Energy grids and bitcoin mining are two fairly complex topics, but understanding generators is key to learning about both.
This article is the first in a series about bitcoin mining and energy infrastructure. Each article offers an introductory level explanation of electric grids and their relationship with mining to better educate miners and other bitcoin investors.
Read Part 2: Transmission, Curtailment, and Behind-The-Meter here.
The most impressive characteristic of bitcoin is the mining algorithm. More beautiful than the math and theory behind the issuance and difficulty adjustments themselves is the fact that it all actually works. That mining death spirals don't occur, that there is a global distribution of hashrate competing for power infrastructure and cheap rates. It’s amazing. What was once cypherpunks competing with laptops in the late 2000’s is now an incredible ecosystem of radically different business models — hosting and licensing, on-grid proprietary behemoths, generator joint ventures, off-grid natural gas, and even pig manure!
Because of this incredible success and diversity of business models, there is no shortage of arguments about bitcoin mining’s effect on the grid. Oftentimes bitcoin investors find themselves on social media (this author included) arguing pointlessly against one belief about bitcoin and energy, and then turn around an hour later to argue with someone else on the complete opposite side of the political spectrum about their beliefs on the same topic.
Energy and bitcoin mining are two fairly complex topics, and many online arguments may well be avoided (or at least be more educated) if all parties read a primer on how well-known grids like the Electric Reliability Council of Texas (ERCOT) function, and how bitcoin mining can interact with these markets or specific pieces of electrical infrastructure from a power systems perspective.
Part 1 of this series will focus on generators and bitcoin mining’s relationship to their operation and function within the power system. But readers should know that concepts found in this article are generalized and simplified to give an introduction to grids, not an exhaustive study of them. Grids are incredibly complex systems, and engineers spend entire careers working on only one small aspect of their operation. Grids are also a marvel of technological coordination that carries all the mathematical complexity, tariff law, and political baggage that one can imagine. With all that in mind, let’s get started.
A power system is a collection of power ‘sources’ (generators), ‘paths’ (transmission), and ‘sinks’ (loads) which must operate at a high level of reliability & within the bounds of physics. Generators are the sources of the power, and thus generate power that must be delivered to loads through the pathways available to them. The power must be consumed immediately by the load upon generation, unless it is stored using energy storage technology. To manage this process of matching real-time generation with load by way of paths available, a central authority (Grid Operator) normally uses a mathematical algorithm to identify the most economic (cheapest) and most reliable (able to still operate in events where you lose an element(s)) manner of operating the grid. In layman's terms, this means fancy math is used to determine the best combination of generator output across the system to serve load while keeping any single element from NOT being overloaded in case something fails or trips.
Below is a simple image illustrating the difference between sources of power (generators), paths of power (transmission lines and substations), and sinks of power (loads, residential load in the case of the image). Transmission and Distribution are usually differentiated by voltage level, but both are ‘paths’ through with power flows. ERCOT and other ISO’s normally only operate on transmission level voltages, and distribution companies manage those lower voltage levels.
Sources of generation have traditionally been thermal generators. Mostly coal, natural gas, and nuclear power plants that expend fuel to boil water & generate steam, which is used to move a turbine to generate power. Recently, solar & wind have become an increasing percentage of the generation mix. These facilities operate passively based on the weather, and rely heavily on solar & wind forecasting to predict their output.
Today, generation developers are companies that specialize in siting & interconnecting generators to the power system. These companies focus on understanding what the future of the grid will look like, and how they can develop generation infrastructure to capture value. Generator owner/operators are companies that specialize in owning & operating the generators once they have been built. Sometimes generation developers also own and operate the facilities they develop, and sometimes they sell them to others.
High power prices are the primary signal for generation developers or investors to build more generators. Recent tax legislation has become another signal which has spurred development in solar, wind, & storage as investors aim to capitalize on the large tax windfalls from building these types of facilities.
