After much discussion with the community and an in-depth look into the economics of ASIC manufacturing, the Sia core team has decided to reset the Sia Proof-of-Work function to brick the Bitmain and Innosilicon hardware. We are making this decision after much discussion and careful deliberation over the implications that this will have for the Sia network. We believe that this decision builds the best future for Sia, and sets an example for others in the cryptocurrency ecosystem going through similar challenges.
Sia’s History with ASICs
In June 2017, we announced that Sia would be pursuing ASICs, and that Nebulous (parent company of Sia) would be spinning up a new subsidiary to manufacture ASICs for Sia. The project gained traction, and long time members of the Sia community collectively contributed millions of dollars to pursue ASICs.
In parallel and in secret, two larger manufacturers also began development of ASICs for Sia. The first one, Bitmain, did not announce their product until 10 days before shipping, effectively disrupting a multi-million dollar project from the community without warning and nearly causing a civil war.
A majority of Sia’s core contributors and biggest supporters had put large sums of money towards the community ASIC, and many viewed the sudden entrance by Bitmain as a direct attack on the community, believing the best course of action was a decisive hardfork to brick the Bitmain machines but protect the Obelisk machines. A sizable minority vocally opposed the fork, believing that a fork would forever centralize Sia under the single manufacturer Obelisk, and that though they agreed Bitmain had done a substantial amount of damage, true decentralization required allowing freedom for ASIC manufacturers.
The core developers ultimately decided against forking. We don’t view blockchains as a democracy, we view them as a source of stability. We like decentralization because it prevents groups outside of our control from making decisions that we don’t like. Decentralization is valuable because there is nobody in control, and we weren’t comfortable releasing an update that threatened to rip the community in half.
The debate effectively closed with the anti-fork faction prevailing. Bitmain would be allowed to stay and mine on the Sia network, despite the trauma it had brought to the Sia community.
A shadow of pessimism and regret fell over the Sia community. Chatrooms slowed down, discussions shifted towards mining, governance, and arguing over what could have been done differently instead of focusing around the powerful technologies behind the Sia core code. People started to leave, and ironically the people leaving in the largest droves were those who most aggressively opposed the fork during the earlier debates. Sia’s biggest supporters and believers were the ones that got hit hardest by the mining catastrophe, and despite this loss, they were also the ones who stuck through the hardest times.
Today, Innosilicon is the dominant mining manufacturer on the Sia network. They have the only 14nm miner on the market, and as such they have the only rig capable of competing. Without competition, there is no price pressure, and it seems that there is close to, if not above, a 100% markup on their hardware. For every machine that gets sold, Innosilicon makes enough profit to produce a machine for themselves to mine. And indeed, Innosilicon is Sia’s largest miner, mining a stable 37.5% of the Sia hashrate.
The Tough Economics of ASICs
Our situation with Innosilicon is not unique. With the exception of Bitcoin, all cryptocurrencies today have a single dominant mining hardware manufacturer. Ethereum has the Innosilicon A10, Zcash has the Innosilicon A9, Dash has the Spondoolies SPx36, Decred has the Whatsminer DCR…the list goes on. Though each cryptocurrency has only a single dominant manufacturer, that manufacturer differs from coin to coin. This is not a coincidence, but rather a result of cryptocurrency ASIC manufacturing economics.
ASIC development requires a substantial amount of up-front investment. The leading ASIC for most cryptocurrencies today is a 14nm ASIC. Including engineering design costs, tooling, IP, firmware, etc, a 14nm ASIC costs a minimum of $5 million to develop, but more realistically $6–8 million. These costs do not include any actual chips or machines, but are purely non-recurring costs, the one time price of being able to make machines at all.
When the machines come to market, they have to be sold with enough markup to pay back the initial development investment. Because the ASIC business is risky, profit targets are typically between 2x and 10x the initial investment. For a $6 million initial investment, the expected gross margin on the machines will need to be between $12 million and $60 million. Before development ever even starts, a market opportunity of tens of millions in margins needs to be present.
