The Complicated Relationship Between Ethereum-based NFTs and the Environment

Co-written with Wei Dai.

Part 0: Intro / tl;dr

There has been a controversy regarding Ethereum network energy consumption, its CO2 emissions and “gas.” Below are two excerpts from two popular articles that directly contradict each other.

NFTs are harming the planet (Statement 1):
“From the total amount of Ethereum gas and energy consumed …, we can calculate the energy footprint per unit of gas. .. Using the footprints per unit of Gas .., I can calculate the (energy and carbon) footprints relating to the NFT” –  Memo Akten – The Unreasonable Ecological cost of #CryptoArt

In other words: Ethereum transactions have high carbon footprints due to usage of network resource (gas).

NFTs are not harming the planet (Statement 2):
“.. claims that more computationally-intensive transactions such as minting NFTs cause more energy consumption in the Ethereum network are untrue. Higher gas transactions … are not correlated with higher energy consumption.” – SuperRare – No, CryptoArtists aren’t harming the planet

In other words: High usage of network resource (gas) does not increase network energy consumption.

Who is correct? In this article, we will show and hopefully convince you why neither of them are correct. The truth is more complex and somewhere in between the two claims.

Thesis of this article
Attention drawn towards NFT / Ethereum space, as well as increases in demand for transactions, lead to higher Ethereum price and gas price, which lead to rising mining incentives that eventually result in increasing hash rate, energy consumption, and CO2 emission of the Ethereum network. The impact of individual transactions on the environment is (1) non-zero, but (2) not as dramatic as claimed by Akten, and unfortunately (3) hard to quantify and calculate.

Precisely pinpointing the footprint of a single transaction on a PoW blockchain is not easy. There is also no single universally-accepted method for carbon foot-printing Scope 3 emissions. (Carbon footprint: current methods of estimation, ‘Carbon footprinting’: towards a universally accepted definition, EPA’s Guideline, calculating scope 3 emissions, Technical guidance for calculating scope 3 emissions)

Outline
In part I, we first give some background on energy consumption and revenue of Ethereum miners. In particular, only ~50% of total mining revenue comes from transaction fees, and only <1% of energy used goes towards processing these transactions. We then demonstrate why both statement 1 (Akten & Carbon.fyi) and statement 2 (SuperRare) are misleading via two simplified examples.

In part II, we directly look at how energy consumption of the Ethereum network changes in response to changes in mining incentives, in particular the price of Ethereum and gas.

In Part III, we apply findings from Part II to investigate the environmental impact of NFTs on the Ethereum network.

A technical overview:

Costs (Energy consumption)
C1: > 99% energy consumption toward running computational lottery
C2: < 1% energy consumption toward direct processing of transactions
Total: 100%
Revenue (ETH received by miners)
R1: 2 ETH created 
R2: ~1-3 ETH from transaction fees (paid by initiators of transactions)
Total: ~3-5 ETH

Statement 1 (Akten & Carbon.fyi): R2 is responsible for (C1 + C2)
Statement 2 (SuperRare): R2 is only responsible for C2, because C1 is “fixed” and needs to be spent even if R2 is reduced to zero
Our thesis: Costs (dominated by C1) changes in response to changes in overall revenue in dollars, i.e. (R1 + R2) * price of Eth

Disclaimer: The purpose of this article is not to take a side, but to (1) educate and to clear up the possible confusion caused by some misleading statements of both Pro-NFT and Anti-NFT crowds and (2) discuss the relationship between NFTs and the environment using facts about the proof-of-work Ethereum blockchain.

PART I – Costs (energy consumption) and Revenue (mining rewards) of the Ethereum PoW network

We will not be including the infrastructure cost of running Ethereum mining nodes (space, hardware, maintenance, etc). Instead, we only look at power consumption and CO2 emissions from power consumption, similar to the approach taken by Akten. In the short to medium term, power sources of Ethereum miners are relatively constant. Hence, cost is not only the cost of running Ethereum miners (electricity) but also directly related to CO2 emissions from consumption of electricity. For the rest of the article, we will use “cost” (dollars spent on electricity) and “emission” (CO2-equivalent emitted from production of electricity consumed) interchangeably.

Costs (Energy consumption)
C1: > 99% energy consumption toward running computational lottery
C2: < 1% energy consumption toward direct processing of transactions
Total: 100%

How does the Ethereum network work? How much energy does a block consume?

