Bitcoin’s Difficulty Is Really a Bidding War for Power

The Number That Is Really a Power Bill

On 19 September the Bitcoin network raised its mining difficulty by 4.16%, to about 132.76 trillion, the biggest single step up since June and the second increase in a row. As of 29 September the figure still reads 132.76 trillion at block 969,078, and the next automatic adjustment, due around 3 October, is tracking a further rise of roughly 0.62% on CoinWarz.

Most coverage treats difficulty as a statistic about computing: how hard it is to find a block, how much hashing the network is doing. That is true, but it buries the more useful reading. Difficulty is really a number about electricity. It rises when power is cheap and plentiful enough that operators keep switching machines on, and it falls when power gets expensive, gets rationed, or gets outbid by something willing to pay more for the same megawatt. Read that way, the 2026 difficulty chart is not a graph of chips. It is a map of where the world’s cheapest electrons are and who is fighting over them.

Bitcoin itself trades near $83,200, about 34% below its October 2025 record of $126,080, per CoinGecko. Here is the network at a glance.

This piece reads difficulty as an energy signal from end to end: what it measures, why it tracks the price of power, how each 2026 swing maps onto a real electricity event, where the hashrate physically sits and why, and what it means that in 2026 Bitcoin mining picked up a rival bidder for its power, the AI and high-performance-computing buildout, with far deeper pockets.

A 60-Second Refresher on What Difficulty Does

Bitcoin has no meter that measures how much computing the network is doing. It has a clock. Satoshi Nakamoto’s white paper fixed the target block time at ten minutes and built a self-correcting rule to hold it there. As the white paper puts it, “to compensate for increasing hardware speed and varying interest in running nodes, the proof-of-work difficulty is determined by a moving average targeting an average number of blocks per hour. If they’re generated too fast, the difficulty increases.”

The mechanism is mechanical. Every 2,016 blocks, roughly two weeks, each node compares how long those blocks actually took against the 20,160 minutes they should have taken at ten minutes each. It then rescales difficulty by that ratio: new difficulty equals old difficulty times 20,160 divided by the actual minutes elapsed. If blocks arrived fast, difficulty goes up; if they came slowly, it goes down. A clamp limits any single adjustment to a factor between 0.25x and 4x, so no retarget can move more than 300% up or 75% down. That is the whole thermostat.

Two wrinkles matter for accuracy. The retarget window spans 2,016 blocks but measures only the 2,015 intervals between them, a rounding quirk worth about 0.05%; and block timestamps are only loosely policed (a block must be later than the median of the last eleven, and no more than two hours ahead of a node’s clock), so the clock the network runs on is honest but coarse. Neither changes the big picture. Difficulty is a two-week moving average of how fast blocks are being found, and blocks are found faster only when more hardware is switched on and drawing power. Difficulty is the network’s read on how much electricity is currently being burned to secure it, expressed as a dimensionless ratio against the very first block.

Why Difficulty Is a Proxy for the Price of Power

Hashrate is not measured directly either; it is inferred from difficulty and block times (hashrate is roughly difficulty times 2^32, divided by the average block interval in seconds). And hashrate, in turn, is the sum of one decision made thousands of times a day across the planet: is it worth paying for the electricity to run this machine right now?

That decision runs through a number miners watch even more closely than difficulty: hashprice, the daily revenue one unit of hashing earns. In late September hashprice sat near $40 per petahash per day, up from roughly $32 in early August and a June low around $27, but still well below the roughly $49 seen in October 2025, according to Luxor’s Hashrate Index. Hashprice is set by the BTC price, the block subsidy (3.125 BTC since the April 2024 halving), transaction fees, and difficulty. Fees currently add less than 1% on top of the subsidy, so the block reward is still almost entirely newly issued bitcoin. When hashprice climbs above a machine’s running cost, operators plug in; when it drops below, they unplug. That running cost is, overwhelmingly, electricity.

This is why difficulty is a lagging proxy for the price of power. JPMorgan’s Nikolaos Panigirtzoglou put the loop in one sentence, via TFTC: “When bitcoin trades below its production cost, higher-cost miners power down, the hashrate declines, and difficulty adjusts lower.” The bank pegs all-in production cost near $78,000 per BTC and estimates the beta of difficulty to price has climbed to 0.62 over the past six months, meaning difficulty now tracks price more tightly than it once did. Difficulty does not move because computing became harder in the abstract. It moves because the economics of buying electricity changed.

The 2026 Chart, Read as an Energy Map

Run the year’s retargets through that lens and the chart tells an energy story, not a computing one.

