Considering a bitcoin mining hardware purchase? Make a more informed decision with our Bitcoin Mining Dashboard.
In early 2013 bitcoin mining saw the start of the application specific integrated circuit (ASIC) revolution. ASICs are customized components that perform the calculations necessary for bitcoin mining significantly faster than any hardware that came before it. To anyone following the exponential growth that has occurred over the last six months it is clear that this story is nowhere near finished.
There are two factors driving ASIC adoption: ASIC chip design, which is limited by technical expertise, and production capacity, which is limited by funding and relationships with manufacturers. The first generation of ASIC chips – produced by ASICMiner (AM), Avalon, and Butterfly Labs (BFL) – were designed in-house and, with the exception of AM, have been fraught with production delays.
ASICs are stamped out using customized dies, where smaller nodes – measured in nanometers (nm) – allow for a more powerful chip. It allows for denser designs, meaning more speed and better power efficiency, but comes at the cost of design complexity and production cost. AM and Avalon chose to use the 10-year old 110nm process while BFL used the 5-year old 65nm process. Only AM was able to produce ASICs without significant delay, while Avalon has been having continuous issues with their manufacturer and BFL struggled with design problems for almost a year.
The newest crop of ASIC companies have focused on the more modern 28nm process. To reduce risks they have partnered with ASIC manufacturers to design and produce the chips, or brought significant industry knowledge and experience. KNC miner has partnered with ORSoC, HashFast has partnered with Uniquify, and Cointerra has brought significant industry knowledge from Samsung, to name a few.
The most significant costs in ASIC production revolve around node design and the creation of screens and dies that stamp out the chips. All of these are fixed costs that occur before production, meaning that the per-unit variable cost is insignificant compared to the upfront fixed costs. Companies have funded the first batch through pre-orders or IPOs, pushing much of the risk on to customers.
The 28nm chips due at the end of this year are one die shrink away from 22nm, the smallest ASIC commonly available. Die shrinks allow for magnitudes of improvement, so it is likely within a year massive increases in chip design will slow down and competitive advantage will come from efficiencies of design and production, rather than vastly superior hardware. The ability to design better nodes, establish relationships with manufactures, and optimze supply chain logistics will dominate over the first-to-market benefits currently driving the mining landscape.
As design efficiency approaches Moore’s Law – doubling in power every every 18 months, instead of every month as has occurred recently – production capacity and operating efficiency will drive profit margins. The natural evolution of this will be large data centers that can take advantage of economies of scale.
A decentralized mining network can be extremely inefficient. It requires thousands of people learning how to setup and maintain their miners and thousands of people checking on their miners regularly to make sure they’re still running, then spending time fixing them when they’re not. Many early adopters are enthusiasts who enjoy learning what each component does and optimizing settings to squeeze out as much speed as possible, but this can only go so far.
Of course it could be Apple-ified, simplifying the process to be as plug-and-play as possible, but with that comes higher margins on products and the associated inefficiency. Contrast that with a centralized system where those same thousands of mining units can be housed in a single location and maintained by a single team. The marginal time spent per unit to learn the specialized knowledge required for mining decreases, the ability to monitor and quickly respond to downtimes increases, and no money is wasted on sleek cases and intuitive user interfaces. We’ve seen the first steps of this with Alydian and KNC both offering hosting options for mining.
A favorite expression of many ASIC hardware detractors is “in any gold rush the ones making money are the ones selling shovels.” While this has largely been true so far, much of the cause for pre-selling equipment is due to the millions of dollars in upfront costs that an aspiring ASIC mining company faces. Mining companies are becoming better funded on a daily basis and eventually they will not require pre-orders. AM has proven that companies can be profitable mining on their own, with a $150M market cap and a 30% annualized yield.
The large catch here is that companies can’t mine all bitcoins on their own without jeopardizing the security of the network. As a single entity approaches 50% of the total network, bitcoin runs the risk of double spend attacks, among other significant security risks. Because of this risk mining companies will have to limit themselves to a fraction of the network, but an acceptable value of this limit has yet to be determined.
The question then arises of what to do with extra capacity. How can a mining company take advantage of their core competencies, provide potential customers with maximum value, and remain agile enough to respond to rapid changes network speed? The natural solution is to rent out extra capacity, much the way cloud servers operate for Amazon Web Services. A user could choose to rent a defined capacity for set period of time, say 1 TH/s for 1 month. This has several advantages over purchasing mining equipment and running it locally:
In order to absolve the hosting company from ownership of its customers’ mining, it will have to relinquish full control of the hashing power to be directed towards whichever mining pool the customer desires. There is an argument that this is not really reducing their total hashing power since they could commandeer it and use it as they please, however, but there are several counterpoints to this.