ERCOT is the nonprofit entity that operates the Texas grid, which accounts for 90% of the electric load in Texas (the name of the grid is also ERCOT, I know, it’s confusing). ERCOT owns no wires, substations, or generators, but rather are the “manager” of the system. This means that ERCOT has special responsibilities for planning the system, handling the actual transactions, and operating the market in real-time.
ERCOT has a “deregulated” market, which means that different entities must own different portions of the industry. No transmission owner can also own generation, etc. Power generators are privately funded in ERCOT, and ERCOT generators operate in one of two distinct categories (markets): Energy and Ancillary Services.
Here are the basics of the Energy market.
All generators produce and submit to ERCOT (hereafter referred to as “the grid operator” ) a “bid curve”, which serves as an indication of how much money they would need to spend in order to produce a megawatt (MW) of power. For instance, an offline coal facility needs the grid price to be at a certain dollar per megawatt hour ($/MWh) value before it comes online (a.k.a., starts generating power) in order for them to breakeven on their costs associated with turning on. Once online, the plant needs a marginal bit of more money to pay for a marginal bit more fuel to ramp up production a marginal bit more, and so on. The resulting bid curve represents the marginal cost, or the dollar value required to produce the next megawatt of power. This ends up looking like a curve starting at the facility's minimum output, and going up and to the right, as thermals need more fuel to produce more power.
Unlike thermal generators, wind and solar are passive energy sources with a production cost of $0, meaning that their marginal cost to produce the next megawatt is always $0 (a flat line at $0, instead of a curve going up). With the tax incentives for renewable energies referenced in the introduction to this article, some renewable energies can become negatively priced, meaning a generator receiving certain tax benefits could get a $26/MWh credit just for being online, and will therefore bid -$25/MWh instead of $0/MWh into the market, and capture $1/MWh even though the grid price is reflecting their negative bid (-$25/MWh). Since renewables bid $0 or even negative into the market, this is often why they are referenced as “cheaper” than thermal generators.
Confused yet? Let’s keep going…
Grid operators generally dispatch generation (meaning, they order a generator’s output to a specific level) starting with the lowest marginal cost generators and then move in the direction of the highest cost generations by working through the “bid stack.” Operators also optimize (and therefore set the grid price) for something called “constraints”, not just the cheapest generation.
But typically in any local area, the cheapest generation is dispatched first, and the bid stack is incrementally worked through upward to dispatch enough generation to match the forecasted amount of load on the system. Eventually, operators will climb the bid stack until they reach a total number of generators who collectively will provide the amount of power needed to meet demand for the given time interval.
It is crucial to note that a generator's marginal cost data (the values they submit as part of their bid curve) do not contain profit or information about the capital deployed to build the generation facility. It only communicates variable cost, like fuel cost and regular maintenance.
The entity sitting at the top of this bid stack during any interval who provides the final, most expensive megawatt hour sets the grid price for all generators for that interval, and so they are sometimes called the “price setter.” Because wind and solar both bid the market at $0, they are in contrast often called a “price taker”. This is because these energy sources who bid $0 never plan on being the marginal cost unit (the one at the top of the bid stack). If they did, then everyone would be paid $0/MWh for their energy…and no one would make any money! There is a whole field of study regarding how power system markets will work when renewables start to “set the price”, which this author will dutifully ignore in this article for the sake of time. We are omitting “constraints” here in favor of simplicity, but just know that generally, this is how it works.
This dispatching process normally ends up looking like the graph below, where power generators below the market clearing price (gray line) and to the left of the demand curve (blue line) are turned on and get paid the clearing price (gray line) for their energy. Generators above the market clearing price and to the right of the demand curve stay offline. This process runs continuously throughout the day, redispatching generation based on expected load levels.
Decisions about whether to build a new generation asset are made after careful research and power system modeling by engineers (like this author) to understand what the expected future price at a specific generation location on the grid will be. These price models give no certainty, so in some regards they’re akin to using a very powerful, precise, but inevitably inaccurate crystal ball to foresee market conditions, transmission upgrades, future retirements, load growth, and other generation competition.
