For seven issues, I've been examining the heuristics that analysts use to trace Bitcoin transactions — what they assume, where they break, and why it matters. In Issue 7, I argued that the most important factor in real-world Bitcoin privacy isn't cryptography but economics: most transactions are private because nobody will spend the money to trace them.
This issue is a deliberate pivot. Instead of forensic methods, I want to examine something more fundamental about Bitcoin's design — the relationship between energy and scarcity. Specifically, I want to stress-test a claim that comes up constantly in Bitcoin discourse: that proof-of-work "converts" energy into monetary value.
The thought experiment I'll use is extreme. What happens to Bitcoin if a civilization has access to essentially unlimited energy? The answer reveals something about Bitcoin's design that I think is genuinely underappreciated — and it centers on a mechanism that rarely gets the attention it deserves.
What mining actually costs right now
Before we can reason about energy abundance, we need to understand what energy scarcity looks like in practice. The numbers are stark.
The April 2024 halving cut Bitcoin's block reward from 6.25 to 3.125 BTC — roughly 450 BTC per day. JPMorgan's mining research team (Panigirtzoglou et al., January 2026) estimated average industry production costs at $77,000–$92,000 per BTC by early 2026, including hardware depreciation, energy, cooling, and overhead. The Cambridge Digital Mining Industry Report (Neumueller et al., April 2025), a survey-based study covering 49 mining firms representing 48% of global hashrate, puts the picture more granularly: median electricity cost of $45/MWh, with all-in operational costs ranging from roughly $14,000 per BTC in the lowest quartile to $36,000 in the median. With Bitcoin trading around $67,000–$69,000 in late February 2026, the largest miners remain profitable on a cash basis, while smaller and less efficient operations are not.
The variance across operators is enormous. Marathon Digital, the largest public miner at roughly 57–60 EH/s, reported energy-only costs of about $33,735 per BTC in its Q2 2025 earnings, with fleet efficiency of 18.3 J/TH. Riot Platforms reported roughly $48,992 per BTC excluding depreciation in the same period. CleanSpark, mining in Georgia with wholesale rates as low as $0.012/kWh, achieved direct production costs in the $40,000–$45,000 range for FY2025. At the extremes, data compiled by CoinGecko and BestBrokers shows Iran's subsidized electricity enables mining at roughly $1,324 per BTC, while Italy and Ireland push costs above $300,000.
This raises an obvious question: if the average miner is underwater on an all-in basis, why are they still operating? The answer lies in a basic microeconomic distinction. Hardware costs are sunk once purchased — miners continue operating as long as revenue covers variable costs, primarily electricity. Depreciation is an accounting figure, not a cash outflow that can be avoided by shutting down. Fu (2020, IACR ePrint) models this formally as a real options problem: even negative all-in margins don't trigger shutdown if expected future revenue exceeds restart costs. Derks, Gordijn, and Siegmann (2018, Electronic Markets) provide empirical confirmation that mining profits converge to zero via competitive entry and exit — with exit occurring at the variable cost floor, not the accounting breakeven.
Energy dominates everything. It accounts for 60–80% of operational expenses across the industry. JPMorgan found that each $0.01/kWh change in electricity price moves production cost by roughly $18,000 per BTC — a figure that incorporates not just raw power draw but electricity-correlated overhead costs (cooling, grid infrastructure, demand charges). Raw power consumption alone at current network difficulty runs approximately 1,600,000 kWh per BTC for efficient operations, implying a baseline sensitivity closer to $16,000 per BTC; the $18,000 figure reflects JPMorgan's broader model including overhead. Hardware costs have compressed — ASIC pricing fell from peaks of $80–$120/TH in late 2021 to roughly $15–$16/TH for efficient machines in 2025 (per Hashrate Index data) — but depreciation remains massive.
Here's the counterintuitive part. Despite the halving cutting revenue in half, global hashrate surged from roughly 505 EH/s pre-halving to over 1,000 EH/s by early 2026. The network hit an instantaneous peak of 1 zettahash on April 4, 2025 — a thousandfold increase from when it first hit 1 EH/s in 2016 — though the 7-day moving average didn't reach that level until September 2025. Mining difficulty first crossed 150 trillion in October 2025, then set an all-time high of 156 trillion in November.
