We’ve all been there, you’re locked in, desperately fighting with spreadsheets, the dishwasher’s going, and a banana cake is in the oven. Suddenly, all goes quiet, lights are off, oven’s off, PC dead as a doorknob. You check the fusebox, it’s fine. Must be a blackout. Usually, the power comes back on after 30 minutes to an hour, but you’re left standing there wondering what happened. 

We’ve just seen a country-wide scale event like this, which was much more extreme. On 28th April, the Iberian Peninsula experienced a catastrophic breakdown of the grid (see here).

Power was off for several days. I won’t dig into the details of this particular event; doing so is way beyond my pay grade. Suffice it to say it follows the ideas I want to introduce in this article.

We’ll begin with the simplest model I can think of, getting to grips with the issue from first principles. Then I’ll walk through how the basic premise scales up the Australian grid (or at least the east coast, the NEM), before introducing the shadowy supergeniuses in control of it all (AEMO) and one of the most important market mechanisms they use - Frequency Control Ancillary Services (FCAS). 

Part one: First principles

Imagine a way-over-simplified system with a single source of power and a single load. Our source of power is just a battery that produces 10MW of power at an assumed 1000V. The battery has no rotating components. It is connected to a very basic inverter. The inverter is designed to produce 50 Hz AC power; however, let’s imagine it has no sophisticated voltage controls or protections (unlike real inverters). All that our inverter comprises is a basic switching and timing circuit.

This specific frequency of 50 Hz is a fundamental design standard for almost all AC electrical equipment. Just like a clock needs to tick at a precise rate to tell accurate time, electrical devices like motors rely on the AC power cycling exactly 50 times per second. If the frequency deviates too much, it can cause motors to speed up or slow down incorrectly, or sensitive electronics to malfunction. Our simple inverter, despite its limitations, is designed to produce 50 Hz because that’s what the heater and any other AC device expect.

Our load is a large but mechanically simple industrial heater. It has a large coil that produces heat by running a current through it. Importantly, the heater has a dial to adjust temperature (which can change its power consumption). The heater is currently set to draw 10MW of power.

That’s the start, with a simplified system in equilibrium, all is well. The heater is producing heat, and you can make up your own story about what the heat is used for. The important point is that the voltage, current, and frequency are all stable.

Next, the foreman operating the heater turns the dial up to increase the heat output. Assume the heater now wants to draw 20MW of power. We’re simplifying, so also assume that in this instant, the voltage remains at 1000V.

To draw this much power, the heater now has to draw more current (because power = voltage x current, it’s the rules). Since we assumed the voltage remains at 1000V, there is an instantaneous demand for more power from the battery than it was previously supplying.

The battery now attempts to supply this current, which means the current must flow through the battery’s components, the inverter’s components, and the wire connecting the battery to the heater. All of these physical components have resistance. This means the actual voltage the heater receives will be lower than 1000V.

Now that the heater is receiving voltage lower than 1000V, things start to go wrong. The power consumed by the heater is proportional to the square of the voltage across it (rules are rules), so power = voltage^2 / resistance. Following this rule, since the voltage has decreased, then that means the actual power the heater draws will be less than the 20MW desired. For example, it might only draw 15MW, but it depends on the resistances in all the components, which we haven’t set.

Things just get uglier. The battery and inverter are operating under extreme stress, trying to deliver the higher current demanded, but at reduced voltage. Inside the inverter and the battery, the components start getting hot (rules, again). Since we assumed the battery and inverter are super simple, the inverter can’t adjust its power output based on actual grid conditions (like a sophisticated one would); it can only fail. Then the heater goes out, and the foreman gets mad; it’s just a bad day all around. 

The fix: add a spinny thing.

Reimagine the situation just as before, but now, instead of a battery, imagine the source of power is a contemporary synchronous generator system. Say it’s a hydro plant, where the spinning turbine is directly connected to the grid. This generator, unlike the battery-inverter, inherently uses its large rotating mass to move a metal rod up and down inside a magnetic coil and produces AC power. Importantly, it synchronises with our system’s required 50 Hz rhythm.

In this case, the story proceeds as before. When the foreman turns the heater dial up the heater wants to draw 20MW. In this case, the increased current flows through the impedance of the generator’s internal components and the wire. Just as before, this causes an immediate, rapid voltage drop at the heater’s terminals. In that instant, the heater will not get its desired 20MW because the voltage is too low.