Ethereum 1.X does 2 things:

    1. C1: Run a large scale computational lottery, which accounts for more than 99% of the energy consumption of the network. 
      • Ethereum miners (programs running on computers with powerful GPUs) compete with each other to create blocks, which contains numerous (usually hundreds of) transactions.
      • In the current version (1.X) of Ethereum, the computational lottery is called “proof-of-work” (PoW). The more computational resources a miner has, the more it is likely to create a block and then claim the mining incentive.
      • This is necessary in securing the Ethereum 1.X network. Ethereum 2.0 swaps out PoW with Proof-of-Stake, which is a different mechanism of securing the blockchain that does not consume a ridiculous amount of energy.
    2. C2: Process transactions (Examples: Minting/trading NFTs, sending ETH, deposits/withdrawals to defi accounts, etc.). This accounts for less than 1% of the energy consumption of the network. 

Revenue (ETH received by miners) per block
R1: 2 ETH created 
R2: ~1-3 ETH from transaction fees (paid by initiators of transactions)
Total: ~3-5 ETH

What is the incentive offered by the Ethereum network for creation of one block?

    1. R1: Flat reward. Currently, the Ethereum network creates 2 Eth every 13 seconds as reward for miners.
    2. R2: Transaction fee is the amount of Ether (ETH) paid to the miner for processing the transaction. Transaction fees increase or decrease depending on the congestion on the Ethereum network. More people on the network trying to make a transaction = higher transaction fee. Usually, blocks contain hundreds of transactions. At the time of writing, the average transaction fee per block is about 1-3 ETH.

Overall mining incentive is obtained by adding up the flat reward and transaction fees.

Mining incentive (in $) = (2 ETH + transaction fees per block) * price of ETH (in $)

Transaction throughput of Eth 1.X is fixed, regardless of energy consumed.
Ethereum is designed to produce a block on average every 13 seconds, regardless of how many miners and how much computational resources are involved. Less miners and/or less computational resources involved = less energy consumption. Number of transactions that can be processed remains the same.

Why Statement 1 (Akten & Carbon.fyi) is misleading:
“We could place the burden of emissions entirely on miners, but that would not be fair to them… For this reason, we allocate emissions based on your usage …” Carbon.fyi

Methods used by Carbon.fyi and Akten place the burden of carbon emission generated by miners (C1 + C2)  entirely on the users of the network (R2). This overestimates the emission responsible from each single transaction. To see this we consider these numbers in a different context.

Example

      • Alice is the sole customer of a coal power plant and purchases $50 worth of electricity from them.
      • The power plant costs $100 to operate, and will go out of business if there are no other revenue sources, so the government subsidizes the power plant and gives them $50 for free.
      • How much is Alice responsible for the carbon emissions of the power plant?

Some might say that Alice should be held 100% responsible for all emissions of the power plant since she is the sole user. Indeed, this is the approach taken by Akten and Carbon.fyi. We argue that this method is inaccurately accounting the responsibility of Alice. To see this, consider the economic contribution of the parties involved: Alice and the government each paid $50 to the company. Without either of these two payments of $50, the powerplant will not be able to cover the operating cost, and likely to shutdown in the long-term. From the perspective of the powerplant, it does not matter who their product (electricity) goes to, only how much they are receiving as revenue. Alice and the government each paid $50 in this example, and they share the responsibility of keeping the powerplant in business, and in turn the emissions generated from the powerplant.

This is exactly what is happening on the Ethereum blockchain. Two Eth (R1) is created out of thin-air and given to the miners, every 13 seconds! In other words, the Ethereum protocol is subsidizing the miners, via creation of newly minted Ether (R1). On the other hand, transactions fees (R2) only account for ~50% of the miners revenue. Transactions (R2) should not hold 100% responsibility for all emissions (C1 + C2), since transactions are only responsible for paying ~50% of mining revenue. Instead, the Ethereum protocol should also be responsible since ~50% of mining revenue was paid by the protocol. We will explain how transaction fees affect energy consumption of the Ethereum network in Part II.

Why Statement 2 (SuperRare) is misleading
SuperRare says that transactions do not directly result in higher energy consumption. While this is true in the short term, higher transaction fees (R2) does have an impact on energy consumption of the Ethereum network (C1).