The result has been severe concentration. Foundry USA and AntPool together now mine approximately 50–55% of all blocks, with just six pools collectively responsible for 96–99% of block production (per b10c.me's 2025 centralization analysis). When accounting for proxy pooling arrangements — where Bitmain-affiliated entities share infrastructure — effective concentration reaches an estimated 60–70%. This level of concentration is not merely a business story. Eyal and Sirer (2014, FC) demonstrated that selfish mining becomes profitable for any pool controlling more than 33% of hashrate under adversarial network conditions, meaning the top-2 pools are already past this threshold. Cong, He, and Li (2021, Review of Financial Studies 34(3)) model pool equilibrium dynamics in "Decentralized Mining in Centralized Pools," finding that larger pools charge higher fees and attract miners disproportionately slowly — the centralization pressure comes not from pool advantage but from the subsidy-dominated fee environment reducing the penalty for miners staying in concentrated pools.
A parallel concentration risk sits upstream in hardware manufacturing. Three Chinese firms — Bitmain, MicroBT, and Canaan — produce over 99% of all Bitcoin ASICs globally, with Bitmain estimated at approximately 82% market share alone (Cambridge Digital Mining Industry Report, 2025). Bitmain simultaneously operates AntPool, one of the two dominant mining pools, creating vertical integration across the supply chain. All leading ASIC chips are fabricated at TSMC in Taiwan or Samsung in South Korea, adding a semiconductor choke point. The academic literature has not formally modeled ASIC supply-chain concentration as a systemic risk — a gap that the pool-centralization literature (extensive) has not replicated for hardware (essentially absent).
More energy in. Same coins out. This is the pattern that makes Bitcoin unlike anything else in economics, and it's entirely the product of one mechanism.
The thermostat — and its limits
The difficulty adjustment is, in my view, the single most important innovation in Bitcoin's design. More important than proof-of-work itself, more important than the UTXO model, more important than the 21 million cap — because without the difficulty adjustment, none of those other features would produce the monetary properties that make Bitcoin interesting.
The mechanism is simple. Every 2,016 blocks — roughly two weeks — the protocol recalculates the mining target. If miners found those blocks faster than expected, difficulty increases. If slower, it decreases. A safety clamp caps any single adjustment at 4×. The formula is straightforward: new difficulty equals old difficulty multiplied by the ratio of actual time to expected time, where expected time is exactly 1,209,600 seconds (14 days).
The economic consequence is what matters. When Bitcoin's price rises, more miners enter, hashrate increases, and blocks temporarily arrive faster than every 10 minutes. Then difficulty adjusts upward, and block production returns to roughly 144 per day. The same 450 BTC are produced whether the network consumes 1 GW or 100 GW.
More energy produces more security. Never more coins.
Josh Hendrickson, Senior Fellow at the Bitcoin Policy Institute and Professor of Economics at the University of Mississippi, described the difficulty adjustment in December 2024 as "the innovation, and arguably the genius, of the design of Bitcoin." Saifedean Ammous puts it more bluntly in The Bitcoin Standard (2018): Bitcoin is the only asset in the world where supply is completely inelastic to demand. Garay, Kiayias, and Leonardos (2017, CRYPTO) provided the first formal cryptographic analysis of the variable-difficulty mechanism, proving that the protocol maintains consistency and liveness even as miner populations change — the underlying security guarantee that makes the economic analysis meaningful.
The most dramatic stress test came in May–July 2021, when China's crackdown on mining caused network hashrate to fall by approximately 50% — a collapse unprecedented in Bitcoin's history. Multiple consecutive downward adjustments followed, with one single epoch dropping roughly 28%. Block times slowed temporarily, but the mechanism performed exactly as designed: difficulty fell, remaining miners became more profitable, new miners entered, and by early 2022 hashrate exceeded pre-ban levels. (China's share of global hashrate had already declined from its peak of 65–75% in 2019–2020 to around 46% by April 2021 — the "over 60%" figure that circulates in Bitcoin discourse reflects earlier dominance, not share at the time of the ban.)
A smaller but notable test occurred in late January 2026, when Winter Storm Fern curtailed US mining operations — MARA alone powered down roughly 70% of its global hashrate — and difficulty dropped 11.16% to 125.86 trillion, the largest single negative adjustment since the 2021 ban, according to The Block. Within days, hashrate was already recovering.
The 4× cap has never been seriously tested. The closest approach came during the China ban itself, when hashrate dropped roughly 50% and difficulty fell approximately 28% across several epochs — still well within the cap's 75% single-epoch limit. The cap protects against catastrophic, instantaneous hashrate loss, but the more relevant stability constraint is the two-week adjustment window: during any period between adjustments, a sudden hashrate departure leaves block times proportionally slower, with no mechanism for faster correction.