Since the generator’s power output is less than the load demand, the lack of energy causes the rotating mass of the generator to decelerate (energy is conserved), and the frequency of the AC power in the system drops below 50 HZ. Because the generator’s turbine is a large mass, it has substantial inertia (pesky rules again), so the drop in frequency happens much more slowly than in our first story, and there is no catastrophic failure of the system. Instea,d this inertia buys time for the system to react.

The reaction comes from the manager of the generator. In the real world, this is a piece of software called a governor, but it’s useful to imagine it’s a person. The manager sees the drop in frequency and presses some buttons to let more water flow to the turbine, the frequency climbs back to 50 HZ and the power output of the generator increases.

Then the heater gets its required 20MW of power at 1000V and 50Hz, and the foreman is happy, everyone’s happy, it’s just a good time.

Frequency, in this simple system, indicates power imbalances and triggers the manager of the generator to respond. Equally, we could have assumed the foreman can be notified that he shouldn’t increase the heater’s dial - we can manage the imbalance either on the supply (generator) or demand (heater) side, or a mix of both. 

This hints at a crucial question: how the hell do we coordinate this? Even with just a single load and a single generator, what sort of contracts would the foreman and generator manager need to keep everything running?

Part deux: AEMO, the not-so-invisible hand coordinating the market

Above is a story of a system with one load and one generator, without protections for the management of voltage or frequency. I walked through how this simple system could suffer a catastrophic breakdown as a result of the load increasing. Followed by a slight added complexity, which solves the issue. 

Everything that applies in the simplified model also applies in the real world; the basic principles scale up and create complex incentives, coordination, and knowledge problems. 

In my simple story, there was a manager in charge of the generator and a foreman in charge of the heater. In the much more complex National Electricity Market (NEM), there’s a third main actor: Australian Energy Market Operator (AEMO, because it operates the energy market, perfect naming convention). Among a host of competing priorities, AEMO has one role we’re particularly interested in: making sure we have a good time, i.e making sure the grid is stable. How it does so is the economics bit.

The NEM covers a huge geographic area (all the way up the East Coast from South Australia up to North Port Douglas). Within the vast area it covers are around 10 million customer connections (source). AEMO provides an interactive map that shows you how many substations there are - lots! 

here’s a map from AEMO

With all those connections and interconnections, the only constant is change. Strong winds can suddenly break connections, generators can trip for a host of reasons, and sometimes physics just says “no”.

On the flip side, sometimes large loads connect unpredictably and want instant power to be available. The physics says supply must equal demand or things get bad, so how does AEMO make sure supply equals demand?

In short, dispatch instructions. AEMO coordinates the supply-side market participants using signals to their control rooms. The economics of it could be summed up as: AEMO is a visible hand trying to approximate the role of Smith’s invisible hand in the market*.

Though not the focus of my current article, dispatch can be (over)simplified as follows: Every five minutes, generators send an offer to AEMO containing a price-quantity pair for power output. AEMO sorts the offer from lowest price to highest and matches supply to forecasted demand, starting from the lowest offer to create a dispatch interval/dispatch window, within system constraints. Importantly, generators all receive the highest marginal price (spot price) of all offers within the dispatch interval. Then, all generators included in the dispatch interval receive revenue equal to their power dispatched times the spot price.

So, assuming the system is in balance. This plays out, and everyone’s happy. But that’s not guaranteed in such a complex interconnected system like the NEM. So, what happens if, after offers are placed and power is scheduled, a generator suddenly trips off (cutting supply) or a large load connects unexpectedly (rapidly ramping up demand)?

AEMO needs some sort of backup power supply or backup demand as a contingency for when imbalances happen. One of such contingencies is what we call this service provision Frequency Control Ancillary Services (FCAS). So named because AEMO uses system frequency to understand where, when, and how much extra supply or demand (load) is required to return the system to the normal stable operation. 

But AEMO doesn’t own any generators, batteries, or large loads itself. So, how does it get them? Someone came up with the idea that you can sell backup supply or demand (load) to AEMO, for a fee. This idea implies that AEMO has to find the right incentives to bring backup supply or backup demand (load) online in a flash.

There’s a market for that! 

Actually, there are eight markets, but we’re simplifying.

To procure FCAS, AEMO runs a similar auction process to the market for power. Participants submit bids to AEMO every five minutes, containing a service type and availability across up to 10 price bands. The service type is one of: increase/decrease frequency in one of the three time intervals.

AEMO then considers all bids and finds the least-cost solution that meets all system requirements, matching the supply of FCAS to the demand for FCAS. Just like in the market for power, AEMO orders the FCAS bids into a “merit order” based on price for each service type (lowest price to highest price). It then decides which bids it will accept and pays all accepted bids the highest marginal price of all accepted bids. This creates a price signal for FCAS in the market.