Example

Hash rate share of miners – Total computational power that is being used to mine and process transactions on a PoW blockchain by a single miner. Purchasing and running more GPUs = more hash rate. Higher hash rate = more energy consumed

Network hash rate – Sum of all hash rate shares of miners, the total hash rate of a PoW network.

      1. Alice (Ethereum miner 1) spends $5 and controls 50% network hash rate.
      2. Bob (Ethereum miner 2) spends $10 and controls 50% network hash rate.
      3. They split the mining profits, a total of $25 evenly.
      4. If there are less transactions occurring on the network, then the transaction fees will decrease and miners will make less profit. If the mining profits decrease by $10, Bob will no longer make a profit since he is spending $10 to make only $15 * 50% = $7.5. So, Bob will quit being an Ethereum miner and the overall energy consumption of the Ethereum network decreases.

Transactions (R2) are not only responsible for the energy consumed by processing transactions (C2). We will explain how transaction fees affect energy consumption of the Ethereum network in Part II.

Part II – The relationship between transactions and energy consumption of a PoW blockchain

A better method to describe how transactions relate to potential CO2 is to model miners as profit-seeking agents. In this section, we discuss how the price of Ethereum and average gas price influences miner activity. (Disclaimer: Some miners may mine for a speculative future value of ETH or simply mine to secure the network at a cost. We are assuming most miners mine if they can make immediate profit. We will present empirical evidence which agrees with the conclusions made from our assumption and model.)

The network hashrate (C1) increases or decreases in response to change in mining incentives: (R1 + R2) * price of Eth

    • Miners join the network (increasing energy consumption) as mining incentive increases.
    • Miners leave the network (decreasing energy consumption) as mining incentive decreases.

Overall, at the network level, network hash rate and energy consumption is always directly correlated with the dollar value of overall revenue (R1 + R2). In upshot,

    • Network energy consumption is dominated (>99%) by network hashrate (C1).
    • The block reward of 2 ETH (R1) miners receive is fixed and does not change.
    • Energy consumption of the Ethereum network changes in response to transaction fees (R2) and the price of ETH.

Example
MH/s = mega hash per second = 1,000,000 hashes per second
GH/s = giga hash per second = 1,000,000,000 hashes per second
TH/s = tetra hash per second = 1,000,000,000,000 hashes per second

Let’s say that you (a miner) rents a RTX3080 GPU for $1 an hour (which covers hardware cost and electricity), and it gives a hash rate of ~86 MH/s (denote this as X). Suppose the overall network hash rate is Y hashes per second. Your expected revenue from mining Ethereum (or joining any mining pool) is:

Expected block revenue = (X / Y) *  (2 ETH + transaction fees per block) * price of Eth

Multiply the above by ~277 blocks per hour (13 seconds per block), we can compute the expected hourly reward:

Expected hourly revenue = 277 * (X / Y) *  (2 ETH + transaction fees per block) * price of Eth

Obviously, this mining operation is only profitable if the hourly revenue is more than hourly cost. Setting the above equation to equal to the cost ($1/hr) and solving for Y, we derive that at a network hash rate of Ythreshold = 161 TH/s, this operation breaks even (also assuming 2 Eth average transaction fees per block and $1700 per Ethereum). This means that, rationally, you would only mine Ethereum if the network hash rate is less than 161 TH/s (assuming block reward and price of Ethereum stays relatively constant).

Of course, operating GPUs cost differently depending on many factors. But the general principle remains the same. For each fixed hourly cost X, there is some break even network hash rate Ythreshold. If network hash rate ever drops below Ythreshold, miners with hourly cost X will start mining to generate profits. Moreover, the value of Ythreshold increases as mining incentives increase. For example, if the price of Ethereum doubles, our Ythreshold becomes 2*161 = 322 TH/s.

Price of Ethereum is a variable that can be easily checked, but overall transaction fees of the network is not as readily available to the end users. We will explain how a more accessible metric, gas price, is directly correlated with overall transactions fees.

Transaction fees (R2) per block is directly correlated with average gas price.