For an ASIC that is going to obsolete existing hardware, margins can be anywhere from 50% to 100%. The story is different however for ASICs that intend to compete without being strong enough to become the new monopoly. For these machines, margins are likely to be less than 25% because the presence of competition heavily forces prices downwards. In addition to the reduced margins, the market is going to be split across multiple manufacturers, and the first manufacturer to market will have already consumed much of the opportunity.
For a new product that is going to obsolete all existing hardware and become the new monopoly, the market size required to justify investment is in the mid tens of millions of dollars. But for a new product that is going merely break up an existing monopoly rather than replace it, the market size required to justify investment is hundreds of millions of dollars due to the much reduced margins and presence of competition.
Based on what we’ve seen out in the market, the total amount of hardware sales that a cryptocurrency can support is roughly equal to its annual block reward. Cryptocurrencies with a block reward below hundreds of millions of dollars per year will not be able to profitably sustain multiple ASIC manufacturers. This cleanly explains why Bitcoin is the only cryptocurrency with multiple reasonably competitive machines.
ASIC Monopoly vs. GPU Mining
I’ve spoken about this at length in my other posts, however we believe that an ASIC manufacturer monopoly is strongly preferable to being a GPU mined coin. To the best of my knowledge, no ASIC mined coin has ever been hit by a 51% attack*. And yet, something like a dozen different GPU mined coins have been hit by 51% attacks, typically to steal from exchanges.
ASIC mining works even when there is a monopoly manufacturer because the ASICs are incredibly expensive pieces of hardware that only have value so long as the underlying cryptocurrency has value and also keeps a compatible Proof-of-Work function. The manufacturer is strongly incentivized to protect the coin and even improve the value of the coin, because that directly impacts the value of their machines. Even when secret mining is occurring, the secret miners have avoided attacking the cryptocurrency, even when they had more than 50% of the hashrate. The incentive system behind ASICs is very powerful, even when it is not very decentralized.
We also believe that ASICs are inevitable. If you take a stance of ASIC resistance, typically you just end up with secretly built and operated ASICs. And while secret ASICs are probably better than GPUs, public ASICs are better and decentralized public ASICs are even better still.
All told, we believe that embracing an ASIC monopoly is better than resisting ASICs altogether. That said, we don’t believe that a cryptocurrency needs to embrace a parasitic or abusive ASIC monopoly. ASIC manufacturers ultimately exist to serve the network, and specifically to protect the network against 51% attacks. Even small cryptocurrencies pay millions of dollars per year for this protection, and it’s reasonable to hold any incumbent manufacturer to high standards.
*This resistance to 51% attacks comes from the fact that ASICs are only useful for mining one cryptocurrency, and beyond that cryptocurrency the ASICs do not have value at all. Some cryptocurrencies however share ASICs. For example, multiple cryptocurrencies use the same Proof-of-Work algorithm as Bitcoin. These cryptocurrencies don’t share the same incentive benefits, a miner can attack one of the smaller cryptocurrencies, trashing it, and still expect the hardware to be valuable for mining the larger cryptocurrency. If a cryptocurrency does not represent the majority of the block reward for a particular ASIC, then that cryptocurrency is operating under the same weak security model as GPU mined cryptocurrencies. There are historic cases where a small cryptocurrency that shares an ASIC with a much larger cryptocurrency has been hit by a 51% attack.
Components of Healthier ASIC Economies
The purpose of Proof-of-Work is to make attacking as expensive as possible. And ideally, attacking the network is as expensive as possible for all parties, which includes the existing miners and ASIC manufacturers.
For this reason, one of the things that we like to see is low margins for miners and manufacturers. When there are high margins, at least one player (the benefactor of the high margins) is able to acquire hashrate more cheaply than everyone else, and therefore is able to more easily attack the network. If margins are lower, then this asymmetry is less severe.