The historical record is striking. Hashrate has grown from literally zero — Satoshi's CPU — through kilohash, megahash, gigahash, terahash, exahash, and into zettahash territory. That's a span of roughly 10²¹. Block production has never deviated meaningfully from 144 per day. Rudd and Porter (2025), writing in the Journal of Risk and Financial Management, model Bitcoin as exhibiting a perfectly inelastic supply curve at each time step, and the data supports this framing for the total cumulative supply schedule.
But "perfectly inelastic" deserves some qualification. Within a single 2,016-block epoch, deviations are real. During the China ban, block times stretched to 14–19 minutes when half the network disappeared. During hashrate surges, blocks temporarily arrive every 8–9 minutes. These within-epoch timing deviations explain why halvings have occurred slightly ahead of their theoretical schedule. The total supply remains absolutely fixed; the issuance rate is approximately fixed but not exactly so between adjustment periods.
More significantly, the formal economics literature has found a deeper vulnerability. Noda, Okumura, and Hashimoto, in a paper published in the International Economic Review (DOI: 10.1111/iere.70028), apply what they call a "Lucas Critique" to Bitcoin's difficulty adjustment: the mechanism is backward-looking, using the previous epoch's block times to set the next epoch's target. It doesn't anticipate how miners will respond. Their analysis proves the DAA becomes dynamically unstable — failing to converge rather than merely converging slowly — when the reward elasticity of hash supply exceeds 1 (when hashrate is highly responsive to reward changes). Their simulations show Bitcoin approached but did not cross this threshold during the November 2018 market crash, when per-hash reward hit its historical minimum. The implication is pointed: under severe price stress simultaneously reducing mining reward, the DAA could produce divergent block times rather than smooth convergence. The authors note that Bitcoin Cash's ASERT algorithm is formally stable under substantially weaker conditions — an elasticity threshold of 575 versus Bitcoin's 1 — raising a question Bitcoin developers have not publicly resolved.
None of this undermines the thermostat's elegance in normal conditions. But the article would be incomplete without acknowledging that formal analysis has identified real boundaries on its robustness — and that a known protocol exploit called the "timewarp attack" remains unpatched. First identified around 2011, the timewarp attack would allow a majority miner to manipulate block timestamps across epoch boundaries to artificially suppress difficulty. Poinsot's analysis for BIP 54 (Great Consensus Cleanup, April 2025) specifies that an attacker with majority hashrate could reduce difficulty to near-1 within 38 days. BIP 54 has been implemented on Bitcoin Inquisition's signet and has broad developer support, but has no mainnet activation timeline.
Is Bitcoin "stored energy"?
This is where I want to be careful, because the popular framing and the defensible framing are different things — and because the academic literature makes this harder than it looks.
The popular version comes mainly from Michael Saylor: you trade labor (energy) for money; fiat currency leaks 3–7% per year via inflation (a "bad battery"); Bitcoin, requiring real energy to create and fixed in supply, preserves purchasing power — a battery that doesn't leak. Elon Musk has echoed similar views.
The problem is that this is physically wrong. From a strict thermodynamics perspective, Bitcoin stores zero energy. When miners run SHA-256 computations, electrical energy converts irreversibly into heat. No energy remains "in" a bitcoin any more than energy remains in a solved crossword puzzle. A bitcoin is a ledger entry — pure information. You cannot extract the mining energy back.
The more defensible framing runs through Nick Szabo's concept of "unforgeable costliness" — the idea that money requires objects whose production cost is genuinely high, verifiable by others, and impossible to fake cheaply. Szabo articulated this in his writings on Bit Gold (conceived 1998, published 2005) and in "Shelling Out: The Origins of Money" (2002). Adam Back's Hashcash (1997) provided the technical implementation, tying digital tokens to real-world computational expenditure.
I want to be honest about the status of this framing, though: Szabo's "unforgeable costliness" has never appeared in a peer-reviewed economics journal. It exists in self-published essays and blog posts. The formal tradition in monetary economics points elsewhere. Kiyotaki and Wright (1989, Journal of Political Economy) demonstrate that money can emerge endogenously from search frictions using intrinsically worthless objects — production cost is neither necessary nor sufficient. Lagos and Wright (2005, JPE) ground monetary value in trading frictions and liquidity needs, with no role for production costs at all. More recent work applying search-theoretic models to cryptocurrency — Schilling and Uhlig (2019, Journal of Monetary Economics) model Bitcoin as a speculative asset with multiple equilibria; Fernández-Villaverde and Sanches (2019, JME) find private money equilibria generally fail to deliver price stability — treats production costs as economically irrelevant to fundamental value. Luther (2019, Journal of Institutional Economics), applying Kiyotaki-Wright to Bitcoin's actual historical emergence from the bitcoin-list forum, frames it as a coordination problem rather than a costliness story. Szabo's intuition is valuable, but it's an intellectual tradition, not an established economic framework.