For example, if a company expects natural gas to become the “price setter” for the local area and they’re planning a solar farm, they now have to analyze whether or not the price set by the natural gas generator will be enough to recoup the capital cost of the solar farm. So, while solar generators might bid $0 into the market, that does not mean that capital expenses can be paid back at a price of $0/MWh.
Enough about energy generators.
Let’s move on to the second category: ancillary service providers.
Aside from simply producing enough energy to match demand, grid operators need to be able to maintain a constant frequency of 60 Hertz (HZ). In real power systems, frequency is actually a dependent variable of many things on a network, and thus maintaining it and keeping the system afloat is of the utmost importance. Frequency crashes can lead to blackouts, which can mean weeks or months without power. Readers can see ERCOT real-time frequency and inspect the different types of ancillary services and their deployment here.
To maintain frequency, grid operators buy & reserve ahead of time “ancillary services” from generators and loads that are able to rapidly increase or decrease their output. By rapidly increasing output, a generator can increase system frequency. By rapidly decreasing their output, a generator can decrease system frequency. Grid operators have a certain “amount” of MW they need to buy (reserve) ahead of time to ensure they have enough "firepower" online to deploy (order up or down) during real time to maintain frequency through the day. When real-time comes, grid operators deploy these procured reserves to manage frequency fluctuations during the day.
This frequency market is separate from the energy generation market, there are two markets. The energy market, in which generation is turned on or “matched” with expected load (or demand). And ancillary services, in which frequency is maintained around 60 HZ to keep the system operational and avoid blackouts. The energy market is the big, heavy dial, the ancillary services market is the small, precise dial.
Certain types of loads can also provide ancillary services. And this is where bitcoin mining – often cited as a “grid balancing agent” – comes into the picture.
After the basic explanation of energy generators and ancillary services above, it’s appropriate to revisit the same topics while adding bitcoin in the explanation to understand the role that mining can play for the grid.
This section describes miners who are exposed to nodal pricing, or wholesale pricing, the same type of pricing that generators receive. This price goes through a couple layers of hedging, aggregation, and abstraction before it gets to industrial loads or residential consumers. (These layers are why residential customers are able to pay flat rates.) But ERCOT has special rules on the horizon that would allow some transmission level loads such as bitcoin miners to receive nodal pricing directly. For now though, most bitcoin miners are exposed to nodal pricing through PPAs and some fancy arbitrage. All of these details can get hairy quickly, so for educational purposes, readers can assume that everything described here about miners getting variable pricing is (or could be, if miners were to act rationally) true.
A fun exercise for bitcoin mining machines is to calculate their $/MWh rate. Similar to how all generators have a “bid curve”, which they must submit to grid operators, bitcoin mining machines can be abstracted into a $/MWh value that they turn energy into. This number is also sometimes referred to as their “breakeven price”, since, if the machines were being fed with electricity that cost that amount, they would stop being profitable.
The tables below show some back-of-the-napkin math using a hashprice of $0.22 to determine what the $/MWh rate would be for various mining machine models. This math includes some extra steps in order to find what quantity of machines would be needed to constitute one MW of load.
The math for determining the breakeven price without a machine count is a bit simpler.
So, what does this have to do with the grid? There are two key takeaways.
Remember that generators are dispatched in order of their marginal cost. Cheaper generators are dispatched first, and more expensive generators are dispatched last (if at all). Load Resources can also register with a grid operator, receive a small payment, and be part of this bid stack as well, not just generation. But instead of adding to the generation side of the market, this load works as a tool to decrease the expected load target.
From the tables above, Antminer S9’s have a breakeven cost of about $90/MWh, or 9 cents/KWh. By participating in the market as load resources, bitcoin miners coordinate with grid operators in ways similar to generators, but instead of ramping up generation they power down load demand in response to wholesale pricing. The result is that the miners flexibly push the marginal price of power downward for the grid. Put differently, they do not push the marginal price higher than their breakeven.
Continuing to use miners with S9 machines as an example, here’s how the grid affects their operations. As ERCOT operators work up the bid stack and turn on generation until it matches expected load, they face two options when they reach the $90 marginal cost range.