When a manufacturer is also a miner, there is an incentive against manufacturing and selling more machines. As a manufacturer sells more machines, the difficulty increases, and the profitability on the machines they are mining themselves will decrease. This incentive interferes with the goal of having as much hashrate on the network as possible. Therefore, we do not like manufacturers that also mine.
We like to see high availability of hardware. Ideally, the best hardware in the world is available to everyone easily, and any deals that are presented by manufacturers to some group are also made available broadly. We like to see manufacturers with wide distribution channels, and we would like an ecosystem where anyone can enter the mining space competitively without needing pre-existing relationships with suppliers and manufacturers.
A lot of these elements manifest naturally when there is high diversity in manufacturing. When there is competition, narrow distribution channels mean missed profit opportunities, and high margins mean you will be out competed. Beyond this, the ecosystem as a whole becomes more tolerant to black swan events such as the tsunami that wiped out a good chunk of Taiwan’s manufacturing capability.
High manufacturer diversity is currently limited by the extreme barriers to entry. Money aside, the crypto mining ecosystem has an enormous amount of specialized expertise that is required to produce a competitive miner. There are a large number of digital design tricks, analog design tricks, PCB schematic tricks, firmware tricks, and even things like manufacturing assembly line tricks that get used to make modern mining machine. Whether it’s open sourcing chip designs and firmware code, releasing manufacturing guidelines, or even just explaining innovative parts of the design process, we like to see manufacturers that share the knowledge and encourage a vibrant competitive environment.
One of the more critical things to miners and investors is predictability within the ecosystem. Today, there is a lot of risk associated with mining that comes from not knowing how much hashrate is currently being produced, and also that comes from not knowing when your machine is going to be made obsolete by new competition. These risk factors increase the barrier to entry for mining, increase profit requirements, and overall end up reducing the total hashrate on the network. Ideally, all manufacturers announce new products months in advance, especially if those products are going to obsolete hardware that is actively being sold to the market. And similarly, ideally manufacturers disclose in full how much hashrate has been produced, so that buyers of mining equipment can make more informed decisions.
Finally, a healthy mining system has failsafes. In the case of Sia, Obelisk built a tiny, secret extra circuit into the mining chip that would allow the Sia developers to perform a hard fork that breaks manufacturers without that extra circuit, but does not break the Obelisk machines. There is a nearly infinite number of ways to add such a circuit to a chip, and a flexible ASIC would not be able to cover all of those opportunities without paying a substantial performance and efficiency penalty. We encourage all manufacturers to add secret extra circuits that can act as failsafes.
These failsafes will typically never be used. However, in the event of an unhealthy monopoly miner, or even an outright attack from a majority hashrate owner, these failsafes can be leveraged to break the attacker’s hardware without breaking all of the ASICs on the network, allowing the network to continue to be protected by ASICs, but escape the attacker or abuser.
The Value of Mining to a Cryptocurrency
The purpose of Proof-of-Work is to make attacking the network expensive. It is a service provided by miners to benefit the network. And while miners are typically handsomely rewarded for the service they provide, this reward does not exist as some inalienable right for the most successful miner. If the network’s miners are parasitic, abusive, monopolistic, or otherwise inhibiting the success of the network, the network has every reason to kick them out.
For the Sia network, an important line was crossed when secret ASIC projects superseded a public project that had substantial community investment. Though the Obelisk project got a few things wrong, largely Obelisk went about ASIC development and manufacturing in the right way. Obelisk was focused on adding value to Sia, to the Sia community, and to the cryptocurrency community as a whole, which is what we expect from a service provider.
Sia did not fork initially because there was a lot of confusion, a lot of emotion, and a great fear that the heavy conflicts of interest would cause the development team to make the wrong decision. Since then, there has been time for emotions to cool, for level heads to prevail, and for a second community fork proposal to come forward. Unlike the first fork proposal by the community, this second proposal experienced widespread support and virtually no opposition at all from regular members of the community.
Sia is forking today to reprimand the current ASIC monopoly for the damage it did to the Sia community, to make whole the supporters of Sia’s community ASIC project, and to send a clear message to all future Sia ASIC manufacturers: we will not tolerate an abusive ASIC monopoly.