The design has obvious intentions: FCAS suppliers win big if they can be super low cost while their competitors are high cost. 

Just like in the market for power, once bids are accepted, AEMO sends a signal to the control panels of the FCAS participants. 

You can see how , by following this process, AEMO acts to approximate the emergent order (or invisible hand) of a market by creating price signals for both power supply as well as FCAS. Then, physically matching those price signals with the actual power flows using a (pretty amazing, honestly) control scheme. 

The idea that AEMO is only approximating the invisible hand is important; I’ll explore that more deeply in future articles.

That’s the physics and market mechanisms in a nutshell. But who sells FCAS to AEMO, and what about the mullah, dollars, buckerinos, fat stacks?

Part tres: How big is the market

Firstly, who can participate in the FCAS market? As far as I understand, it’s quite open; if you invent a device that meets the standards and go through the application process, then you can participate. AEMO’s concern is system stability, so from their POV, any particular device is fair game as long as you meet the engineering standards. 

One common example is certain electricity retailers who use their customers’ hot water cylinders as a demand response. By sending an electrical signal to your hot water cylinder, the company can turn it off for a short period, reducing demand (load). There’s a complex “annoyance cost” paid by the consumer of their hot water cylinder being turned off, but then that’s balanced against potentially lower bills as the electricity retailer can substitute FCAS revenue. 

Another example is grid-scale batteries; these can build or release charge on a whim. For batteries, the incentives are about whether their charge is worth more in the spot market or in the supply response FCAS market. If the battery is connected to a facility like a solar plant, then the incentives get more complex; is it worth selling power instantly or building charge in the battery for potential spot market or FCAS later? Add in a futures market and you just get more complexity.

A final, still emerging, example is bitcoin miners, these are basically just computers. These machines can be powered up or down relatively quickly and easily with little ill physical effects, and no ill effects to the Bitcoin network or protocol. When they are powered up, they earn a stream of income for the owners in the form of bitcoin.

This creates an interesting dynamic when bitcoin miners participate in the FCAS market; their FCAS bid, all else equal, should be at least the value of the opportunity cost of not mining bitcoin.

A completely unanswered question is what happens if bitcoin mining becomes a large proportion of FCAS? Consider a case where the FCAS from bitcoin miners drops out of the market because bitcoin fees spike and miners find it more profitable to mine rather than provide FCAS.

Alright, but the money! Surely FCAS is lucrative?

AEMO publishes data on FCAS. The most readily understandable data is the payment data. This is, as the name suggests, how much AEMO paid to FCAS participants over the eight service types (and in each State/Territory in the NEM) each month. This is the raw revenue of FCAS market participants. One way to think of it is as the cost of stability. 

In total, FCAS payments have been about $39 million this year (up to week 20, which ended May 18). 

AEMO has also provided total revenue for FCAS from batteries in the NEM, specifically in its Quarterly Energy Dynamics - Q1 2025, covering 1 January to 31 March 2025 report. This data suggests that batteries in the NEM have earned $69 million in 2024, which is 42.9 percent of all FCAS costs. This is remarkable to me, because grid-scale batteries are relatively new to the market as compared to other generation and storage technologies. 

The cause lies in the fact that batteries can* *have a low marginal cost to charge and discharge, so the fact that they captured such market share this fast highlights the idea that he who produces FCAS cheapest wins.

This data release also contains aggregate FCAS costs for all states. The data tells us that in 2024, total FCAS costs were $161 million.

The fact that there is such a healthy pot up for grabs speaks to the fragility of our grid, but also to the resilience; it’s a strange duality.

With continued investment in solar generation, because such systems have little to no inertia, we could see increased frequency excursions (technical term for being out of range). Engineers are busy solving this using inverters, which simulate inertia. These are genius, but such innovations take time (and standards) to be adopted.

Then there’s the question of critical infrastructure risk - how tolerant can we be of instability when datacenters mining Bitcoin and running AI models are on the line

At the risk of repeating myself, the whole construction is very much a visible hand trying to approximate what Adam Smith’s invisible hand might do. 

Fin

Full disclosure, I do use AI (Gemini specifically) to help me research, learn, and write my articles, but I go over everything multiple times to make it my own voice and reflect my own ideas.

*Adam Smith described the idea of an invisible hand, which directs the market, coordinating people’s desires, as a metaphor to explain how people in a market somehow end up serving each other’s best interests even without directly communicating.