Tuesday, October 5, 2021

Cryptocurrency's Carbon Footprint Underestimated

Back in April I wrote Cryptocurrency's Carbon Footprint about the catastrophic carbon emissions of Proof-of-Work cryptocurrencies such as Bitcoin. It now turns out that I didn't know that half of it; the numbers I and everyone else has been using are greatly underestimated. Below the fold, based on my no doubt somewhat inadequate methodology, the real story.

The leading source of data on which to base Bitcoin's carbon footprint is the Cambridge Bitcoin Energy Consumption Index. As I write, after the Bitcoin price took at 10% hit on news from the People's Bank of China, they estimate 12.94GW or 0.45% of total global electricity consumption, with theoretical bounds of [4.65,31.17]GW.

Their methodology is explained here. Greatly oversimplified, it is this:
  • The hash rate of the Bitcoin blockchain is known (see graph).
  • The energy consumed per hash by the various mining ASICs is known.
  • The population of each type of mining ASIC is not known, and changes all the time, so:
    The lower bound estimate corresponds to the theoretical minimum total electricity expenditure based on the best-case assumption that all miners always use the most energy-efficient equipment available on the market. The upper bound estimate specifies the theoretical maximum total electricity expenditure based on the worst-case assumption that all miners always use the least energy-efficient hardware available on the market, as long as running the equipment is still profitable in electricity terms. The best-guess estimate is based on the more realistic assumption that miners use a basket of profitable hardware rather than a single model.
  • Mining ASICs are profitable if the mining rewards they generate more than cover the electricity cost.
  • The electricity cost varies among miners, but:
    Electricity prices available to miners vary significantly from one region to another for a variety of reasons. We assume that on average, miners face a constant electricity price of 5 USD cents per kilowatt-hour (0.05 USD/kWh). This default value is based on in-depth conversations with miners worldwide and is consistent with estimates used in previous research
The energy consumption estimates can be converted to carbon footprints using the known carbon intensity of electricity production in the regions where Bitcoin is mined. CBECI also produces estimates of the proportion of hash rate in different regions although, unlike their energy consumption estimates, these lag considerably. As you can see from their most recent map, they do not yet show the exodus of miners from China due to the government's continuing cryptocurrency crackdown.

Using the average carbon intensity of electricity in each region is, of course, a simplification. There is a tidal wave of Bitcoin greenwashing, such as:
  • Suggesting that because Bitcoin mining allows an obsolete, uncompetitive coal-burning plant near St. Louis to continue burning coal it is somehow good for the environment.
  • Committing to 100% renewable energy use by signing the Crypto Climate Accord, then implementing it:
    By buying 15 megawatts of coal-fired power from the Navajo Nation! They're paying less than a tenth what other Navajo pay for their power — and 14,000 Navajo don't have any access to electricity. The local Navajo are not happy.
But the desperate greenwashing suggests it is likely that in most regions Bitcoin mining uses electricity more carbon-rich than the average.

There were already many reasons why the CBECI energy estimates are too low. They exclude, for example, the energy consumption of the computers driving the ASICs, the cooling needed to stop them overheating, and the networking to tie them together. These are relatively small corrections, but there is another that is rather large.

A full accounting of the carbon footprint of IT equipment such as Bitcoin mining rigs requires estimating not just the emissions from its use, but also the embedded emissions produced during its manufacture and disposal. Fortunately, a paper from a year ago entitled Chasing Carbon: The Elusive Environmental Footprint of Computing by Adit Gupta et al from Harvard, Facebook and Arizona State provides these estimates for both mobile devices and data center equipment, which is a reasonable proxy for Bitcoin mining rigs:
we split emissions into Scope 1 (blue), Scope 2 (green), and Scope 3 (red). Recall that Scope 1 (opex) emissions come from facility use of refrigerants, natural gas, and diesel; Scope 2 (opex) emissions come from purchased electricity; and Scope 3 (capex) emissions come from the supply chain, including employee travel, construction, and hardware manufacturing
Figure 11
They plot the history of these emissions in Figure 11, which:
illustrates the carbon footprint of Google and Facebook over six years. Although the figure divides these emissions into Scope 1, Scope 2, and Scope 3, Scope 2 comprises two types: location based and market based. Location-based emissions assume the local electricity grid produces the energy — often through a mix of brown (i.e., coal and gas) and green sources. Market-based emissions reflect energy that companies have purposefully chosen or contracted — typically solar, hydroelectric, wind, and other renewable sources.
Thus the green line on the graph is "opex" (i.e. electricity) corresponding to the CBECI estimates, and the red line is "capex", i.e. embedded emissions. Note two things:
  • Until Facebook and Google switched to renewable electricity, opex and capex emissions were approximately equal. The fraction of renewable energy in Bitcoin mining is unlikely to be large.
  • In 2017 Facebook and Google changed their hardware footprint disclosure practice, resulting in an increase of 7x for Google and 12x for Facebook. It is safe to assume that neither would have done this had they believed the new practice greatly over-estimated the footprint.


In 2018 Christian Stoll et al estimated Bitcoin's carbon emissions:
We determine the annual electricity consumption of Bitcoin, as of November 2018, to be 48.2 TWh, and estimate that annual carbon emissions range from 21.5 to 53.6 MtCO2.
Adjusting this to the current CBECI estimate gives a range of about 43 to 108 MtCO2/yr for Bitcoin's opex emissions. Using data from Wikipedia, this is between Cameroon (~42) and Uzbekistan (~108).