The June collapse, a 10.09% cut on 14 June and the largest of the year, came as BTC traded far below production cost and margins went negative for the least efficient fleets; operators pulled machines off the grid because the power was no longer worth buying. The choppy summer that followed (up 7.15%, down 5.00%, down 0.74%, up 0.99%, down 1.31%) is the signature of a fleet toggling on and off around breakeven as price and power costs jostled. Then came the autumn turn: +1.31% on 5 September, then +4.16% on 19 September, as BTC’s recovery back above the roughly $78,000 to $80,000 cash-cost line made it worth buying power again and idled machines came back online. Panigirtzoglou’s loop, run in reverse: price up, hashrate holds, difficulty up.

Notice that difficulty spent almost the entire year below its November 2025 record of about 155.97 trillion, and still sits roughly 15% under it. For most of Bitcoin’s history difficulty paused only during government bans (China in 2021) or deep capitulations (2018, 2022). As CoinDesk noted, 2026 marked only the second time ever that difficulty fell below its year-earlier level, and it happened during neither a ban nor a crash. The power did not disappear. Some of it simply found a better-paying customer.

The Marginal Megawatt

Here is the deeper reason difficulty behaves like a power gauge. Bitcoin mining is a commodity business with fixed output. The protocol issues the same roughly 450 BTC a day no matter how many machines compete for it. So when a new operator plugs in cheap power, they do not add to the supply of new bitcoin; they raise difficulty, which lowers everyone’s share, until the marginal miner (the one with the highest power cost still running) is pushed back to breakeven. Difficulty is the mechanism that continuously drags the entire network’s cost of production toward the cost of the cheapest marginal megawatt anyone is willing to commit.

This is what makes mining unlike oil or gold. When the price of gold rises, miners dig more of it and supply grows. When the BTC price rises, miners cannot produce more bitcoin; issuance is fixed by the protocol. Extra competition instead raises the cost of winning the same fixed reward, so the marginal cost of production is endogenous: it climbs to meet revenue rather than being set by geology. Difficulty is the variable that carries that adjustment, which is why it behaves less like a technical parameter and more like a price.

MARA chief executive Fred Thiel described the endgame bluntly. Mining, he told CoinGeek, “is a zero-sum game. As more people add capacity, it gets harder for everybody else. Margins compress, and the floor is your energy cost.” That floor is the whole game. Two miners running identical machines have identical hashrate and face identical difficulty; the only thing separating a profitable operation from a bankrupt one is the price it pays per kilowatt-hour. Difficulty sets the revenue every machine earns; power price decides who keeps any of it. This is why difficulty and geography are the same story told two ways: difficulty is the global average of the competition for cheap power, and the world map of hashrate is where that cheap power physically turned out to be.

Revenue Per Megawatt-Hour, the Line Difficulty Draws

To see the link at the meter, convert hashprice into the unit that matters to a power buyer: revenue per megawatt-hour. Luxor’s Hashrate Index publishes this by efficiency tier. At a hashprice near $40, a modern sub-14 J/TH fleet earns roughly $136 for every megawatt-hour it consumes, while a 25-to-38 J/TH machine from a couple of hardware generations back earns only about $51.

That table is really a survival chart. A miner is cash-profitable only when its all-in power cost per megawatt-hour sits below its revenue per megawatt-hour. So the newest hardware can survive power around 13 to 14 cents per kilowatt-hour, while an older 25-to-38 J/TH machine needs power below roughly 5 cents to stay above water. Every time difficulty rises, it drags that survival line down, and the oldest, least efficient machines on the most expensive grids are the first to fall below it and switch off, which in time pulls difficulty back. The thermostat and the meter are the same instrument.

This is also why efficiency and location have partly swapped roles over the cycle. A sub-10 J/TH machine can profit on ordinary industrial power almost anywhere, which loosens the old imperative to chase the very cheapest, least stable frontier electricity and rewards grids that can also host something else. Hold that thought; it is the hinge of the 2026 story.

The World Map of Hashrate

If difficulty is the average price of committed power, the hashrate map shows where that power sits. Luxor’s Hashrate Index Q2 2026 heatmap puts the network near 1,004 EH/s (down about 5.8% from the prior quarter) and distributes it like this.