The strongest academic critique of cost-based value theories is Thomas Umlauft's "Input Fallacy of Value" — a cognitive error analogous to the labor theory of value where participants believe mining costs determine Bitcoin's value. Umlauft, of the University of Vienna, argues that value is solely determined by utility, not by labor and capital deployed. His 2018 paper (SSRN #3182646) is worth reading, though I should note it remains an unpublished working paper rather than peer-reviewed literature.
The empirical evidence for gold supports the Umlauft direction. O'Connor, Lucey, and Baur (2016), using mine-level data from 1981–2013, found Granger causality running from gold prices to production costs — not the reverse — published in the Journal of International Financial Markets, Institutions and Money. Bitcoin's mining economics show the same pattern: costs follow price, not the other way around.
There is, however, a counterargument that the article would be incomplete without acknowledging. Adam Hayes (2017, Telematics and Informatics; 2019, Applied Economics Letters) presents a cost-of-production model showing Bitcoin's market price gravitates toward marginal mining cost. This is not a value theory — it's a competitive equilibrium argument: free entry and exit push prices toward marginal cost in any competitive industry. Hayes himself acknowledges the distinction. The finding is real and peer-reviewed: production costs act as a gravitational attractor for price in competitive equilibrium, without causing value in any fundamental sense.
The best empirical evidence for what actually does drive cryptocurrency value comes from Liu and Tsyvinski (2021, Review of Financial Studies 34(6):2689–2727). Their key finding: cryptocurrency returns are significantly exposed to network factors — proxied by total wallet addresses and active user growth — but are not exposed to production factors, proxied by hashrate and electricity cost variables. This is about as direct a test as exists, from a top finance journal, and it supports the value-from-network-effects framing over any cost-based alternative. Prat and Walter (2021, Journal of Political Economy) provide a complementary equilibrium model where the Bitcoin price determines hashrate through rational miner entry, with causation running entirely from price to production rather than the reverse.
Major Jason Lowery's 2023 MIT thesis "Softwar" pushed the energy framing furthest, arguing Bitcoin is an "electro-cyber security technology" for projecting physical force into cyberspace. The thesis attracted significant national security attention and sharp criticism. Jameson Lopp, in a detailed review, found the power-projection framing partly compelling but noted the thesis never provides a practical example of securing non-Bitcoin data, that the word "governance" appears twice while "watt" appears 149 times, and that the connection between the "why" and the "how" remains unresolved.
The synthesis position I find most defensible: Bitcoin does not store energy. The energy expenditure creates an unforgeable proof that real-world resources were irreversibly committed — Szabo's insight, regardless of its informal status. But the difficulty adjustment is what converts this expenditure into monetary scarcity. Without it, cheaper energy would simply mean faster coin production. And the value comes not from the energy itself but from the combination of absolute supply scarcity, network effects, and the practical resistance to censorship and confiscation that proof-of-work provides.
That last point — censorship resistance as a value driver — is worth flagging as underexplored. It features prominently in Bitcoin advocacy but has received almost no rigorous academic treatment. The network effects story is empirically well-supported by Liu and Tsyvinski. The censorship resistance story is intuitive but formally underdeveloped.
The aluminum precedent
Before we go to Dyson spheres, there's a historical parallel worth examining, because it shows exactly where the Bitcoin-energy analogy breaks — and where it introduces some unintended implications.
Before 1886, aluminum was rarer and more precious than gold. According to a widely repeated account — though no surviving physical evidence has been found — Napoleon III reserved aluminum tableware for honored guests while others ate with gold. What is well-documented: Napoleon III funded Deville's aluminum research, and aluminum was displayed alongside crown jewels at the 1855 Paris Exposition. Global production was about 2 tons in 1869. Then Charles Martin Hall and Paul Héroult independently discovered electrolytic reduction. In the eight years following their 1886 discovery, aluminum's price fell approximately 94% — from around $17.60/kg in 1887 to $1.10/kg by 1894. By 1900 it was under $1/kg. Aluminum became an industrial commodity.
The key input was now electricity: modern smelting requires roughly 13,000–15,000 kWh per ton, and smelters chase cheap power globally. Iceland's Kárahnjúkar hydropower project — built specifically to power Alcoa's Fjarðaál smelter — cost approximately $1.2–1.3 billion for the dam itself, with total development including the smelter and infrastructure exceeding $3 billion. Tiny Iceland now produces more aluminum than the United States.