A Factor Of 2?

If we assume that Bitcoin mining is represented by Facebook and Google before they switched to renewables, then the embedded Scope 3 emissions are around the same as those based on the CBECI numbers, and the carbon footprint of Bitcoin is double previous estimates. A carbon footprint of 86 to 216 MtCO2/yr would place Bitcoin mining between Tanzania (~76) and Pakistan (~215).

Note that this would mean that even were Bitcoin using 100% renewable energy, it would still have a carbon footprint between 43 and 108 MtCO2/yr.

A Factor of 10?

If we assume that the newer hardware footprint disclosures increase the embedded emissions by between 7x and 12x, say 9x, including the embedded emissions in mining rigs increases the emissions computed from the CBECI estimates by a factor of 10 (1 for electricity and 9 for embedded emissions). Thus applying the 9x factor for capex emissions gives a range for the total carbon footprint of 430 to 1080 MtCO2/yr. Or, about the same as South Africa (~440) to about the same as Russia (~1050).

A Factor of 19?

That is bad enough, but it isn't the end of the problem. Data center equipment is run in carefully controlled environments at carefully controlled duty cycles. Because of the phenomena outlined by Dean and Barroso in The Tail at Scale, even batch processing cloud infrastructure aims for a duty cycle around 80%, whereas interactive service infrastructure aims between 30% and 50%. Bitcoin mining rigs, on the other hand, are run at 100% duty cycle in sub-optimal environments.

Both factors impair their useful life, but a bigger factor is that the highly competitive market for mining ASICs means that they have only a short economic life. This month's Bitcoin's growing e-waste problem by Alex de Vries and Christian Stoll estimates that:
The average time to become unprofitable sums up to less than 1.29 years. While this concerns an unweighted average, we can refer to the case study on Bitmain's Antminer S9 ... to show that weighting the average lifetime by sales volume does not significantly change the results.
de Vries and Stoll estimate that:
  • Bitcoin's annual e-waste generation adds up to 30.7 metric kilotons as of May 2021.
  • This level is comparable to the small IT equipment waste produced by a country such as the Netherlands.
  • On average Bitcoin generates 272 g of e-waste per transaction processed on the blockchain.
  • Bitcoin could produce up to 64.4 metric kilotons of e-waste at peak Bitcoin price levels seen in early 2021.
  • The soaring demand for mining hardware may disrupt global semiconductor supply chains.
David Gerard points out that this means:
That's half an iPad of e-waste average per transaction.
Because a mining rig's life is perhaps only half that of data center equipment, it means that the embedded emissions in mining rigs are amortized over only half as long, and are thus in effect twice as great. Thus it is plausible that the multiplier for the embedded emissions is not 9 but 18, leading to Bitcoin having a carbon footprint between 817 and 2052 MtCO2/yr. This compares to between Brazil (~812) and India (~2400).

Note that this would mean that even were Bitcoin using 100% renewable energy, and we ignore the updated Scope 3 disclosure, it would still have a carbon footprint between 86 and 216 MtCO2/yr.

It Gets Worse

  • The numbers above are for Bitcoin alone, but it is far from the only Proof-of-Work cryptocurrency. In 2018, Tim Swanson estimated that the next four Proof-of-Work cryptocurrencies added an extra 35% to Bitcoin's electricity consumption. This could mean that in the worst case the top 5 cryptocurrencies had a carbon footprint of between 1100 and 2770 MtCO2/yr, or between Japan (~1074) and the EU (~2637).
  • Gupta et al add a caveat:
    Note that because industry disclosure practices are evolving, publicly reported Scope 3 carbon output should be interpreted as a lower bound.


These later estimates seem excessive. Please critique my methodology in the comments.

To be conservative, lets assume the value for Scope 3 emissions before the disclosure change, i.e. that the Scope 3 and 2 emissions were about the same, and simply adjust the Scope 3 emissions for the short working life of mining rigs. Thus the Scope 3 emissions are double the Scope 2 emissions, and the factor by which the emissions exceed the previous estimates is 3. This leads to a carbon footprint of between 65 and 161 MtCO2/yr for Bitcoin alone, or between Morocco (~65) and Columbia (~162). The top 5 cryptocurrencies would be between Kuwait (~89) and Kazakhstan (~217).


Keith said...

You wrote:
"There were already many reasons why the CBECI energy estimates are too low. They exclude, for example, the energy consumption of the computers driving the ASICs, the cooling needed to stop them overheating, and the networking to tie them together. These are relatively small corrections"

The CBECI does actually purport to account for this with their Power Usage Effectiveness (PUE) parameter. You're right it's an ultimately trifling factor, with them estimating 20% overhead at the high end.

David. said...

Thanks for the correction, Keith!

David. said...

Olivia Solon's Bitcoin miners align with fossil fuel firms, alarming environmentalists is indeed alarming:

"What happened at Scrubgrass highlights a growing trend within the crypto world that alarms some environmentalists. Bitcoin mining is breathing new life into America’s aging fossil fuel power plants, creating a demand environmentalists say discourages investment in renewable energy sources at a time when shifting away from carbon-emitting sources of energy is essential."