The United States alone hosts roughly 37% of global hashrate, on Texas and Wyoming grids built for deregulated, interruptible industrial load. Russia’s near-17% runs largely on cheap Siberian power. China, banned since 2021, has quietly returned to roughly an eighth of the network, though illegal mining is inherently hard to measure and estimates run wider. The rest of the top ten is a tour of stranded and subsidised energy: Paraguayan hydro from the Itaipu dam, Gulf sovereign gas in the UAE and Oman, Ethiopian hydro from the new Grand Ethiopian Renaissance Dam.

Each entry is really a power story. Russia legalised mining nationally in late 2024, yet its largest industrial miner sits under US sanctions and several regions have imposed year-round bans over grid strain, a legal-at-home, isolated-abroad paradox. China still manufactures the majority of the world’s ASICs even while banning domestic mining, so it profits as the exporter of the machines that run everywhere else. Paraguay’s rise rode surplus hydropower, though the state utility has since raised rates and warned of grid strain. Kazakhstan, which briefly absorbed the 2021 China exodus to reach roughly 18% of the network, is the cautionary tale of what happens when cheap power turns out to be cheap for a structural reason: grid strain forced caps and monitoring, and its share collapsed to under 2%.

The concentration is real: the top three countries hold about two-thirds of all hashrate, and at the pool layer the picture is tighter still, a point worth reading alongside our look at the mining-pool leaderboard and who actually secures the network. But the map is not fixed. Difficulty is the force that keeps redrawing it: when one jurisdiction’s power gets more expensive, its machines fall below the survival line first, and the hashrate effectively migrates to wherever the next-cheapest committed megawatt is. Iran’s loss of about 7 EH/s to regional conflict earlier in 2026 did not shrink the network; the work redistributed.

Texas, Where Difficulty Meets the Grid

Nowhere is difficulty-as-power-signal clearer than in Texas, home to the largest single concentration of US hashrate. On 22 July 2026 the ERCOT grid set an all-time demand record of about 91,000 megawatts, topping the previous mark of 85,508 MW from August 2023, as CryptoSlate reported. Bitcoin miners are one reason the grid held. Because a mining machine’s only cost of powering down is the block reward it forgoes (no spoiled product, no broken process, no restart penalty), miners are the most flexible large load on the system. They sign demand-response agreements that pay them to stand ready to curtail and pay them again when curtailment is actually called, so a facility can earn from electricity it never consumes.

The programs come in tiers by how fast a load can respond, from slower non-spinning reserves to near-instant responsive reserve service, the best-paying. Texas also runs a Four Coincident Peak system, in which curtailing during the single highest-demand interval of each summer month cuts the following year’s transmission charges, an incentive that rewards exactly the kind of instant, penalty-free shutdown mining can provide. The numbers are not small. Riot Platforms curtailed more than 95% of its power during the August 2023 heat wave and booked $31.7 million in power credits against $8.9 million of forgone bitcoin; across 2025 it earned $56.7 million in such credits, and $21 million in the first quarter of 2026 alone.

When a heat wave forces mass curtailment, hashrate drops for days, blocks slow, and the next difficulty retarget prints lower. Summer power stress is legible in the difficulty chart. The US Energy Information Administration now lists data centers and cryptocurrency mining together as a primary driver of Texas power-demand growth. That pairing is the tell for what changed in 2026.

The Seasonal Clock Inside the Chart

Because difficulty tracks power, it inherits power’s seasons. The clearest historical example predates the 2021 ban: Chinese miners used to migrate twice a year, running on cheap hydro in Sichuan and Yunnan through the May-to-October wet season, then moving back to coal-fired Xinjiang and Inner Mongolia for the dry months. Hashrate, and therefore difficulty, rose and fell with the rains. The ban scattered that pattern, but China’s quiet return to roughly an eighth of the network means a muted version still exists.

The dominant seasonality now runs through the United States. In Texas, the summer months bring both the highest power prices and the most curtailment events, so mid-year is when the largest US fleets spend the most time offline; the June and July difficulty cuts of 2026 line up with exactly that window. Northern grids invert it, as winter heating demand raises power prices and squeezes margins in the coldest months. None of this is written into the protocol, which knows only block times. It leaks in through the electricity bill, and difficulty is where it becomes visible. A reader who knows the power calendar can often anticipate the direction of a retarget before the estimators settle.

The New Bidder for the Same Megawatts

For a decade Bitcoin miners competed mainly with one another for cheap power. In 2026 they picked up a rival with a fatter wallet: AI and high-performance computing. The same warehouses, substations and interconnection-queue slots that host ASICs can host GPUs, and GPUs earn far more per megawatt.

