How much would it cost to do a 51% attack

Question

how much would it cost to carry out an 51%-Attack? If there is a place service in the web that you can rent ASIC's than i would rather hear what renting this would cost. If not then buying the asics is fine too.

To keep the calculation KiSS following assumptions:

1. Attacker chooses the cheapest way of renting/buying hardware. But he is NOT renting/bribing whatever the existing machines from miners from the existing bitcoin network. The gear has to come from outside the network.
2. Don't worry about that renting/buying such a huge qt would drive the price for the hardware up.
3. Ignore what the community would do, the cored devs or whoever to counterstrike your attack.
4. you don't need to include electricity cost in the calculation. If you do include it, please tell what you took for US-Cents/kw/h

Thank you!

P.S. is there a website that does this calculations on a regular basis? Would be cool to check this from time to time..

Update 4 Years later ond June 2019: The Hashrate of the Entire Network is right now (according to Bitinfocharts.com): 55,42 ExaHashes. Most efficient Miner is the DragonMint which has exactly (and realisticaly) 16 TH/s and costs including with power supply 2844USD. One would need 55 420 000 TH/s /16 TH/s =3,46 millin Dragon Mints * ~ 2800USD (included a small rebate)~ 9,8 billion USD. Crazy how the cost increased in the last 4 years from 145 Million to almost 10 billion...

EDIT X Years old answers to this question don't matter at all. This question needs to answered every 6 months. Entirely new generations of Asic-miners, change picture a LOT

Answer 1

There is a handy chart at https://en.bitcoin.it/wiki/Mining_hardware_comparison.

If we assume that it is complete, and that the figures listed there are accurate (I have not verified them), and that the attacker must buy the hardware, then the most economical device is the Antminer S5+, which produces 7.7 TH/s (7.7e12 H/s) and costs US$2300. It draws 3.4 kW.

The current network difficulty is about 5.6e10. This means that an average of 5.6e10 x 2^32 = 2.4e20 hashes are needed to mine one block. Assuming that one block is mined each 10 minutes (600 seconds) on average, the network as a whole is producing 2.4e20 / 600 = 4.0e17 H/s.

So a 51% attacker needs to produce about 4.0e17 H/s himself, to match or slightly exceed the hashing power of the existing network. This means he needs 4.0e17 / 7.7e12 = 52000 Antminer S5+ devices, which would cost 52000 x $2300 = $120 million.

(This is unrealistic in a couple of ways - buying that many devices would not only drive up the price but probably exceed the production capacity. On the other hand, such a large order might qualify for a quantity discount or wholesale pricing. This attacker might also find it more efficient to license the design and do the manufacturing themselves.)

Running these machines would consume 52000 x 3.4 kW = 177 MW of power. If the attacker is located in the US, this table shows that the cheapest place to get electricity at an industrial rate is Washington State, where it costs an average of 4.5 US cents per kWh. At this rate, the attack would cost US$8000 per hour to supply with electricity.

If they are all to be located in one place, there will be additional costs to build / rent and operate a facility to house them all, with appropriate electrical infrastructure. There will also need to be significant network and server infrastructure to connect them all to the Bitcoin network. Cooling costs will also be substantial - Google says that their data centers have electrical overhead of about 12% (i.e., on top of the electricity used by the equipment itself, an additional 12% is used for cooling, power supply losses, etc). They say that the industry average is more like 70%. You would also have to hire an IT staff to maintain everything.

This overhead might be reduced if the devices were to be distributed (e.g. each of 52000 people keeps one in their house) but then you start to incur administrative costs to keep all those people organized. (Just picture the communications department to tell them what to do, accounting to make sure they all get their cut of the proceeds from this evil scheme, human resources to recruit new participants, etc.)

Answer 2

The cost consists from two parts.

1. Cost of hardware
2. Cost of electricity during attack

The market cost of miner which is not able to mine with profit is... zero. Because it is useless. The problem is only to collect such devices and turn them on for an hour or two.

We will see a lot of obsolete hardware for a very cheap price right after the halving the award.

Answer 3

An update as of Feb 2018.

The hashing power of the network is currently 24M TH/S. The most powerful Antminer costs $1800 and can generate 14TH/S.

You'd need to 24M/14 = 1.7 million of these to control ~50% of the bitcoin network. That would cost ~$3B USD.

Each miner consumes 1.4 kw per hour. The average cost of a kilowatt hour in the U.S. is 12 cents.

So to run all of these miners would cost 12 cents * 1.4 kw/h * 1.7M antminers = $285,000/hour.

So that's $3B up front for hardware plus $285,000/hour to control the network.

Granted that leaves out a lot of practical considerations like how one could obtain 1.7 million ant miners, how much work it would take to get those all up and running, where you'd get that much electricity, etc.

This also is considering the cost of launching an attack for an entity that currently controls 0% of the existing mining pool.