Participating in consensus is a necessity in a permission-less consensus system in order to finalize those rewards, otherwise it would be equally easy for another participant to “steal” those earnings for themselves. The goal of coming to consensus on history is to “finalize” those rewards (as well as fees and other earnings, such as MEV) and ensure they are not taken by others.
The concept of “Finality” is of particular note in a Proof of Stake algorithm design like Ethereum’s. Notably, it is possible to come to consensus that a part of history is “finalized”, or unable to be altered by the participants without violating the rules of the protocol (and suffering the consequences). This is unique to Proof of Stake, as Proof of Work does not have this property, at least explicitly. This property is valuable, especially for a global transaction settlement network, as it is possible to determine an explicit timeline where a transaction can be deemed “final” (again, unable to be altered without violating the consensus rules). We should note that there are practical limits to this property; namely, if an attacker is able to collect a super-majority of collateral at stake in the protocol (>66.6%) and is willing to burn it for some extra-protocol goal (such as causing havoc or reversing a high-valued transaction). The cost of this behavior is quite steep, with tens of billions in value staked in Ethereum’s PoS protocol, it becomes pretty unlikely that a single transaction will be of high enough value or an attacker resourced enough to commit this violation of the ruleset. But it is still possible.
There are still parallels to draw between Proof of Stake and other consensus systems like Proof of Work, fundamentally their structure is usually the same: participants obtain the resources necessary to participate in block proposal and consensus, and earn rewards for doing so. However, the different consensus protocols and even cryptocurrency-specific implementations vary in their every detail: the relative share of rewards earned for performing these duties, the available amount of extra-protocol earnings possible that are accessible to these participants, the opportunity cost of putting those resources to work in participating, and other external factors involved with participation.
Barriers to Entry and Limits to Economies of Scale
In this essay, I wanted to dive deeper into the specifics of these implementations, and how their choices drive emergent behavior in this protocols. These subtle differences can lead to vastly different outcomes, and it’s important that we analyze these academically so we can identify potential issues in their structure so we can build better systems.
For example, the Bitcoin blockchain currently offers a reward of 6.25 BTC every block for mining every 10 minutes (about 900 BTC per day), and pays about an average of 6% of that rewards in fees from users who send transactions. That’s about $20b in revenue annualized. In order to access that revenue, you need to purchase Application Specific Integrated Circuits (ASICs for short) from one of a few hardware manufacturers that offer Bitcoin mining hardware with the level of efficiency required to be profitable in mining (and not mining at a loss). The true costs of this hardware for larger scale buyers is very difficult to speculate on, because the sales data is largely unavailable for these companies, and they tend towards sweetheart deals for larger buyers. However, the largest costs by far are not the capital required to purchase these machines but the operational expense of electricity required to run these machines 24/7 in large pooled configurations to mine Bitcoin continuously.
While in theory Bitcoin mining is accessible to anyone, we can start to see that due to the structure of the network and it’s parameters, there are barriers to entry that exist such as access to ASIC hardware and space to store and operate them, reasonable electricity costs, and capital expenses related to purchasing those assets at a scale necessary to compete with others running similarly sized operations. Not everyone can access these opportunities, there is a certain level of capital necessary, and that’s not even considering the political costs of running a large operation and the variations in electricity cost between different regions.
On the flip side, there are also limits to these economies of scale. For example, you most likely can’t just go into a particular region and purchase *all* the available electricity to run your machines, that would attract the attention of governments who generally want to make sure other citizens also have fair access to that electricity as well. You also can’t find an infinitely large building to house them in, or build a powerful enough cooling system necessary to cool that much waste heat. Lastly, there’s only so much hardware produced by these manufacturers each year, and you can’t purchase them all (at least not without driving up the price past the point of profitability). Every single one of these potential advantages a competitor could optimize comes with a limit to their ability to optimize it. However, the game continues to grow every year in scale, based on these inherent limitations, opening up the ceiling of what is possible over time as more hardware is created, more access to power negotiated, hardware efficiency improves, and participates discover innovations to reduce their operational costs.
Compared to Proof of Stake, there are very big differences in these parameters and differing limitations and barriers that come into play. For example, Ethereum’s version 2.0 Proof of Stake design offers a block reward that decreases with the amount of participating validators (currently about 400k ETH per year). Currently on Mainnet Ethereum (which is still under PoW), there is about 4k ETH a day in transaction fees being paid out by users to miners, which should in theory should be the same or larger when the Merge upgrade of the current PoW network to PoS occurs. Lastly, there is over a half a billion dollars of detected MEV that has been extracted over the past year, a number that grows every day as we find more sophisticated means of measuring it. That’s about $6.4b in annualized revenue for Validators.
In order to become a validator, you must have 32 ETH (about $100k) to obtain a single validator slot, which gives you a one out of N chance of being randomly chosen to participate in the protocol. However, due to how the protocol is structured, validators are effectively in a pooled arrangement natively, which reduces the variability of block rewards they earn since everyone earns about the same. This does not affect transaction fees or potential MEV that the validators earn when they are chosen to publish a block, those are still rewarded to the block proposer specifically, meaning it’s important to stay online and responsive so that you don’t miss out on the opportunity to earn these fees. In fact, fees and MEV account for an increasingly larger portion of Ethereum’s protocol revenue, so up-time is vitally important.