This scenario comes with a few caveats.
First, generators are required to respond to orders from grid operators or face fines. If the grid doesn’t function with consistent mathematical precision and transparent coordination, bad things happen. With the ability to modify the grid in a material way as a generator or load balancer comes great responsibility.
Secondly, for bitcoin miners, adjusting load in response to price signals is currently an ‘opt-in’ situation, where miners are paid for participating regardless of whether or not they are actually dispatched in real time. This means the example bitcoin miner with the S9's will get paid, even if they don’t have to turn off. But if a miner is exposed to nodal pricing (e.g., they face the choice of either paying the $91 dollars to make $90 dollars, or shut off), powering down when prices aren’t economical is simply a rational way to participate in the market just like generators that do not turn on unless it is economical for them to do so.
Okay, caveats aside, back to the two options listed above.
What’s the best deal for minimizing the total cost of the system? Effectively, any type of load that is exposed to wholesale pricing that can turn off during times of high pricing, will turn off. For example, Steel fabricators have been shutting down in response to market pricing for a while. And this same behavior applies to bitcoin miners. Any miners that don’t want to be online when pricing is uneconomical, won’t be. So..
Bitcoin miners are unique because of how fast, transparent, and flexible their response can be to pricing fluctuations.
Riot’s Whinstone mining farm in Texas is a perfect example of miners powering down during tight conditions in ERCOT. During winter storms in February 2022, the site powered down 99% of its operations to reduce load. This flexibility made headlines.
Now imagine if ERCOT has visibility of every miner over 10 MW that’s connected to the transmission network. The price at which the machines breakeven would then become an integral aspect of wholesale power pricing because the bitcoin miners themselves can set the wholesale price.
Consider a high pricing situation where marginal power prices climb to $90/MWh, and bitcoin miners in Houston who are running some S9s are then ordered to turn off. (Remember: they wouldn't want to be online anyways, and they get paid by ERCOT just for being available for powering down at higher prices.) By turning off, let's say these miners actually allow load to match generation, and they therefore become price setters, setting the marginal price for the entire ERCOT system. All generators then get $90/MWh for that interval, assuming no constraints, because that price is how much bitcoin a MW of S9's could have produced with that marginal MW. Bitcoin became a price setter!
Imagine further that the bitcoin miners with the S9’s would submit something like this graph below to ERCOT, where, instead of a bid curve, they offer to reduce their MW load by a certain amount depending on the grid price. This would be the opposite of the previously mentioned generator bid curve where marginal load reduction is offered instead of marginal power generation. And, of course, it’s not difficult to imagine a miner who isn’t only running S9s, but who also has a diverse portfolio of machines with an array of breakeven prices.
This type of paradigm will certainly drive cheaper ASIC’s hard, and thermal cycling will become a real concern for the machines. But what else? This is still adding load to a system, right? Kind of a "push you off a ledge but then catch you before you fall" and claim to save your life type of deal. So, in a place like ERCOT where the market is tight already, $90/MWh is incredibly expensive. But bitcoin mining can do something else extremely well.
Attentive readers will remember from earlier in this article how generation developers rely on engineers like this author to run sophisticated and not-so-accurate models to forecast pricing that determines if a location is lucrative enough to build a generator. Also, remember that ERCOT prices are considered “price signals” for new generation, which makes them inherently lagging indicators of load growth that could underwrite new generation buildouts.
So, what if new generation development could be underwritten in part with bitcoin as a colocated offtaker (buyer of some amount of energy)?
Instead of having to solely rely on grid pricing for revenue, new generators could buy “offtake insurance” that enables them to contract with a bitcoin miner if their grid pricing forecasts turned out to be a bust. This would present an incredibly novel tool for derisking generation development, allowing generators to bring their offtaker (their buyer of energy, a bitcoin miner) with them to a new site.
