Turning a miner's power down cuts the electricity bill. Turn it down to the limit and the hashrate goes to zero. Somewhere between those two points sits the setting that earns a farm the most money, and in an immersion system that setting depends on many variables at once: the price of Bitcoin, the cost of power, the temperature outside, and the temperatures, fan speeds and pump speeds inside each tank. A handful of miners can be tuned by hand. A farm with thousands of them cannot, and that is the problem Arctic Systems set out to address.
Mike Nixon, CFO of Arctic Systems, an immersion mining systems developer, made this case at Mining Disrupt Miami 2023, held July 25–27, 2023. Bitcoin was trading at $29,144 at the time. His talk grew out of conversations with Arctic's clients, and its premise was that as immersion mining becomes more common, the number of decisions needed to keep a farm profitable starts to overwhelm the people making them. His answer was to pair artificial intelligence and machine learning with robotic process automation.
The Overwhelm of Immersion Variables
Nixon opened by acknowledging the scale of the problem. While a small operation with one or two miners could be managed through manual adjustments, the complexity exploded as the number of units grew. The goal for any mining farm is straightforward: maximize income while minimizing expense. However, Nixon noted that both income and expense are volatile. Income fluctuates with the price of Bitcoin, while expenses fluctuate with energy costs and operational parameters. This creates a dynamic equation where the optimal strategy changes constantly.
As immersion mining is becoming more and more prevalent, the decisions that you have to make in immersion start to overwhelm.
Mike Nixon · 0:35
Nixon was plain that immersion is more complicated than an air-cooled system, so it had better deliver additional benefits in return, and those benefits have to hold up when the goal is the most profit at every moment. Operators must manage ambient temperatures, heat rejection rates, and seasonal performance differences. Inside the tanks, variables include inlet and outlet temperatures, fan speeds, and pump speeds. Each of these elements affects the miner’s efficiency, and each must be balanced against the others to achieve the desired key performance indicators.
Profit as a Real-Time Calculation
Nixon framed profit as a mathematical problem that plays out in the present tense. Profit is income minus expense, but both move with time. Bitcoin's volatility pushes operators to earn in shorter and shorter windows, and earlier speakers that day had described curtailment and swings in power pricing, which he expected to matter more and more to a farm's performance. In his framing, total profit over a month or a year becomes a question of what the profit is at any instant, given every parameter in play.
At each point in time, the goal is to maximize both income and at the same time minimize expense in the process.
Mike Nixon · 2:49
That instant can change quickly. Nixon said many of these client conversations began in the November–December low in Bitcoin's price, when the question was whether to overclock or underclock. Most operators chose to underclock: they wanted a return on their hardware, but not to burn so much power that a low Bitcoin price left them running at a loss. Power consumption depends on how hard the miners are pushed, so the decision also needs feedback on hashrate and, on the sales side, the real-time price of Bitcoin.
The Data Gap in Immersion Systems
The core of Nixon’s argument was that optimization is impossible without data. He listed the specific metrics that an immersion system must track: environmental information, ambient temperatures, heat rejection rates, and forecasts for day-night temperature profiles. Beyond the environment, the system must monitor the coolant performance, including tank temperatures and flow rates. It must also track the power consumption and hashrate of each individual miner. Finally, it needs to ingest real-time Bitcoin pricing data. When these data streams are siloed or missing, operators cannot see the full picture.
If you're not collecting the data, you don't really know what you're optimizing.
Mike Nixon · 4:46
Nixon called these conflicting variables. Cutting power cuts cost, but taken to its limit it takes hashrate to zero, so the optimum lies somewhere in between. Each variable can be optimized in its own sphere; what matters, he argued, is optimizing the system as a whole. And as the time frame shrinks toward real time, managing the equation becomes an adjustment problem as well as an analysis problem.
Learning from LED Innovation
To illustrate the importance of system-level thinking, Nixon shared an anecdote from his early career. As a graduate student in the 70s, he worked in a lab at the University of Illinois under a professor named Nick Holonyak, who was known for inventing the light-emitting diode and the first solid-state laser. When Nixon was there, the red LED already existed and the green was a newly developed process. Nixon recalled his professor’s advice: always think of the system.
You can optimize individual variables, but if you're not looking at the system as a whole, you can lose sight of the overall vision.
Mike Nixon · 8:39
Nixon said Arctic sees this with some of its own clients in Bitcoin mining. They tune one parameter at a time, yet the farm as a whole is left unmanaged, and he argued that only a new kind of tool can tune everything together.
AI and Robotic Process Automation
Arctic Systems’ response to this challenge was a two-part technological investment. The first was the application of artificial intelligence and machine learning to analyze the control variables. These algorithms process the inputs—such as inlet and outlet temperatures, fan and pump speeds, and power supply settings—to determine the optimal configuration for maximizing profit. The second part was the implementation of robotic process automation, which allows the system to execute these adjustments automatically across large-scale operations.
Nixon put the scale problem simply: it is easy to control one machine, but a farm with thousands of machines needs new tools. Robotic process automation lets Arctic control entire systems automatically. The work reaches down to the firmware, overclocking and underclocking miners through control of their power supplies, a topic the Alpha Miner speaker before him had just covered, and Nixon said Arctic was seeing very promising results. He added that the company was looking at how to bring management consoles and firmware into the same system.
We believe that the future of that is not only the combination of analyzing the variables with artificial intelligence and machine learning, but also to incorporate the adjustment of the entire system in real time
Mike Nixon · 10:34
Bitcoin was at $29,144 when Nixon spoke in Miami; on October 16, 2026, it was $82,165. He ran over his time slot and closed by inviting operators with questions about their projects to find Arctic Systems at the conference.