The parallel to Bitcoin mining is immediate. Both industries are energy nomads gravitating toward the world's cheapest electricity. Both convert energy into a valuable output. But here's the critical distinction that makes the parallel ultimately misleading: when energy became cheap for aluminum, supply expanded and prices collapsed. This is true of every energy-intensive commodity — steel, cement, desalinated water, nitrogen fertilizer. Cheaper energy means more production means lower prices. Supply is elastic to energy input.
Bitcoin's supply is perfectly inelastic by protocol design, and this makes it categorically different. The difficulty adjustment ensures that cheaper energy produces more hashrate, more security, and more competition — but exactly the same number of coins.
I should note, though, that the aluminum analogy proves something slightly different from what I'm using it to prove. Aluminum's price collapse was not a disaster for aluminum — it created entirely new mass markets. Cheap aluminum built aircraft, modern cookware, skyscrapers. The falling price was the point. Bitcoin's supply inelasticity prevents a supply-side price collapse, but that's a design choice about monetary properties, not a claim about demand resilience. The analogy shows why Bitcoin is different; it doesn't show that difference is inherently valuable.
History offers instructive counterexamples. Silver was demonetized by the Coinage Act of 1873 and spent the following century losing its monetary premium despite relatively inelastic mine supply — fixed supply is a necessary but not sufficient condition for sustained value. NFTs offer a starker recent case: each token has supply of exactly 1 (perfectly inelastic), yet trading volume collapsed over 95% between early 2022 and late 2023, with a September 2023 dappGambl analysis of 73,257 NFT collections finding 95% at effectively zero market value. Bitcoin's supply inelasticity is a genuine and rare property. Whether it translates into enduring value depends on demand dynamics that supply constraints can't guarantee.
Gold comes closest among existing assets to Bitcoin's supply profile. Its stock-to-flow ratio (roughly 60 — the global above-ground stock relative to annual mine production) limits the impact of new mining on existing stockpiles. But even gold's supply responds to price over multi-year cycles as new deposits become economical. Bitcoin's supply schedule has been unaltered since genesis — 21 million coins — regardless of how much energy civilization directs at mining. Post-2024 halving, Bitcoin's S2F ratio sits around 119, roughly double gold's. That's a genuine difference of degree that approaches a difference of kind.
I want to be clear that this S2F comparison is descriptive, not predictive. Popular S2F-based price models have failed dramatically as forecasting tools — peer-reviewed analysis (Morillon & Chacon, 2022, Studies in Economics and Finance) found S2F-based trading strategies substantially underperformed buy-and-hold. More fundamentally, Burger (2020) and subsequent researchers have shown that S2F models exhibit severe serial correlation in their residuals, and a 2024 paper in Journal of Risk and Financial Management found that when time fixed-effects are included in the regression — controlling for the fact that Bitcoin's S2F is mechanically determined by time since genesis — the S2F coefficient becomes statistically insignificant (p=0.39), consistent with spurious regression. The high S2F ratio is a real property of Bitcoin. Using it to predict price is a substantially weaker claim.
Twenty-one trillion times more energy
Now let's push the logic to its extreme.
A Dyson sphere — more precisely a Dyson swarm of orbiting solar collectors — capturing the Sun's full output would harness 3.8 × 10²⁶ watts. Current global energy consumption is roughly 18.4–19.5 TW depending on the measurement methodology (direct energy vs. substitution method; Energy Institute Statistical Review, 2025). The ratio is staggering: a Dyson sphere produces approximately 20 trillion times more energy than humanity currently uses.
On the Kardashev scale, proposed by Soviet astronomer Nikolai Kardashev in 1964, this represents a Type II civilization. Humanity currently sits at roughly 0.73 — not yet Type I. Michio Kaku estimates Type I in 100–200 years, Type II in a few thousand years. Freeman Dyson himself, in his 1960 paper in Science, conceived a swarm of orbiting structures rather than a rigid shell — the "Dyson sphere" as a solid object was a science fiction invention that Dyson distanced from his original proposal.
Bitcoin mining currently draws roughly 17–24 GW continuously (Cambridge Bitcoin Electricity Consumption Index). A Dyson sphere could power approximately 2.2 × 10¹⁶ Bitcoin networks simultaneously — 22 quadrillion copies.