For renewable generators this is especially enticing. A 200 MW nameplate solar or wind farm could see large gains by colocating with a much smaller bitcoin mine (e.g., 30 - 40 MW) who just pulls grid energy whenever the solar or wind farm are not generating enough to meet their needs. This colocated mine could produce enough revenue for the generator to shore up financing while still allowing for 160 - 170 MW of nameplate capacity to the grid during peak hours, all while still bidding $0.
The economics are slightly different for thermal generation (e.g., natural gas), since per the current “marginal cost” paradigm, the marginal cost of the thermally produced megawatt supplied to the grid after serving the colocated mine would technically be more expensive than those megawatts first supplied to the mine. Recall how the normal bid curve is up and to the right, as thermals require more fuel to produce more power. This isn't to say that renewables are better, as all generation types have their tradeoffs, but rather a highlighting of the implications of flat $0 marginal cost generation.
Current financing paradigms for generation asset development aren't ready for this type of underwriting — most companies developing generation have a rigid and conservative set of requirements, and bitcoin is just too novel of a technology to fit this hole as of now. However, the paradigm shift is here, and generation developers that are willing to be the first to leverage this model will certainly reap the rewards. I expect that this model of colocation to improve financing will hit existing generators first, since they have no downside to trying something new if they are already missing revenue targets.
This section explores more advanced implications of generators partnering with miners.
In certain situations where the grid operator knows that the real time conditions in the immediate future will be tight (not much margin between available generation and expected demand), the grid operator will “RUC” (Reliability Unit Commit) thermal generating units, whereby the grid operator orders certain thermal generators to be Online, running at their minimum, and thus available to be dispatched up if needed. Since generators take time to come online, having them ready and online is an extra conservative approach to operations planning.
This is different from standard procedure, since normally those generators decide themselves when to be available via their bid curves & their COP (Current Operating Plan). Earlier in this article, grid operators were described as generation dispatchers, but that’s only half true. Generators actually tell the grid operator when they will be online and available. Then, once generators are online and available, the generator must follow dispatch orders from the operator.
When the grid operator RUCs the unit, the operator has to pay a premium to the generator to be online and available — a premium to what the economic conditions of the grid forecast (else the generator would have decided to turn itself on). With colocated bitcoin mining, a generator would always have an incentive to be online & available, and thus would make RUCing a thing of the past, and move the cost burden of RUCing away from the consumer and onto the colocated bitcoin miner. Environmentalists would probably not like the idea of running thermal units more, but they couldn’t deny that moving RUC costs away from the consumer and onto the bitcoin miner would reduce consumer facing costs associated with RUCs while improving reliability.
The ancillary services potential of bitcoin mining is a bit more straightforward than the generation aspect, since most readers are likely familiar with the narrative of bitcoin mining providing ‘grid balancing’. To quickly recap Ancillary Services: after grid operators ensure via the energy market that they have enough generation to match demand for any interval, they also need to regulate the frequency on the grid as it oscillates between 59.5+ and 60.5- HZ. (For context, a grid’s frequency is always oscillating within this band as operators on a real system work to match generation and load.) Traditionally, keeping the frequency within the bands of this range is done by leveraging generators that are able to quickly ramp up or down their output, and also – for certain acute (fast) events – by big loads that are hooked up to relays such that they immediately drop if the relay senses a frequency event.
The chart below is an example of a generator-forced outage (i.e., a generator tripped or was forced offline for some other unplanned reason) that caused frequency to dip (blue line) resulting in load resources (e.g., potential bitcoin miners, green line) immediately responding by tripping themselves and restoring the grid’s frequency.
Bitcoin’s place as a grid resource is pretty clear. Large flexible loads that have the ability to pay themselves to be online and respond immediately to frequency events is a new asset class for the power system. So what exactly does this look like? Besides just responding to acute events by dropping load, in ERCOT, Ancillary Services are currently sold in the "day ahead", meaning the day before operations. Miners who can qualify to provide these types of services (by proving that they can ramp up or down and follow instructions quickly) will sell their capacity in an auction. The Grid Operator will buy the services, which forces the miner to reserve those megawatts during their operation the following day. During real-time, the miner may be called upon to ramp up or down to fulfill their obligation, depending on which type of service they sold.