There is precedent that ASIC companies have had significant potential power of the network without apparent security risk. Yifu Guo, CEO of Avalon, mentioned at the Bitcoin Conference held in San Jose that he held a tray of ASICs in his hand that contained the entire network speed, and AM has mined 30% of the network while holding stockpiles of ASIC miners to sell to waiting customers. The truth is, if a well funded ASIC mining company decided to attack the network, their main deterrent would be the prospect of their business collapsing from the immediate devaluation of bitcoin.
While ASICs will likely be commoditized quickly, other cloud-based markets indicate that a large player or two will likely control a significant portion of the market. Infrastructure-as-a-Service (IaaS) is a well developed market and dominated by Amazon Web Services. Despite the competitive and commoditized industry, they have continued to control 36% of the market share as of the end of 2012. Given that bitcoin mining is not likely to get as mature as that market soon, it would not be surprising to see major players maintain significant portions of the network for sustained periods.
There are three potential options for optimizing the location of the cloud mining data center: establish a facility near the production site, near cheap power, or near talented personnel. The best decision depends on the businesses’ unique advantages over the competition and all models could provide a competitive edge.
Locating near an ASIC chip foundry will reduce the lead time due to shipping and schedule risk of getting chips through customs. Avalon recently announced they had hundreds of thousands of chips held up in customs for weeks, a risk any company would prefer to avoid. Most major foundries are currently located in China or the US, so those may provide the strongest options. China has larger risk of extortion or bribery, as Avalon also noted in their latest letter, while the US poses greater regulatory risk.
Another option is to locate the production facility to minimize operating expenses, such as electricity costs to keep the miners running and cooling costs to keep them from overheating. Locations such as Greenland and Siberia offer cheap power rates due to government subsidies or natural features that allow them to operate cheaply, and cooler climates that reduce HVAC costs.
Lastly, there is something to be said for attracting top talent in popular locations. There is a reason Google chose to locate some of their data centers in Berkley, Hong Kong and Dublin. Top talent is not as likely to move to remote locations. While much of the focus has been on optimizing hardware as the marginal gains become slimmer, companies will likely begin focusing more on custom software and system-wide solutions that will expand their decreasing margins. This ongoing process will require a broader range of talent than the pure hardware focus of companies today.
An interesting side effect of the new cloud mining economy is the likely increase in demand for mining futures. A futures contract enables the buyer to receive a set amount of a commodity at a specific time in the future. This has been near impossible to execute so far because of the difficulty of enforcement. A legitimate cloud mining company backing the contracts will enable the holders to freely trade a contract that can be legitimately fulfilled at expiration.
For example, the cloud company could sell a customer 1 TH/s of mining power for one month starting in October 2013. Systems could then be established to freely trade these contracts on the open market, enabling the community to determine the fair market value of future hashing power.
Having so much concentrated mining power poses unique risks to the network. The risk of a significant portion of the network going offline due to power or legal issues cannot be understated. Presuming the network has stabilized compared to its current rate, losing a significant portion of the network will have several detrimental effects. The large increase in block time due to decreased hashing power will expand confirmation times until the difficulty can adjust. A difficulty adjustment in this scenario could take weeks due to the slower rate of block discovery. If the cloud company is able to rejoin the network after the difficulty adjusts downwards, they must do so carefully.
Second, like any company connected to the internet, they run the risk of cyber attack. There must be significant safeguards in place to ensure that the hashing power cannot be diverted and used for malicious purposes. One possible solution to a system of renting hashing power that requires cryptographic signatures, possibly using bitcoin addresses, in order to control the miner. This would reduce the risk that the customer’s mining power could be affected by a software attack at the data center if the center itself cannot control the mining power. While the logistics of such a system is still unclear, it does not seem unfeasible that one could be created in the not-too-distant future.
Third, even if such a mining datacenter could withstand cyber attack, the risk of physical overtaking would be quite real. If the government of the country in which it is domiciled decides they want more control over bitcoin or simply don’t like the idea of cryptocurrency in general, there would be little to stop them from commandeering the facility. In less developed nations, the risk of terrorist or hostage situations could be equally as high.
Fourth, there is immense potential for manipulation. If one individual or entity were to control massive shares of the network when bitcoin is a hundred-billion dollar market, as they do today, they could intentionally cause network difficulties for the sake of sending the price down to buy BTC before turning on their equipment and sending prices back up. Considering that everyone from global banks to third world oligarchs would have incentive to do this, one has to wonder if regulations are inevitable.
Exciting times lie ahead for bitcoin mining, but the omnipresent concept of centralized mining creeps closer every day. Much the way the exchanges have established best practices to deal with anti-money laundering regulations, best practices or regulations for centralized mining companies may need to be established. This concept is at odds with the ideology of many current bitcoin enthusiasts, but with the entire network will relying on concentrated mining power, mitigating the risks listed above could be overwhelmingly beneficial. Either way, a debate on the subject is surely on the horizon.