Gas: Gas is a unit of computation that is used to calculate the fee for a transaction (more complicated transactions use more gas)
Gas Limit (in ETH): For transactions: the maximum amount of gas a user is willing to use for performing a transaction. For blocks: the maximum amount of gas usage allowed in a block (sum of all the gas usage of included transactions).
Gas Price (in ETH): Cost of a single gas unit. Transactions compete with each other in gas price, as miners will process transactions that pay them more first.

    • Miners receive the sum of all transaction fees in a block as a reward.
    • The transaction fee of a transaction = gas price of a transaction * gas used for transaction
    • Currently, the Ethereum network is very saturated, meaning gas used per block is 90% of gas limit of a block.
    • When network is saturated, the transaction fees paid to miner for a block is approximately average gas price * gas limit for block.
    • ~160 gwei average gas price corresponds to ~2 Eth transaction fees.

Transaction fees are correlated with average gas price, so energy consumption of the Ethereum network changes in response to average gas price. More congestion in the Ethereum network results in higher average gas price and transaction fees, which in turn results in an increase in hash rate and CO2 emissions of the network.

Comparing Ethereum price vs. gas price in terms of their influence on mining incentives

Mining incentive = (2 ETH + transaction fees per block) * price of ETH

Current (at the time of writing) transaction fees of a block tends to be around 1ETH – 3 ETH. (https://etherscan.io/blocks)

Assuming the average gas price is 160gwei and transaction fees of block is 2 ETH, the price of Eth has double the influence on mining incentive, compared to gas price.

For example, suppose the transaction fees per block is 2 ETH. Then:
If the price of ETH goes up by 100%, mining rewards go up by 100%.
If the transaction fees of a block goes up by 100%, mining rewards go up 50%.
If the price of ETH is reduced by 50%, mining rewards are reduced by 50%.
If the transaction fees of a block is reduced by 50%, mining rewards are reduced by 25%.

    • Before May 2020, fees were low (< 1 Eth) so the change in gas price did not significantly affect mining rewards compared to change in price of ETH.
    • Now at ~2 Eth transaction fees per block, the change in gas price has around half the impact on mining rewards compared to change in Eth price.
    • In the future, if transaction fees keep increasing (significantly more than 2 Eth), then a change in gas price will have a similar impact on mining rewards compared to a change in the price of Eth.

Hypothetical scenario: What would happen if the transaction fee of the block is unbelievably high?

Let’s suppose the transaction fee per block is 8 ETH. Then:
If the price of ETH goes up by 100%, mining rewards go up by 100%
If the transaction fees of block goes up by 100%, mining rewards go up by 80%
If the price of Eth goes down by 50%, mining rewards go down by 50%
If the transaction fees of block goes down by 50%, mining rewards go down by 40%

Can the transaction fees of a block keep increasing?
Possibly, but casual users are already migrating out of Ethereum due to high transaction fees, therefore reducing the network congestion.

Comparing network hashrate with Ethereum price and transaction fees over several years.

Overall Ethereum network hashrate at a given time

Ethereum price at a given time

Total transaction fees of the Ethereum Network at a given time

https://etherscan.io/chart/transactionfee

There is a strong correlation between Ethereum price and hash rate. At the moment, transaction fees have a weaker (~50%) correlation with hash rate. Increase in the price of Eth and transaction fees causes an increase in network hash rate.

The price of Ethereum contributes most to mining incentive
Ethereum price is directly correlated with mining incentive and it currently has double the influence over mining incentive. Hence, potential CO2 emission of Ethereum is correlated with the price of Eth. 

Transaction fees contributes secondly to the mining incentive
Correlation between transaction fees and mining incentive depends on the total transaction fees of a block. Considering the average transaction fees of a block is 2ETH, we can say that today, transaction fees influence mining incentives 50% less than Ethereum price.

Summary of Part 2

    • Higher ETH price = more incentives for miners
    • Higher gas price = higher transaction fees = more incentives for miners
    • More incentives for miners = more network hash rate = more CO2 emissions
    • Influence of gas price over mining incentives depends on the price of gas per block
    • Assuming average transaction fees of a block is around 2 ETH (gas price of ~160gwei) Price of ETH has double the influence over mining incentives than gas price