Firmware that allows the mining machines to be ramped up and down while minimizing long term harm to the machines would be an incredible tool for the power system. But the extent to which ASIC mining machines can or will incorporate this type of ramping capability is unclear.
As briefly mentioned above, thermal cycling will likely be a limit to how often miners can ramp up or down to move frequency around for operators. This author expects that eventually, mining companies will pop up specifically to perform this type of service, using extremely old machines and pairing them with generators that are unable to perform this service on their own (renewables or nukes), that way they can make use of the already existing electrical infrastructure and split revenue with a generator that is already well versed in electrical markets.
Even though this author is very bullish on the integration of bitcoin mining machines into the power system, placing such concentrated rampable loads on the system also creates risk.
One risk is unclear operating schedules. If bitcoin miners are not transparently sharing their operating schedule or bid curve information with a grid operator, the operator has no knowledge of when, and by how much bitcoin miners will turn off or turn on throughout the day. And because transmission level connected large miners are not treated with the same scrutiny as generators, they are currently not required to share this operation schedule or submit a bid curve to grid operators.
Changing load in a drastic manner affects frequency, which can cause blackouts. Large miners that ramp up or down their output have an outsized effect on the grid as their output can drastically affect system frequency and the speed at which the frequency changes. With more load ramping, more ancillary services will need to be procured to ensure that the grid operator has enough firepower to manage frequency during real time operations.
But how much ancillary services need to be reserved to handle bitcoin miners turning off and on? Grid operators won’t know unless the miners give them an idea of how they plan to operate. Grid operators are already having to procure more reserves due to renewable penetration, bitcoin miners who don't transparently operate present another reason for the grid to procure (buy) more ancillary services resources to regulate frequency in real time.
This problem will likely be solved with new rules and additional incentives. Bitcoin miners operating at significant size that interconnect on the transmission network will likely face interconnection rules and responsibilities that more closely resemble existing rules for generators, not rules for loads. For example, bitcoin miners will have to prove that they have redundancies in their network connections, such that a faulty cable won’t drop their entire load.
Miners could also have to submit operations schedules or bid curves, and they could face penalties for not behaving in accordance with their schedule or bid curve — penalties which will in turn be used to pay for more ancillary services. For their service, the miners will likely be allowed to take nodal pricing and avoid pesky transmission, delivery, and coincident peak charges that would add to their all in cost.
Miners connected at a distribution level will likely not face the same scrutiny, but they also won’t be exposed to the same pricing incentives and rates as those transmission connected miners.
Congratulations for making it this far! The goal for this article (and the entire series about the grid) is to develop among bitcoiners a better understanding of how modern grids are managed and operated, so that creative miners and enthusiasts can think of new ways to marry these two industries.
Hopefully this piece provides some useful insight into how ERCOT works and increases the reader’s appreciation for how much potential bitcoin mining has to change the power industry. Many topics discussed in this article are generalized for the sake of simplicity, and most of the above sections probably deserve 50 pages of their own analysis.
Topics like curtailment, negative pricing, and others are also important pieces of the puzzle presented by bitcoin mining, and these will all be addressed in subsequent articles.
Read Part 2: Transmission, Curtailment, and Behind-The-Meter here.
This article was written for the Braiins blog by Blake King. Blake is a power engineer who builds and analyzes software models of electric grids. His views here do not reflect those of any of his past, present, or future employers. Follow Blake on Twitter.
Bitcoin mining software company: Slush Pool, Braiins OS+ & Stratum V2.
By miners, for miners.
Increase hashrate on S9s to 17+ TH/s, improve efficiency as much as 30%, and get 50% lower pool fees on Slush Pool
Reduce data transmission between your farm and pool by 95%. Configure parallel usage of multiple pools. Set a backup pool for the whole farm.
It focuses on making data transfers more efficient, reducing physical infrastructure requirements for mining operations, and increasing security
It focuses on making data transfers more efficient, reducing physical infrastructure requirements for mining operations, and increasing security