Las Vegas is behind me. Three days, one panel, 24 scheduled meetings, and more conversations on the floor than I can count. The Bitcoin Conference 2026 felt more measured than previous years, which makes sense with Bitcoin still below $100K, but in many ways that’s when the real signal emerges. Less noise, more substance. What stood out most wasn’t just the deal flow, but a structural shift that’s now impossible to ignore: AI and HPC are no longer a side conversation in mining circles. The Mining Stage was rebranded to the Energy Stage this year, and that reframing says it all. This industry is repositioning itself from a Bitcoin niche into a broader energy-backed digital infrastructure stack.
The sessions I moderated went straight to the heart of an important efficiency question: Squeezing Profit from the Margins: Uncovering Stranded Sats in Mining Operations. This week’s newsletter pulls the key insights from that panel discussion on firmware tuning, operational discipline, and cost control in a sub-$40 hashprice environment.
- The Margin Problem Hiding in Plain Sight
- Firmware & Hashrate Tuning
- Operational Efficiency
- Cost Control & Margin Discipline
The Margin Problem Hiding in Plain Sight
Hashprice has been below $40/PH per day for most of 2026. At those levels, efficiency is no longer a competitive differentiator. It is the minimum requirement for staying in business.
The industry response has been fleet upgrades, relocation of hardware to districts with cheaper power, and for the public miners a rush toward AI/HPC as an alternative revenue stream. Those are legitimate strategies. But they all share the common assumption that the path to better margins runs through new capital deployment.
What got less attention at the conference, and what this panel was explicitly designed to address, is the opportunity that already exists inside current operations. The sat leakage from suboptimal firmware, poor uptime discipline, and avoidable inefficiencies is real and in a compressed-margin environment, it compounds quickly.
The three panelists — Dan Koehler (FARMGOD), Bradley Peak (VNISH), and Alexander Lozben (Interhash) — brought a combined perspective from fleet management software, firmware development, and large-scale mining operations. The conversation surfaced a consistent theme there are many opportunities in getting more out of what they already have. Each percentage point of improvement in uptime, efficiency, or cost structure does not just add linearly, it compounds across the full fleet, across every operating hour.
Firmware & Hashrate Tuning
Stock firmware is designed for broad compatibility, not peak performance. That distinction matters more than most operators realize. Manufacturer settings are built to work across a wide range of environments, which means they are inherently conservative on voltage, frequency, and thermal targets.
Custom firmware changes all of that. By adjusting voltage profiles, frequency scaling, and thermal ceilings at the chip level, operators can meaningfully shift the efficiency curve of an existing machine without replacing it. The question is not just how much hashrate you produce, but how many joules you spend producing it.
In a low-margin environment, firmware is alpha. It is cheap, fast to deploy, and has an immediate impact on your P&L.
One of the most important applications in today’s environment is underclocking, deliberately reducing power draw and hashrate to improve J/TH ratios. The calculus here is not intuitive. Producing less hashrate at meaningfully lower power consumption can generate more profit per machine.
A more complex question is what role custom firmware plays as next-generation ASICs come equipped with built-in eco and turbo modes. The answer is not that firmware becomes obsolete. Stock eco modes are still generalized. Custom tuning can push those profiles further, adapt in real time to facility conditions, and integrate with broader operational logic in ways that manufacturer defaults simply do not support.
Pool selection is another variable that tends to receive less attention than it deserves. The choice of mining pool directly affects actual revenue per PH. In many cases miners still choose the pool based on headline fee structures and do not analyse actual pay-out performance sufficiently.
Operational Efficiency
Uptime is one of the most directly controllable variables in mining economics, every hour a machine is offline is revenue that cannot be recovered. A facility running at 95% uptime versus 98% uptime is leaving a material amount of BTC on the table.
The challenge is that downtime is rarely caused by a single failure category. Effective management software is crucial in tracking these at the facility level, identify failure patterns before they cascade, and flag anomalies early enough to act on them.
The repair-versus-retire decision was also discussed. This is where operational discipline and market cycle awareness intersect. A machine that makes economic sense to repair at $60/PH/day may not at $35/PH/day. Not because the repair cost changed, but because the opportunity cost of tying up capital and rack space in a marginal unit change. In the current environment, the default toward repair optimism needs to be replaced by a more rigorous ROI framework. What will this machine actually earn over its remaining productive life, net of the repair cost and turnaround time?
Repair only when the economics make sense. Long turnaround times and high component costs can quickly turn repairs into hidden losses.
Custom firmware also provides options to reduce thermal stress and component longevity across different tuning profiles. The warranty question came up, and the panel's position was clear: voiding manufacturer warranty to deploy custom firmware is not a difficult trade-off, it's a straightforward call. Manufacturer warranty processes are notoriously slow, and in practice, the downtime and administrative friction involved often makes them less valuable than they appear on paper. When custom firmware can materially improve chip performance and J/TH ratios immediately, waiting on a warranty claim is rarely the better option.
Cost Control & Margin Discipline
The hydro versus air-cooled debate continues to be relevant in current market conditions. Hydro deployments tend to demonstrate lower failure rates and better thermal consistency, which translates to more predictable uptime and lower maintenance overhead over time. But with ASIC prices at current levels and many operators under margin pressure, the lower CapEx and faster payback of air-cooled infrastructure still drive the decision for many investors. Whether to opt for a hydro-cooled or air-cooled deployment is context-dependent.
Strategic curtailment and demand-response participation are increasingly a core part of the operational toolkit rather than an optional add-on. Miners do not curtail because they have flexible loads, they curtail because the economics are compelling. When power prices spike during peak demand periods, shutting down is not losing revenue; it is harvesting the value of flexibility. Management software is crucial in the ability of evaluating those decisions in real time, integrating live power pricing signals, and executing curtailment logic without requiring manual intervention.
Strategic curtailment is not losing revenue. It is harvesting the value of flexibility and that flexibility has a real market price.
Firmware plays a direct role here as well. Dynamic load management, ramping machines up or down in response to real-time power pricing or grid signals requires firmware that can respond quickly and predictably. The coordination between software-level logic and firmware-level execution is where a lot of the theoretical value of demand response is either captured or lost in practice.
Beyond energy, the full cost stack includes facility overhead, personnel, insurance, financing costs, and hardware depreciation. When margins are this thin, any one of those line items can represent the difference between profitability and loss. The operators finding the most leverage are often not just attacking the biggest cost item, but they are systematically reducing the second and third largest items after power rates.
The integration of GPU-based AI workloads alongside Bitcoin mining is beginning to demonstrate a viable path to margin diversification for operators with the infrastructure to support it. Bitcoin mining and AI compute share the same physical foundation: power, cooling, and connectivity. They are different load profiles running on similar systems. For facilities with available rack space, predictable power, and the operational capacity to manage additional complexity, adding GPU workloads can materially improve revenue per megawatt.
The closing takeaway from the panel was consistent across all three speakers: stranded sats are real, and the solution is often cheaper than operators assume. Whether it is firmware that stretches every joule, operational discipline that keeps machines hashing, or cost structures built to survive multiple cycles, the best operators are compounding marginal advantages into material returns. The path forward is not one big move but a series of percentage-point improvements, each one stacking on the last.
You don't lose money because the model was wrong. You lose money because reality doesn't follow the model. Execution risk is the variable most investors underestimate and the one we specialise in. Book a call today and let's pressure-test the plan.