There are more specific costs associated with downtime in PoS, more than potential loss of opportunity to earn rewards. For example, there is a leakage rate at which your stake is deducted for unresponsiveness. If enough of the pool of validators is unresponsive, the penalties grow exponentially in order to discourage consensus deadlock that leads to loss of up-time. This drives diversification of validator locations and software setups, which ensures up-time is as high as possible. It’s made a lot easier in that there is little variation in cost for hosting in different locations. Lastly, more complex validator setups with software and hardware redundancies in place (to handle larger number of validator slots) drive up engineering cost, cutting into overall profitability.
Calculating Barriers to Entry
In permission-less cryptocurrency security systems, in theory anyone can meet the entry and profitability requirements necessary to ensure you can participate in the continuous, zero-sum game of the protocol. In practice however, there are limitations, either implicit or explicit that prevent participation by lesser resourced individuals who might otherwise like to participate.
Calculating barriers to entry in Proof of Stake is easy. For Ethereum, that barrier is obtaining 32 ETH (again, about $100k), the (minimal) hardware spec (about $1-$2k to purchase or rental fees about $200/mo), and a reliable internet connection (about $100/mo). For an 18 month staking duration, this is about $105k, the majority of which is the cost of capital. Using our numbers above, that’s about $45k of revenue over the 18 month period (assuming no slashing losses or inactivity leakage, and also ignoring the human resource cost to running this setup), for a total annualized return of about 28% in ETH terms (USD terms vary a lot more obviously). That number will likely to go down a lot once we factor in these costs, and the Merge happens driving down the risk of participation. For a more reasonable estimate, let’s say that the annualized return would probably be closer to 5% under these future conditions.
For Proof of Work, calculating this number is a lot more difficult. Firstly, there is no in-protocol limitation to the number of validator slots available, and the capital costs vary a lot more (ASICs, electricity, real estate, maintenance, etc.). The ecosystem is also a lot more opaque to derive a similar profitability comparison, but we will do our best deriving from hashrate charts and publicly known costs for ASIC devices. ASICs have a profitability window of about 18 months before they start to fail or be replaced by new models, so hopefully it makes sense why we chose the same window above in our PoS calculation. Mining is also a lot more established and larger of an industry, annualized returns (in BTC terms) are likely much less than we calculate here.
Firstly, let’s make some more direct comparisons. The maximum amount of ETH is also the limit to the total slots that can be staked. If the limit to the total supply of ETH is 120m (there’s no limit in theory, but in practice it’ll be about this much), that is about 3.75m 32 ETH validator slots that can be filled. Bitcoin’s hashrate is currently 1.15 EH/s, and an Antminer S9 is about 13 TH/s and costs about $500. To own 1/3.75m-th of the Bitcoin network’s hashrate, you would need to own 2375 Antminer S9s, costing you about $1.19m for the hardware. The more efficient Antminer S17 Pro costs about $2k and produces about 53 TH/s, so you would only need to own about 583 Antminer S17 Pros, costing you only about $1.17m in hardware costs to own the same % of the network as one Ethereum 2.0 validator slot.
To compare apples to apples though, Ethereum’s market cap is only about 42% of Bitcoin’s, so the equivalent spend to own the equivalent share of the network is just about $495k (10.75 BTC), or 248 Antminer S17 Pros. It takes about $15.6k (0.35 BTC) in electricity costs for one Antminer S17 Pro to mine one bitcoin over about 4 years with $0.11/kWh electricity costs, so extrapolating out we can mine 248 BTC in 4 years with our Antminer S17 Pros for a cost of $3.87m (84 BTC). Normalizing that to our 18 month interval, that’s 93 BTC mined for an electricity cost of $1.45m (31.5 BTC), which is an annualized return of 47% (in BTC terms)! That number is a lot higher lately because of the Chinese mining crackdown and other structural changes in the Bitcoin hashrate, but we can assume it should be at least half that in practice.
To do a more equivalent comparison, let’s say we only had the ability to spend $105k on mining equipment and electrical costs for 18 months, how much would that be in practice? Well, we can purchase about 6 Antminer S17 Pros ($12k, or about 0.26 BTC), and mine about 2.25 BTC in 18 months using $39.6k in electricity costs (0.86 BTC), for an annualized return of about 33% (in BTC terms). That’s pretty similar to our return for our Ethereum staking slot! But we can begin to see how economies of scale come into play here, the more well-resourced participant (248 miners) can obtain a 42% better return than one who isn’t as well resourced (6 miners). The rich are getting richer!
Conclusion
As we can see, there really isn’t that much difference in profitability between mining and staking if we compare apples to apples in terms of operational and capital expenditures. Leaving aside discussions of “decentralization”, which has been well debated outside of profitability, we can see that the “Rich Get Richer” argument against Proof of Stake is fundamentally flawed. The reality of the situation is that any capital-intensive investment offers participants who can front more capital better returns than those that can’t put in as much. Hopefully this is enough to convince you that this argument is intellectually lazy at best, and falls apart under the lightest of observations. Also, I hope you take home with you an increased appreciation for comparing economic security of cryptocurrencies based on profitability, and an intuition for how economies of scale affect barriers to entry in permission-less security systems.
