Bitcoin is back in the spotlight these days thanks to some wild price movements and central bank meetings. The decentralized currency has recently been trading over its all-time high of $1200 on some exchanges. But the higher the price goes, the more it exacerbates bitcoin's dark side: shocking levels of electricity consumption. In 2015, I wrote that bitcoin had a big sustainability problem. Back then, each bitcoin transaction represented roughly enough electricity to power 1.57 American households for a day— approximately 5,000 times more energy-intensive than a credit card transaction. Since it's been two years, it's time for an update.
First, a caveat: it's impossible to know precisely how much electricity any given bitcoin transaction "consumes," but it's simple enough to estimate a plausible range of energy consumption for overall bitcoin mining. Mining secures transactions on the blockchain, a giant ledger of all completed transactions.
It's worth looking at estimates for a per-transaction energy cost because we can compare that cost to existing payment systems. It's also a more tangible way to represent value-for-electricity. Simply knowing that total bitcoin mining consumes x amount of energy is interesting, but it's better to discuss how many transactions we're actually getting for all that electricity spent.
Updated calculations with optimistic assumptions show that in a best-case hypothetical, each bitcoin transaction is backed by approximately 90 percent of an American household's daily average electricity consumption. So even though that's still about 3,994 times as energy-intensive as a credit card transaction, things could be getting better since 2015.
Unfortunately, it's more likely that things are getting worse. A new index has recently modeled potential energy costs per transaction as high as 94 kWh, or enough electricity to power 3.17 households for a day. To put it another way, that's almost enough energy to fully charge the battery of a Tesla Model S P100D, the world's quickest production car, and drive it over 300 miles.
How could one bitcoin transaction possibly use this much electricity? First, a bit of context about mining.
Bitcoin transactions are secured by computer "miners," which currently compete for a reward of brand-new bitcoins from the network (the block reward). The more computation power you deploy, the better your chance of getting the reward. So it's always rational for someone, somewhere, to add more computing power as long as the bitcoin sale price supports the capital and power costs.
As a general rule, if the price of bitcoin goes up, it becomes more economical to mine, no matter the efficiency of your equipment. Energy consumption should logically increase if the price goes up enough, despite mitigating factors.
The bitcoin network regularly increases the difficulty of mining to account for more mining capacity, so the current situation is akin to an arms race: miners must always add more (or more efficient) mining chips to their operations to compete with other miners for limited rewards. And miners are continuously becoming more efficient, doing more computation for less electricity.
But according to a paper from Adam Hayes at the New School, "this [difficulty] mechanism tends to counteract the downward [price] tendency caused by increasing energy efficiency" of mining equipment. So even as mining equipment improves, bitcoin's code itself supports the incentive to add more mining, requiring more electricity.
With that context in place, we can find a baseline estimate for bitcoin's energy usage overall, as well as per transaction. First, we need a few basic data points:
- Bitcoin can handle a theoretical maximum of about 7 transactions per second as it's presently implemented. The average daily number of bitcoin transactions was 302,150 as of March 1, according to blockchain.info.
- Next, we'll take a look at the bitcoin hashrate, which measures the actual computation power of the network. I use a weekly average to smooth out daily noise. As of March 1 this figure was 3.387 million terahashes/s).
- We also know the power consumption of the most efficient miners on the market. Antminer and BitFury seem to be tied for the lead here, which is about .098 W/GH/s for an Antminer S9 (BitFury does have a 2x more efficient liquid-cooled datacenter, but this likely doesn't represent most mining).
To find a lower bound for total network energy consumption, multiply the miner power consumption per hash per second by the global hashrate per second. As of March 1, that made about 332 megawatts of constant draw. I don't think this sounds unreasonable, given that a bitcoin entrepreneur just announced plans for a new 130 megawatt mining facility in China.
Now, let's express that in a more relatable measure. If all bitcoin miners were running very efficient hardware, bitcoin would be consuming enough power to supply the daily needs of about 268,990 average American homes.
Since the weekly rolling average number of daily transactions was 302,150, each bitcoin transaction represented at least 26 kWh of electricity spent mining, or enough electricity to power 0.89 average American households for a day.
In 2015, that figure was 1.57. Miners have certainly become more efficient since then, performing much more computation for less power, and there are many more transactions per day. So, bitcoin is more sustainable now, right? Not necessarily.
For one thing, not all bitcoin miners are running the most efficient machines. In a post criticizing my 2015 estimate, engineer Marc Bevand assumed a global figure of .15 J/GH, which would add an instant ~50% to the per-transaction electricity consumption—up to 1.34 households' worth of daily electricity use. Nobody knows the true overall efficiency figure, and .15J/GH might be out of date, so I'll stick with more efficient equipment at 0.098 J/GH for my example.
