They built a palace on a fault line. The palace is the sprawling network of Bitcoin mining operations across the United States. The fault line is the U.S. Energy Information Administration’s recent forecast that electricity demand will hit new highs by 2026—driven by the same two industries: artificial intelligence and cryptocurrency mining. The code of the grid is not a smart contract. It is a physical system with hard constraints. And the logic of that system is about to demand a price.
I have spent the last two years auditing mining operations. I watched the economics shift from cheap Texas wind power to the reality of interconnection queues and transmission bottlenecks. The EIA forecast is not a surprise. It is a confirmation. The climate of cheap U.S. electricity for mining is closing. The question is not if, but how many miners will be squeezed out before the next halving.
Context: The Data That Cannot Be Ignored
On January 18, 2026, the EIA released its Annual Energy Outlook. The headlines were bland: "U.S. electricity demand expected to rise 20% by 2030." The footnote was the bomb: cryptocurrency mining and AI data centers together accounted for nearly 60% of new load growth in certain regions—specifically the PJM Interconnection, which covers 13 states and Washington D.C. PJM is the largest wholesale electricity market in the world. It is also home to roughly 35% of U.S. Bitcoin hashrate. The correlation is not coincidental. It is structural.
Mining rigs are flexible loads. They can be turned on and off in seconds. That makes them ideal for grid balancing—when there is excess power, they soak it up; when demand spikes, they shut down. But that flexibility comes with a cost. Miners sign long-term power purchase agreements (PPAs) at fixed prices, or they buy on the spot market. In a world where demand is rising faster than supply, spot prices will trend upward. Fixed PPAs will be renegotiated at higher rates. The era of sub-$0.03/kWh industrial rates in the U.S. is ending.
Data does not lie, but it does not care. The EIA forecast is a cold, clinical extrapolation of existing trends. It does not account for regulatory interventions. It does not model the possibility of state-level bans on proof-of-work mining. It only says: more demand is coming. The market will clear through price. And miners, being the most price-sensitive industrial consumers, will be the first to be cut off.
Core: The Systematic Teardown of U.S. Mining Economics
Let me deconstruct the balance sheet of a representative American mining operation. Assume a fleet of Antminer S21s, each consuming 3500 watts and hashing at 200 TH/s. At $0.05/kWh—still considered cheap by industrial standards—daily electricity cost per unit is $4.20. At current Bitcoin price of $65,000 and network difficulty of 55 trillion, each machine earns roughly $8.50 per day in gross revenue. Gross profit: $4.30. After hosting fees, maintenance, and hardware depreciation, net profit per unit is around $2.00 per day. That is a healthy margin—about 25% net.
Now move to $0.08/kWh. That is the current average industrial rate in PJM during off-peak hours. Daily electricity cost jumps to $6.72. Gross profit falls to $1.78. Net profit shrinks to near zero after overhead. At $0.10/kWh, the operation is cash flow negative.
The EIA forecast implies that in regions like PJM, the average industrial rate could climb to $0.09-$0.12/kWh by 2027, driven by demand from AI data centers willing to pay premiums for firm power. Miners, who are interruptible loads, will be forced to accept higher rates or be curtailed. The result: a wave of miner migration out of PJM to regions with lower rates—like the Midwest (MISO market) or the Pacific Northwest (underutilized hydro). But those regions also face demand growth from AI. There is no safe harbor.
In my audit of three Texas-based mining facilities in 2024, I saw this first-hand. One operator signed a 5-year PPA at $0.035/kWh with a wind farm. By 2025, the wind farm was selling into the ERCOT spot market at $0.12/kWh during peak hours. The miner was paid to curtail—demand response programs exist. But the average cost across the year crept up to $0.055/kWh. That facility is now considering relocation to Argentina. The same story is playing out across the industry.
The core insight: U.S. mining is not a technology business. It is an energy arbitrage business. The arbitrage window is closing because the marginal buyer of electricity is no longer the miner. It is the AI data center, which can justify costs of $0.15/kWh or more. The miner is the swing producer. And in any commodity market, the swing producer is the first to be squeezed.
Trust is a variable you cannot hardcode. Miners trusted that U.S. electricity would remain cheap due to abundant natural gas and renewables. They built multi-hundred-megawatt facilities on that trust. The grid does not care about trust. It only follows the laws of physics and marginal cost.
Contrarian: What the Bulls Got Right
Not all is lost. The contrarian view—the one that will likely be repeated by mining CEOs and public relations teams—is that miners are not passive victims. They are active participants in grid reliability. The demand response programs that paid miners to shut down in 2024 are expanding. The Federal Energy Regulatory Commission (FERC) Order 2222 explicitly allows distributed energy resources—including mining rigs—to participate in wholesale markets. Miners can earn revenue by providing frequency regulation and reactive power support. This is not theoretical. I have seen a facility in upstate New York that earns $0.02/kWh just by being available to curtail within 5 seconds. That reduces effective power cost.
Furthermore, the narrative that AI will steal all the cheap power ignores the physical reality of transmission constraints. AI data centers need to be near fiber optic backbone and low latency networks. Miners can be anywhere with stranded energy: flared natural gas, remote hydro, orphaned wind farms. The economic sweet spot for mining is not next to a high-voltage substation in a dense urban area. It is in the middle of nowhere, where land is cheap and renewable projects are desperate for baseload off-takers. The EIA forecast does not invalidate that model. It reinforces it.
Another bull argument: efficiency gains. The next generation of miners—like the Antminer S22 or the MicroBT M70—boast efficiency improvements of 15-20% per generation. If Bitcoin price doubles over the next two years (a plausible scenario given ETF inflows and institutional adoption), then even at $0.10/kWh, mining remains profitable. The elasticity of demand for electricity in mining is high in the short term but low in the long term because hardware is a sunk cost. Miners will not shut down immediately. They will operate at thin margins until the next bear market washes them out.
But this is exactly the point. The bull case assumes continued exponential growth in Bitcoin price. That is not an assumption a cold dissector can rely on. Price is a variable. Power cost is a quasi-fixed cost once the facility is built. The risk is asymmetric: if price stays flat or drops, the miner with high power costs is dead. The miner with low power costs survives. The EIA forecast tilts the playing field against the majority of U.S. miners.
Takeaway: The Accountability Call
The EIA forecast is not a short-term catalyst. It is a slow-moving glacier. But glaciers carve canyons. The canyon here is the reconfiguration of the global Bitcoin hashrate map. Over the next three years, I predict that U.S. hashrate share will drop from roughly 35% to below 25%. The winners will be miners in the Middle East (low-cost associated gas), Scandinavia (hydro), and Latin America (stranded renewables). The losers will be publicly traded U.S. miners with high leverage and fixed power contracts that cannot be renegotiated.
This is not an investment thesis. It is an audit finding. The capital being poured into U.S. mining today is building infrastructure on a fault line. When the next bear market comes—and it always comes—the cost of power will be the variable that breaks the balance sheet. The smart money is already signing PPAs abroad. The rest are hoping the grid will remain forgiving.
Data does not lie, but it does not care. Neither should you.