The scale of that reallocation is now the dominant force on the difficulty chart. CoinShares research chief James Butterfill has called this “one of the most challenging periods” for miners and estimates listed miners could draw as much as 70% of revenue from AI by the end of 2026, up from roughly 30%, backed by more than $70 billion in announced AI and HPC contracts across the sector, per News.Bitcoin.com. Needham analyst John Todaro is blunter on the economics: “The revenue per megawatt and EBITDA margins are far higher for HPC and AI colocation than for mining.” When the same megawatt earns multiples more pointed at an AI tenant, the rational operator diverts it, and the hashrate it would have produced never appears. That missing hashrate is exactly what the summer’s difficulty cuts recorded.

Riot’s Rockdale campus, once one of the largest Bitcoin mines in North America, now hosts a leased AMD data center, a concrete example of megawatts crossing from mining to AI; a Texas A&M and Harvard preprint noted by CryptoSlate models the pattern, finding mining load declines as electricity costs climb, precisely because the flexible, low-value use loses the bidding war to firm, high-value compute. There is a grid cost too. Nearly 90% of ERCOT’s interconnection queue is now AI and HPC data centers, and those tenants sign uptime commitments that leave them unable to power down in an emergency, unlike miners. As sites convert from flexible mining to firm AI load, the grid’s cheap emergency brake shrinks.

The same scramble for compute power is what decentralized-AI projects run into from the other side, as our report on Gensyn and the bandwidth wall lays out. Thiel’s forecast follows directly: “By 2028, you’ll either be a power generator, be owned by one, or be partnered with one. The days of being a miner plugged into the grid are numbered.” Difficulty, in that world, becomes a scoreboard for who won the power auction.

How Much Electricity Is That, Really?

Put a size on it. The Cambridge Centre for Alternative Finance estimates the Bitcoin network drew roughly 138 terawatt-hours over the past year, about half a percent of global electricity and comparable to the annual consumption of a mid-sized industrial country, with a wide uncertainty band (a lower bound near 74 TWh, an upper bound near 242 TWh) because the exact hardware mix cannot be observed directly. That range lives on the Cambridge index.

From the protocol’s point of view that electricity is not waste; it is the security budget. The network pays miners roughly 450 BTC a day, about $37 million at current prices, which annualises to well over $13 billion, and the majority of that flows straight into power and hardware. Difficulty is the dial that sets how much work, and therefore how much energy, stands between an attacker and rewriting the chain. When people ask whether Bitcoin’s energy use accomplishes anything, the honest answer is that the energy is the thing: it is what makes the ledger expensive to attack.

The mix is shifting toward miners that own or co-locate with generation, from flared-gas capture to grid-balancing renewables, which is the logic behind Thiel’s power-generator forecast: the cheapest, most defensible megawatt is increasingly one you produce yourself. How clean that energy is varies enormously by jurisdiction, which is the other reason geography matters as much as the raw difficulty number.

Difficulty vs Price, the Loop That Never Rests

Because difficulty chases power, and power-buying chases the BTC price, difficulty ends up chasing price with a lag. JPMorgan’s 0.62 beta quantifies it: a move in price now pulls difficulty most of the way in the same direction within a retarget or two. The lag is the interesting part. Price moves in seconds; difficulty moves once every two weeks. So there is always a window where price has jumped but difficulty has not yet caught up (margins fat, every machine worth running), or price has fallen but difficulty is still high (margins thin, marginal machines switching off). September 2026 was the first kind of window: BTC’s rally repriced hashprice up about 22% before difficulty clawed part of it back with the +1.31% and +4.16% steps, as News.Bitcoin.com tracked.

That makes the near-term difficulty path a bet on the macro that drives BTC. The September FOMC, which delivered a hawkish quarter-point hike, set the price backdrop miners are now underwriting; the market read is in our coverage of crypto’s reaction to a hawkish Fed. If price holds above the roughly $78,000 production line, more idled and AI-flexible capacity swings back to mining and difficulty keeps rising; if price rolls over, the marginal megawatt goes back to AI or goes dark, and difficulty stalls. The forward hashprice market, which has traded below spot for much of 2026, is effectively a crowd-sourced forecast that some of the recovery fades and that difficulty keeps compressing margins back toward the marginal miner’s cost, exactly as the thermostat intends. Difficulty is not a leading indicator of price; it is a trailing receipt for the power that price justified buying.