The most extensive analysis of Bitcoin under Kardashev-scale energy abundance comes from Dhruv Bansal's "Bitcoin Astronomy" series for Unchained — an industry blog rather than peer-reviewed literature, but the reasoning is worth engaging. Bansal introduces the concept of "hash horizons" — spatial limits within which proof-of-work mining is feasible, determined by block time and the speed of light. Bitcoin's 10-minute block time creates a hash horizon where Mars, at 12 light-minutes, is outside Earth's mining range. Martians could use Bitcoin but couldn't competitively mine it. The speed of light becomes a constraint on consensus participation — a problem that no amount of energy solves. This intuition is physically grounded even if the framework hasn't received formal academic treatment; Decker and Wattenhofer's 2013 measurements of block propagation latency on Bitcoin's network provide some empirical footing for thinking about how propagation delays affect mining incentives. Pass, Seeman, and Shelat (2017, EUROCRYPT) demonstrated formally that Bitcoin consensus security is determined by the ratio of mining difficulty to propagation delay — making the hash horizon a genuine security constraint at interplanetary scales.
So what would actually happen to Bitcoin if a Type II civilization directed a Dyson sphere's output at mining?
The difficulty adjustment answers this cleanly: nothing changes in the supply schedule. Difficulty would adjust upward by astronomical factors. Block production would remain at 144 per day. The same 450 BTC would be issued. What would change dramatically is security: the cumulative hashrate would make the network essentially impervious to attack by anything short of another Dyson sphere.
The energy expenditure per coin would become cosmically enormous. But the coin production rate would be identical to a world where a single laptop mines Bitcoin.
This is the thermodynamic paradox: more energy doesn't create more Bitcoin. It just makes Bitcoin harder to attack. The difficulty adjustment absorbs any amount of energy input and converts it entirely into security, preserving scarcity absolutely.
When energy costs zero, what's left?
The deeper question the Dyson sphere scenario surfaces is this: if energy becomes free, what determines mining profitability?
Mining profitability isn't just about energy costs. It's about total revenue available to miners. Total miner revenue equals the block subsidy plus transaction fees, denominated in BTC and valued at market price. This revenue caps total mining expenditure — miners collectively won't spend more on mining than they earn.
Under energy abundance, the constraint shifts entirely to hardware.
TSMC dominates fabrication of the most advanced-node chips, with near-exclusive capacity at the 3nm node where leading Bitcoin ASICs now compete alongside AI accelerators — TSMC holds roughly 90% of 3nm production capacity, with Samsung's yields at this node remaining well below TSMC's. TSMC's CEO C.C. Wei acknowledged in an April 2025 earnings call that advanced-node capacity falls about three times short of AI demand alone. A modern semiconductor fab costs $20–40 billion and takes 3–5 years to build.
Even with unlimited energy, you cannot fabricate chips without gallium. China has controlled approximately 98–99% of worldwide primary low-purity gallium production (USGS Mineral Commodity Summaries 2025). China's August 2023 export controls sent Rotterdam prices surging over 150%. China escalated to a full export ban to the US in December 2024 (MOFCOM Announcement No. 46). Following the Trump-Xi meeting in November 2025, the total ban was suspended until November 27, 2026 — returning to a licensing regime — but the strategic vulnerability of a single-country supply for a critical semiconductor input has been demonstrated. Diversification efforts are underway but years from meaningful alternative production.
There are also thermodynamic limits to computation itself. Landauer's principle (Landauer, 1961) sets a minimum energy cost of kT ln(2) — approximately 2.9 × 10⁻²¹ joules per bit erasure at room temperature, experimentally verified by Bérut et al. in Nature (2012). SHA-256 is logically irreversible — a many-to-one function — so this floor genuinely applies to each of the roughly 50,000 to 250,000 irreversible logical operations involved in a double-SHA-256 hash.
The important context: current ASICs operate roughly 20,000 to 80,000 times above this theoretical limit — far from the Landauer floor but far closer than the commonly cited figure of "10 billion times." The arithmetic: the Antminer S21 XP Hydro consumes approximately 1.2 × 10⁻¹¹ joules per hash at 12 J/TH. The Landauer minimum for a full SHA-256d hash (using ~50,000 irreversible bit operations as a conservative estimate) is 50,000 × 2.87×10⁻²¹ ≈ 1.4 × 10⁻¹⁶ joules. The ratio is roughly 80,000×, or about five orders of magnitude at the low end of the operation-count estimate; using more realistic midrange estimates of 100,000–150,000 operations, the ratio drops to 28,000–42,000× (~four to four-and-a-half orders of magnitude). This is consistent with the broader semiconductor reality: modern CMOS transistors typically operate 10⁵ to 10⁶ times above Landauer per switching event.
But even at the Landauer limit, computation generates waste heat that must be radiated away. At civilizational scales, the surface area available to radiate heat into space becomes the binding physical constraint.