The Way Forward for Sia
In v1.3.6, Sia will be releasing hardfork code which bricks the Bitmain and Innosilicon ASICs, but continues to allow Obelisk ASICs to mine on the Sia network. This revokes the Innosilicon monopoly and instates Obelisk as the incumbent ASIC monopoly.
In the short term, this is a step backwards for Sia’s security. The hashrate will decrease substantially, which means our security margin against an attacker will be lower. In the long term, this sets the stage for a healthy mining community and a higher overall difficulty. We fully expect that the 28nm Obelisk ASICs will be replaced by a 16nm chip from another manufacturer, who will become the new manufacturing monopoly for Sia.
So long as the new manufacturer is contributing to the health of the Sia mining ecosystem and avoiding abusive practices, the seat of power will be passed along gracefully. However, the Sia community is not afraid to take action a second time to break a parasitic or abusive ASIC monopoly. In the event that we need to brick another manufacturer in the short term, Obelisk has a chip design prepared that we can tape-out and use to defend the network. In the long term, we hope that there will be good manufacturers who build ASICs with secret circuit extensions that we can leverage instead of needing to explicitly fall back upon Obelisk.
Hardforking While Preserving Decentralization
When we talk about decentralization at Nebulous, we aren’t talking about spreading things out or ensuring that at least multiple people control something. We’re talking about creating a system where there is no point of failure and no point of control at all. And while a perfectly decentralized system only exists in theory, we strive to make Sia as decentralized as possible.
Sia is an ungoverned blockchain. There is no built-in mechanism on the Sia network to change the consensus rules, and there is no mechanism in the software that the developers can use to force people to upgrade. The only way that Nebulous can encourage a fork is to release new code, and then encourage people to upgrade.
This leaves people with the opportunity to reject the upgrade, and to instead continue using the old software and the old blockchain. If enough people rally around the old software, there could be a network split, and Sia could divide into two blockchains, in the same way that Ethereum split into Ethereum and Ethereum Classic, and in the same way that Bitcoin became Bitcoin and Bitcoin Cash.
At Nebulous, we view these cryptocurrency splits as one of the most powerful innovations of the blockchain space. Under traditional governance structures, a single decision gets made and everyone has to live with that decision. But when the network is able to split, you can get solutions where two groups of people with incompatible demands can both get what they want.
We will be structuring the Sia hardfork code to enable a group of dissenters to easily split off and be on a separate blockchain where the hardfork was never implemented. The hardfork will be released as its own release, v1.3.6, where the only code updated is a handful of lines of code + tests required to implement the hardfork. The code will be implemented in a way that easily allows a dissenting group to remove the hardfork code and yet continue merging changes that are made to the primary Sia repo. So long as the siafund ownership is maintained on this fork, members of the dissenting community will be welcome in the Sia community, on the Sia discord, on the subreddit, and will be able to receive support and help directly from the Nebulous support staff.
Perhaps the most amazing thing about a potential Sia network split is that all users will be able to continue to use their current files that they have on Sia. Uploads and downloads will continue to work, no matter what side of the split you are on, and so long as the minority side of the split has enough hosts (50–80 is what most users will require), the repair mechanisms of the Sia network will be able to repair your files from across both networks and ensure that your files continue working into the future. If the minority side of the fork does not have enough hosts, users will have time (most users will have several weeks) after the split to download their files and find an alternative way to back them up.
Overall, we are sad that things have come to the point where we need to take action to reject an existing ASIC monopoly. We hope that in the future, the mining ecosystem is more self regulating and that this is the only mining hardfork required for Sia. However, we believe that we have set this hardfork up to be as decentralized as possible, and we believe that any substantial dissent from the users can be accommodated fully and gracefully through a Nebulous supported network split.
David Vorick is Sia’s lead developer. In 2014 David co-founded Nebulous, the company that funds Sia development. In 2017 Nebulous launched its Obelisk subsidiary, a producer of ASIC mining hardware.