On the cost of Bitcoin Transactions

I recently posted a tweet on the cost of Bitcoin transactions, and my numbers were disputed. They were based on things I heard and read and some quick guestimations, so I took a little time to make more detailed calculations based on first hand data for everyone to check. It turned out that my first estimates were in the right ballpark.

The point I wanted to make in my tweet was that Bitcoin is quite different from what I thought: It is not about paying bills or replacing credit cards or bank accounts. Bitcoin is more like a system of safes filled with digital gold, and it comes with many of the same problems that stem from storing your wealth as gold in safes: Gold mining is not exactly environment friendly, the value fluctuates, it is not of much use other than storing value and you have to guard it. One the positive side it is not bad to have a lot of it.

Now let’s look at some numbers regarding the underlying technology and the total costs of the operation of the Bitcoin network. You will see why Bitcoins are not about payment any more.

Transactions and Hashes

Stats on the hash rate of the Bitcoin network can be found here: https://blockchain.info/charts/hash-rate.

For our further calculations we use a hash-rate of about 11 Mio. TeraHashes/second.

Stats on the daily transactions are here: https://blockchain.info/charts/n-transactions

For the number of daily transactions we assume 300.000 transactions per day or about 3.5 per second.

This means that for one transaction the bitcoin network currently computes about 3 Mio. TeraHashes (TH).

Transaction Fees

Stats on daily average transaction fees can be found here. Historically the idea of Bitcoins was to be used for all kinds of payments, even micropayments, but with average transaction fees in the order of $5 it is currently one of the most expensive payment or money transfer methods, about as expensive as Western Union. So why did the transaction fees get so high?

A contributing factor surely is that the Bitcoin value in USD got high, but looking at various charts it does not seem to be the main factor. I could not find any single factor that determines the transaction fees; they seem to form an independent market in itself, neither directly related to transaction volume, Bitcoin value or any other stats I looked at.

The fact is that transaction fees are already way too high for micropayments, and it can be assumed that transaction fees will increase further when mining will get more difficult, but the economic relationship between mining yields and transaction fees is  complex and not well understood. It seems obvious though that a long time before the last coin will be mined in the year 2140 the system needs to support itself by fees.

However, in the long run we can expect that Bitcoin will transform itself into something very different from today, so I do not dare to predict how transaction fees will actually develop when the network undergoes major changes. In it’s current form it seems to be more like a vault than a payment system, and people put money in mainly because they believe they will get more out than they put in. So far they have been right.

Another strange fact is that the transaction volume in Bitcoins did not really change that much over the last year.

This might be attributed to the increased value, but looking at the graphs I see no correlation that supports this assumption.

ASIC-Miner Efficiency

The most efficient ASIC-Miner I am aware of in Nov. 2017 is the Antminer S9, delivering 13.5 TH/second while consuming 1323 Watts of electrical power, with a list price of  $1415 plus $105 for the power supply. This price is for a delivery in about 3 months though, and if you want one right now you find them on E-Bay for $3000 – $4000. Funfact is that bitmain, the producer of these devices also maintains the worldwide largest mining pool Antpool, accounting for over 20% of the worldwide hash power alone.

Assuming everyone in the world would be using this most efficient miner, the electrical energy to compute the 3 Mio TeraHashes for one transaction would amount to about 300 MegaJoules or 83 kWh.

Cost of Electricity

Prices for electricity the miners actually pay in China are not public, but the average price is about 8 cent/kWh, which is one of the cheapest in the world, but this can’t be the main reason why China currently amounts for over 70% of the Bitcoin hash rate. The USA has even cheaper electricity in some places but runs only 1% of the hash rate, so my guess is that it is superior entrepreneurial spirit and technological leadership why China rules the Bitcoin network. You find more information about mining in China here.

When we assume a wholesale price for industrial electrical power of 5ct/kWh, we get a lower bound on the energy costs for one Bitcoin transaction of about $4 if everyone runs an Antminer S9. In reality, not everyone is running these most efficient miners, and the next most efficient rigs consume 2-3 times of this energy.

5ct/kWh also seems to be a realistic global low bound; even in Iceland the industrial energy prices are over 7 Cents/kWh, and except in some places in the US this is as cheap as you can get in the western world. Also, if you would take into account the purchasing power parity conversion factor, then the adjusted value for China electricity price would be over 10 US-Cents adjusted for the price of other goods and services in both countries.

All this means that the actual energy price of a bit coin transaction can be considered a multiple of $4, and the factor might be as high as eight (2 x PPP adjustment, 2-3 x less overall efficiency, 1.5 actual average energy price), possibly bringing the energy costs per transactions up to $32. It is safe to say that the real energy costs per transaction are somewhere between $4 and $32. Even at $4 per transaction the bitcoin network currently needs electricity worth about half a billion US-Dollars for one year to operate.

Depreciation Cost

Another cost factor that can be estimated is the depreciation of the mining hardware. It is safe to assume that in order to make profit you can not run miners older than a year because next years hardware will outperform last years rigs so much that you will run them at a loss if you don’t update. This means that you can write off your investment in one year unless you find stupid people who will buy your practically worthless last year’s hardware.

Within a year, the Antminer S9 will facilitate about 150 transactions, which means that the depreciation costs of one transaction are about $10. For other less efficient miners the price is going to be higher.

One Antminer S9 will consume about 12400 kWh electricity per year, worth $620 at 5 cents per kWh, so with such low energy prices the depreciation costs are even higher than the energy costs. However, we only know the retail price for the Antminer S9, the productions costs may be higher or lower, depending on whether bitmain makes money by „test running“ them for a while before they delivering them to their customers.

Other Costs

We also so far ignored the costs for manpower, cooling, wiring, network operation and the costs for the end-users performing transactions. Especially the per-transactions costs for casual end users can be substantial when doing very few transactions per year if you want to do it right and safe. It can be also quite cheap if you run it with free software on a device you already own and you have an appropriate backup policy anyway. It would be hard to put a number on these costs though, but it is way more effort compared to using a credit card.


Bitcoin is a very strange universe, and it might not be entirely fair to base all the above calculation on the number of transactions, but that was actually the point: Bitcoin isn’t about transactions.

Currently the bitcoin network is headed away from making transactions cheap and easy. It seems to be headed towards being a vault where money is mainly accumulated and moved around less frequently. To some degree that might make sense in itself. However, given the number of „ifs“ that all pose risks to the values stored there I have serious doubts that it can fulfill this role in the long run.

But what do I know.

, ,