This is why Dyson spheres re-radiate in infrared — and why a computation-optimized Dyson sphere (what Robert Bradbury called a "Matrioshka brain") uses nested shells at progressively lower temperatures. Jason T. Wright's review "Dyson Spheres" (Serbian Astronomical Journal 200, 2020) noted that waste heat radiation is what makes them detectable — the basis for SETI searches.
Seth Lloyd's "Ultimate Physical Limits to Computation" (Nature, 2000) provides the theoretical upper bound: a 1-kg computer operating at absolute physical limits could perform approximately 5 × 10⁵⁰ operations per second and store around 2 × 10³¹ bits. These figures are almost incomprehensibly large — and still subject to Landauer heat dissipation. Even ultimate-limit computation doesn't escape thermodynamics.
The binding constraints on Bitcoin mining under energy abundance are: first, the revenue available to miners, which caps total mining expenditure regardless of energy cost; second, hardware fabrication capacity, which determines how much hashrate can actually be deployed; and third, the difficulty adjustment itself. The system's elegance is that it doesn't need energy to be scarce. It uses the difficulty adjustment to make the output scarce regardless of the input.
Where I might be wrong
I should flag several places where my analysis could be off.
First, the difficulty adjustment is elegant, but it's not without vulnerabilities beyond the Noda et al. instability finding and the timewarp attack I discussed earlier. The timewarp vulnerability specifically: Poinsot's BIP 54 analysis specifies that a majority miner could reduce difficulty to near-1 within 38 days without any patch. My analysis treats the difficulty adjustment as a reliable invariant; in practice its robustness depends on no single entity holding majority hashrate — and given that Foundry USA and AntPool already exceed 50% of hashpower combined, that assumption is not guaranteed.
Second, I've argued that production costs don't determine Bitcoin's value — costs follow price, not the other way around. That's what the Granger causality evidence on gold (O'Connor et al., 2016) and the Liu & Tsyvinski (2021) finding on network factors both suggest. But Hayes's equilibrium argument that production costs act as a gravitational attractor in competitive markets is also real. These claims are compatible: production costs don't cause value, but they respond to it in ways that create real-world pricing dynamics. The full story is more nuanced than either "energy stores value" or "energy is irrelevant to price."
Third, I should acknowledge the dominant academic view on Bitcoin's monetary status. The literature I've cited frames Bitcoin as a monetary system in development, but a substantial body of peer-reviewed work reaches a different conclusion. Baur, Hong, and Lee (2018, Journal of International Financial Markets, Institutions and Money) found that roughly one-third of bitcoins are held by investors who never transact — suggesting Bitcoin is used primarily as a speculative asset rather than a medium of exchange. Yermack (2015) found Bitcoin's volatility is an order of magnitude above conventional currencies, making it a poor unit of account. If Bitcoin's primary value proposition is speculative rather than monetary, the entire energy-security framework must be evaluated differently — not as protecting a payment system, but as protecting a store-of-value whose long-term demand dynamics remain uncertain.
If you think I'm wrong about any of this, I'd like to hear why.
The genius is in the thermostat, not the furnace
The deepest insight from this analysis is one I didn't fully appreciate before working through it.
Bitcoin's value proposition does not depend on energy being expensive. It depends on the difficulty adjustment making energy expenditure irrelevant to supply. A civilization with a Dyson sphere mining Bitcoin would produce exactly the same number of coins as Satoshi's laptop in 2009 — just with incomprehensibly more security protecting those coins.
This inverts the popular narrative. Bitcoin is not stored energy or a digital battery. The energy is consumed, converted to heat, and gone. What remains is a proof of expenditure — the Szabo intuition, regardless of its formal academic status — and a supply schedule enforced by mathematics rather than politics.
The energy expenditure matters for security, not for value. The value appears to come from network effects (empirically supported by Liu & Tsyvinski, 2021), absolute supply scarcity, and the practical resistance to censorship and confiscation that proof-of-work enables — though that last property, while central to Bitcoin's cultural identity, has received almost no rigorous academic treatment.
But the thermostat that makes all of this work has real boundaries. The Noda et al. instability result, the unpatched timewarp attack, the within-epoch timing deviations — these aren't theoretical curiosities. They're the engineering tolerances of a mechanism that has performed remarkably well for fifteen years but hasn't been stress-tested at its formal limits.
The next issue will examine a vulnerability the thermostat can't address: what happens when the revenue that funds Bitcoin's security starts to disappear.
Sources & Further Reading
This issue draws on significantly more external research than my usual forensics pieces. I've tried to be explicit throughout about which sources are peer-reviewed academic literature, which are industry publications, and which are advocacy writing — a distinction that matters when assessing how much weight to put on a claim.