More importantly, most bitcoin experts will tell you that the overall electricity consumption of bitcoin mining isn't mainly determined by miner efficiency. As Ittay Eyal, a Cornell University computer scientist and assistant director of the Initiative for Cryptocurrencies and Contracts, told me via email, "bitcoin's energy usage is a function of the currency's exchange rate and the block reward." In other words, if the price goes up enough, miners will add whatever computing power they can afford to profitably capture mining rewards.
Enter the Bitcoin Electricity Consumption Index (BECI), a real-time modelling tool from Alex de Vries of Digiconomist. The BECI gathers daily figures and uses a reasonable set of assumptions to estimate the economically viable amount of electricity that the bitcoin network could consume while still covering miners' electricity costs.
Most of the data collection for Digiconomist's index is automated, and the output figures are recalculated daily. As the price drops or rises, so should the estimated power consumption of the bitcoin network.
I asked de Vries what sets his estimates apart from the kind of simple calculations I've done above. He noted that while an estimate like mine is useful as a lower bound, economic incentives will drive miners to eventually approach a break-even point where the bitcoin price equals the marginal cost to produce an additional bitcoin. This is in line with Hayes's 2015 paper, "A cost of production model for bitcoin." So his model may not always predict the current energy consumption, but in the long run it should provide a decent average.
"At worst (during very big price increases) my index is predicting energy consumption of the
Bitcoin network two months from now," after additional mining machines have either been delivered or returned into service, he wrote.
To wit, De Vries noted that my optimistic power consumption figures were likely too low because some older, less efficient mining equipment could turn a profit today, especially with access to cheap electricity.
"People have bought machines like the Antminer S5+ (at 0.44 W/GH/s still close to being profitable at 6 cents per KW/h) and many more before the newest became available, so there's a need to account for these and the constant stage of transition (rat race for more efficiency) the network is in," he said. "An economic measure then makes a lot more sense."
Since he first published the model, De Vries has adjusted its parameters to provide more generous assumptions. His figures, he says, now account for things like barriers to entry in bitcoin mining, capital expenditures, lag times, and variable electric costs. Even so, the BECI's average electricity cost of a single bitcoin transaction currently sits at 94 kWh, over three times my optimistic assumption of 26 kWh.
Critics could likely still find a few bones to pick with De Vries's index, but its fundamental value doesn't come from perfect, impossible accuracy. Rather, it gives us a ballpark estimate of bitcoin's energy consumption according to basic economic theory.
Between my optimistic lower-bound estimate, and the BECI, we're still left with a staggering amount of electricity embodied in each bitcoin transaction—anywhere from 26 to 100+ kWh, or enough to power 0.9 to 3.6 US households for a day. To repeat, this is thousands of times more energy-intensive than an estimate for a credit card transaction.
Even if 100 percent of bitcoin mining was powered by wind and solar sources, the fact remains that this activity is displacing other uses of electricity that might be more energy-efficient. With a lot less electricity, a Visa datacenter can power thousands of times more transactions per second. Are bitcoin transactions really powering economic activity thousands of times more valuable? I doubt it, and I say this as no great friend of Visa.
Critics could even argue that a number of bitcoin transactions are facilitating unproductive activity. For darkweb weapons-dealing (nasty but uncommon), ransomware payments (a serious problem), and very high-fee transactions, the value-for-electricity may be downright negative.
There may be hope for the future, however. It's important to keep in mind that while bitcoin transactions are secured by mining, the total amount of mining isn't related to the number of possible transactions. So projects like Teechan show initial promise as a way to massively increase the number of bitcoin transactions per second, as do "payment channels" like the Lightning Network. This would make bitcoin able to perform more useful transactions without requiring additional electricity, which is quantifiably good. Both of these, though, involve relinquishing some of the transparency central to bitcoin.
Whether scaling improvements will be adopted by the majority of users also remains to be seen. Bitcoin's lack of centralized governance is a feature, but it's also a bug. Ideological blocksize battles continue and diverging opinions have split the userbase into various interest groups.
Another environmental positive is the ever-declining block reward. Bitcoin's code dictates that miners will eventually earn more from transaction fees than they do from the regular creation of new bitcoins (supply is capped at 21M BTC). It's doubtful that a future fee market would support the ludicrous energy costs per transaction we see today.
In absolute terms, bitcoin's electricity consumption is still small potatoes: De Vries' index estimates it at 0.05 percent of world energy consumption as of March 1, or about as much as the country of Paraguay. Generously, it could even be a third of that figure.
But the estimated energy cost per transaction, at any end of the range, is still staggering by modern standards. In a future where carbon prices and climate change are almost guaranteed, that's something that should greatly concern bitcoin users, programmers, and advocates.