PART III: NFTs and its potential CO2 emissions

How do NFTs affect potential CO2 emissions

    1. NFT related transactions take about 5-20% of the transactions on ETH today (based on a quick estimate using “% gas used” on etherscan.io/gastracker). We are in fact congesting the Eth network. The network utilization rate of NFT-related transactions is highly variable. We will update this article with better estimates when available.
    2. The main contributing factor to mining incentive however, is the price of Ethereum. Are we contributing to the appreciation of Ethereum? Yes, most likely, but the exact effect is hard to measure.
    3. NFT hype brings more people into the Ethereum network, which might influence more people buy into ETH, which then increases network congestion, ETH price and hash rate (Similar to how Elon Musk tweets about doge. More news about doge = more people buy doge = doge price increases). Obviously, we don’t think NFT artists have nearly as much influence over the Eth network in the way Musk has over doge. But we have some influence over it at least. We think the heated discussion and news around NFTs contributes more to potential CO2 emissions rather than the overall transactions we make. 

“If all NFT related transactions stopped occurring, what will happen?”

You might think that the following happens:
A defi-related transaction will queue instead. Nothing will change. Network congestion remains the same and potential CO2 emissions remains the same.

This is inaccurate. In the language of microeconomics, removing NFT-related transactions creates a downward shift in the demand curve for transactions, which results in an decrease in equilibrium gas price (the supply curve is flat here). Will this make a huge impact on potential CO2 emissions? No. But to say it will do absolutely nothing to the Ethereum network is a misleading statement.

On all the blogs scaring people with numbers

Memo Akten opened up a great discussion on the ethics of using a network that makes a significant impact on our environment. It is correct to say that NFT artists have a carbon footprint because we are involved in the entire Ethereum network ecosystem.

Figuring out the exact amount of CO2 emissions responsible from transactions is not easy. While using gas usage to come up with CO2 per transaction is straightforward, it only gives an “upper bound” and lacks accuracy, and does not give numbers that facilitate comparison.

There is no single universally-accepted method for carbon foot-printing Scope 3 emissions (Carbon footprint: current methods of estimation, ‘Carbon footprinting’: towards a universally accepted definition, EPA’s Guideline, calculating scope 3 emissions, Technical guidance for calculating scope 3 emissions). Applying intuitive methods (that could be correct elsewhere) here is particularly problematic, as mining is only ~50% funded by users doing transactions (purchasing gas). This is in stark contrast with traditional manufacturing where most revenue comes from selling produced products. This is further complicated by the fact that the transaction market is a highly-competitive auction market with a fixed supply of ~12 million gas units per block. There does not seem to be any universally-accepted method for calculating the responsibility of an individual transaction here.

The fundamental issue is that accounting indirect responsibility (carbon footprint) is philosophical in nature. To illustrate the issue, we will leave those interested with the following philosophical question:

“A company produces two units of a product P, costing $10 and emitting 10kg of CO2. Alice donates $6 to the company. Bob pays $5 for 1 unit of product P and Carol is gifted 1 unit of product P. What is the carbon footprint of each party?”

If a philosopher or economist has an answer, please let us know. We first need to determine the relevant ethical frameworks to account responsibility before we can make claims about the exact carbon footprint of a transaction.

Conclusion

    • Any blog dividing overall CO2 emission of Ethereum by the number of transactions made or total gas used is using an over-simplified model and probably over-accounts the responsibility  by a significant amount. However, it does give a good “upper bound” for carbon footprint.
    • Any blog that claims NFT artists have absolutely zero impact on the Ethereum network are also misleading. Higher gas prices means more incentives for miners, which increases network hash rate and raises overall potential CO2 emissions. By simply discussing NFTs in any context, we are directing attention to the Ethereum network, which is influencing the price of Ethereum and the congestion of the network. 
    • Energy consumption of the Ethereum Network is affected by changes in Ethereum price and average gas price. The impact of individual transactions on the environment is non-zero, and unfortunately hard to quantify.

Some solution(s) to lowering energy consumption of the Ethereum network

    1. A temporary solution is to lower the mining incentive so there are less miners mining.

      EIP1559 aims to lower mining incentive and aims to make Ethereum more user friendly, but Miners are actively going against this because it undermines their mining income.

    2. A more long-term solution is the Eth 2.0 upgrade, which removes the computational lottery completely and should decrease total network energy consumption by >99%.

“What can I do as an NFT artist to lower my environmental impact?”

    • Mint on a PoS network 
    • Mint less frequently
    • Mint at times when network is less congested (when gas is low)