What Difficulty Is Not

Two cautions. First, difficulty is not a security dial you can read in a straight line. A higher number means more work per block and a more expensive network to attack, but it does not scale one-for-one with safety, and what actually protects the chain from a majority attack is the real-world cost of acquiring hashrate and power, not the difficulty integer itself. Duke finance professor Campbell Harvey has modelled a one-week majority attack at roughly $8 billion once paired with a derivatives short, which he framed to News.Bitcoin.com as “about 50 basis points of the value of bitcoin,” a reminder that the security budget is large but finite.

Second, difficulty is a proof-of-work idea with no proof-of-stake analogue. Ethereum secures itself with staked capital rather than electricity, so it has no difficulty, no retarget and no meaningful power bill; who controls that stake is a governance question rather than an energy one, as our comparison of who controls staked ETH explores. Bitcoin’s security is metered in megawatt-hours, and difficulty is the meter. That distinction also shapes US oversight: the SEC’s Division of Corporation Finance stated in March 2025 that proof-of-work mining, whether solo or pooled, does not by itself involve a securities offering, treating it as an industrial-energy activity rather than an investment contract. (There is one genuine flaw in the timestamp machinery difficulty relies on, the timewarp loophole, and a 2026 soft-fork fight to fix it, but that is a consensus story rather than an energy one.)

Reading the Next Adjustment Like an Energy Trader

The next retarget is due around 3 October. As of late September, CoinWarz projected a small rise near +0.62%, with the current epoch about 70% complete and recent blocks arriving a touch under ten minutes apart. Early estimates are noisy by design. As Luxor’s Kaan Farahani put it in a Hashrate Index roundup: “Early on, difficulty predictions are shaky because of short-term variance. As time (or blocks) pass, the noise fades and the signal sharpens.” The single most useful field to watch is the average block time in the current epoch: under ten minutes means machines are being added faster than difficulty expected (a rise is coming), over ten minutes means capacity is leaving (a cut is coming).

Read as an energy signal, the tells are physical. A heat wave across Texas in July or August, a cold snap that spikes power prices, a wave of AI conversions taking sites offline, a new hydro build in Paraguay coming online: each shows up first as a change in block times, then, two weeks later, as a difficulty print. To track it yourself:

• Watch the average block time this epoch on a tracker such as mempool.space or CoinWarz; that is the leading edge.
• Watch hashprice; it tells you whether the marginal machine is above or below its power cost.
• Watch the BTC price against the roughly $78,000 production line; that is the on/off switch for flexible capacity.
• Watch grid events in the big mining states, especially ERCOT curtailment days in a Texas summer.
• Treat any single-day hashrate reading as noisy, and lean on the seven-day average instead.

Do that and the difficulty number stops being a mysterious computing statistic and starts reading like what it is: a two-week receipt for the global auction over the world’s cheapest power. The next print will not tell you how fast chips got. It will tell you where the electrons went.

Frequently Asked Questions

Why does Bitcoin mining difficulty change?

Difficulty readjusts automatically every 2,016 blocks (about two weeks) to keep the average time between blocks near ten minutes. When more machines switch on and blocks arrive faster, difficulty rises; when machines power down and blocks slow, it falls. Because switching machines on or off is mostly a decision about the price of electricity, difficulty ends up tracking the cost of power and the BTC price.

What is Bitcoin’s mining difficulty right now?

As of late September 2026 difficulty is about 132.76 trillion, set by the +4.16% adjustment on 19 September, roughly 15% below the record of about 155.97 trillion from November 2025. The next adjustment, due around 3 October, was tracking a small further rise near +0.62%.

How does difficulty relate to electricity use?

Difficulty determines how much hashing, and therefore how much power, the network draws to secure itself. A higher difficulty lowers the revenue each machine earns per megawatt-hour, so only miners with cheap enough power stay profitable. Cambridge estimates the network uses roughly 138 terawatt-hours a year, about half a percent of global electricity.

Is AI competing with Bitcoin miners for power?

Yes. The same power, buildings and grid connections that host mining machines can host AI and high-performance-computing hardware, which earns far more per megawatt. In 2026 analysts estimated listed miners could get up to 70% of revenue from AI, and much of the year’s difficulty decline reflected operators diverting power from mining to AI rather than any ban or price crash.

Does higher difficulty make Bitcoin more secure?

Broadly yes, because more work and more energy stand behind each block, making the chain more expensive to attack. But security depends on the real-world cost of acquiring hashrate and power, not on the difficulty number alone, and more difficulty does not scale safety one-for-one. Difficulty is best read as a measure of committed energy rather than a direct safety score.

By Marcus Okafor, Bitcoin and mining desk, HOGE Wire.