Every five minutes in Australia’s electricity market, prices are recalculated. Power generators submit bids. Supply and demand are matched. Increasingly, this process is shaped not just by human traders but by algorithms.
In its 2021 article in the UNSW Law Journal, “Algorithmic Collusion and Australian Competition Law” explores how pricing software and artificial intelligence are reshaping competition and why markets like electricity provide a powerful example of both the benefits and the risks.
The Australian electricity market, particularly the Australian Energy Market Operator run National Electricity Market, is already highly automated. Generators bid electricity into the market at different price bands. The lowest cost supply is dispatched first. Prices fluctuate rapidly based on demand, outages, weather and network constraints.
In theory, this dynamic environment should encourage competition. But it also has characteristics that make it fertile ground for algorithmic coordination:
A small number of major generators
High transparency (bids and prices are visible)
Repeated interaction every few minutes
Heavy reliance on automated bidding systems
The UNSW article below explains that algorithms operating in such a setting can learn patterns over time. If aggressive price cutting triggers retaliatory undercutting from competitors, an algorithm may “learn” that restraint is more profitable. Without ever being instructed to collude, it may adopt bidding strategies that avoid destabilising high price outcomes.
In an electricity market, that could look like this:
A generator’s software observes that when it bids aggressively low, rival firms respond in the next interval by undercutting further, driving prices down across the board. But when all generators bid at higher levels during peak demand, profits rise for everyone. Over thousands of iterations, algorithms may converge on strategies that sustain elevated prices without any explicit agreement.
No phone calls. No secret meetings. Just machines responding rationally to market signals.
The article distinguishes between deliberate misconduct where firms intentionally design algorithms to coordinate and autonomous algorithmic collusion, where pricing alignment emerges organically from machine learning systems. The electricity market helps illustrate why the latter is so concerning. Because prices are set so frequently and data is so transparent, algorithms can adapt faster than humans ever could.
Australian competition law traditionally targets agreements or “concerted practices” between competitors. But in a highly automated electricity market, what counts as an agreement? If pricing alignment is the product of code interacting with code, rather than executives agreeing to fix prices, proving cartel conduct becomes far more complex.
The authors argue that electricity markets demonstrate both the sophistication and vulnerability of modern competition systems. On one hand, algorithmic bidding improves efficiency and responsiveness. On the other, it may increase the risk of subtle, durable coordination in concentrated markets particularly during periods of peak demand when prices can spike dramatically.
For consumers, the stakes are tangible. Electricity prices flow directly into household bills and industrial costs. If algorithms sustain higher than competitive prices, the impact is not abstract it appears on monthly statements.
The paper ultimately calls for regulators to adapt. Agencies must understand how bidding software operates, develop technical expertise, and refine evidentiary approaches. The law may need to focus less on proving a “meeting of minds” and more on analysing market outcomes and strategic design choices embedded in code.
The electricity market is not evidence of widespread unlawful collusion. But it is a vivid case study of how automated systems can reshape competitive dynamics. As firms increasingly rely on AI driven pricing, whether in energy, airlines, retail, or digital advertising, the question becomes urgent:
When machines learn that cooperation is profitable, who is responsible, and is the law equipped to respond?
44 pages.