Mining economics: JPMorgan, "Bitcoin Mining Economics Post-Halving" (Panigirtzoglou et al., January 2026 — industry research, paywalled). Cambridge Digital Mining Industry Report (Neumueller et al., April 2025, SSRN 5236060 — preferred source, survey-based, freely available). Marathon Digital Q2 2025 Earnings Report. Riot Platforms Q2 2025 Earnings Report. CleanSpark FY2025 Operational Updates. Hashrate Index ASIC pricing data. CoinGecko/BestBrokers international mining cost comparisons. Fu (2020, IACR ePrint 2020/1574) for real-options model of miner shutdown behavior. Derks, Gordijn & Siegmann (2018, Electronic Markets) for empirical mining profitability dynamics.
Mining concentration: b10c.me, "Bitcoin Mining Centralization in 2025" (2025). Eyal & Sirer, "Majority Is Not Enough: Bitcoin Mining Is Vulnerable," Financial Cryptography and Data Security (2014). Cong, He & Li, "Decentralized Mining in Centralized Pools," Review of Financial Studies 34(3) (2021).
Hashrate & difficulty: Cambridge Bitcoin Electricity Consumption Index (CBECI). CoinDesk, "Bitcoin Hashrate Crosses 1 Zettahash" (April 2025). The Block, "Bitcoin Difficulty Drops 11.16% After Winter Storm Fern" (late January 2026). Hashrate Index difficulty data confirming November 2025 ATH of 156T.
Difficulty adjustment, stability, and vulnerabilities: Josh Hendrickson, "Bitcoin and the Genius of the Difficulty Adjustment" (Bitcoin Policy Institute, December 2024 — industry/policy, not peer-reviewed). Saifedean Ammous, The Bitcoin Standard (Wiley, 2018). Garay, Kiayias & Leonardos, "The Bitcoin Backbone Protocol with Chains of Variable Difficulty," CRYPTO 2017 (LNCS 10401). Noda, Okumura & Hashimoto, "An Economic Analysis of Difficulty Adjustment Algorithms in Proof-of-Work Blockchain Systems," International Economic Review (DOI: 10.1111/iere.70028). Bitcoin Core pow.cpp for the 4× clamp specification. Antoine Poinsot & Matt Corallo, BIP 54 "Consensus Cleanup" (April 2025). Pass, Seeman & Shelat, "Analysis of the Blockchain Protocol in Asynchronous Networks," EUROCRYPT 2017.
Formal economics of money and value: Kiyotaki & Wright (1989, JPE 97(4)). Lagos & Wright (2005, JPE 113(3)). Luther (2019, Journal of Institutional Economics 15(2)). Schilling & Uhlig (2019, JME 106). Fernández-Villaverde & Sanches (2019, JME 106). Rudd & Porter (2025, JRFM). Hayes (2017, Telematics and Informatics 34(7); 2019, Applied Economics Letters). Liu & Tsyvinski (2021, RFS 34(6)). Prat & Walter (2021, JPE 129(8)). Baur, Hong & Lee (2018, JIFMIM 54).
Energy-value debate: Nick Szabo, "Bit Gold" (2005) and "Shelling Out" (2002). Adam Back, "Hashcash" (1997). Umlauft (2018, SSRN #3182646). O'Connor, Lucey & Baur (2016, JIFMIM 40). Lowery, Softwar (MIT Thesis, 2023). Lopp, "Softwar Thesis Review" (lopp.net, 2023).
Aluminum history and S2F critique: Science History Institute. International Aluminium Institute. Bertilorenzi (2015). Morillon & Chacon (2022, Studies in Economics and Finance 39(3)). Burger (2020). JRFM 17(10):443 (2024). dappGambl (September 2023).
Dyson sphere & Kardashev scale: Dyson (1960, Science 131(3414)). Kardashev (1964, Soviet Astronomy 8(2)). Bansal, "Bitcoin Astronomy" (Unchained, 2018–2021). Decker & Wattenhofer (2013, IEEE P2P). Suazo et al. (2024, MNRAS 531(1):695).
Computation limits & hardware: Landauer (1961, IBM JRD 5(3)). Bérut et al. (2012, Nature 483). Lloyd (2000, Nature 406). Wright (2020, Serbian Astronomical Journal 200). USGS Mineral Commodity Summaries 2025. MOFCOM Announcement No. 46 (December 2024). TrendForce/SemiWiki (2025).
Geo Nicolaidis
Builder, TrailBit.io
If you found this useful, subscribe to get the next issue in your inbox. Each issue breaks down a different heuristic used in Bitcoin forensics — what it assumes, where it breaks, and why it matters.