There is a number Bitcoiners love and a number they ignore, and in the current cycle the gap between them has become the entire story. The loved number is hashrate. It printed a new all-time high, sustaining a seven-day moving average well above the peaks of the last bull run. The ignored number is fee revenue as a share of total miner income โ currently lodged in the low single digits, punctuated by spikes that decay within days.
Put those two lines on the same chart and the contradiction is immediate. The network is spending more to secure itself than at any point in its history, while earning less per unit of work than at any point since the subsidy was worth something. This is not a crash. It is an equilibrium. And like most equilibria in proof-of-work, it resolves in favor of whoever already owns the cheapest capital.
Code doesn't care about your thesis. It only executes the reward schedule you agreed to.
Context: what the halving actually changed
April 2024 was mechanical, not dramatic. The block subsidy fell from 6.25 BTC to 3.125 BTC at block height 840,000 โ a deterministic step function written into the coinbase reward logic and executed without debate, governance, or a vote. That clarity is the entire point of the design, and it is also the source of the current stress.
The difficulty adjustment algorithm is the other half of the mechanism, and it is worth restating precisely because most commentary misreads it. Every 2016 blocks โ roughly two weeks โ the network recomputes difficulty so that block production averages ten minutes. If hashrate rises, difficulty rises; if miners unplug, difficulty falls. The algorithm is a control loop, not a jury. It does not reward virtue and it does not punish over-extension. It just retargets.
Here is the part the market routinely gets wrong: hashrate is a lagging indicator of capital already deployed, not a real-time measure of profitability. Miners do not power down ASICs the moment a block's revenue dips. They power down when the machine's marginal electricity cost exceeds the marginal coin it produces โ and even then, only after they have exhausted their hedging, their power contracts, and their sunk-cost psychology. A rig that has been paid for is a rig you run until it loses money, not until it loses margin.
I spent enough of 2022 reverse-engineering collateral and liquidation mechanics to recognize the pattern. When a system's inputs are fixed and its outputs are variable, the stress does not surface as a price move. It surfaces as a slow migration of capacity toward the participants with the lowest cost basis. That is precisely what the post-halving data is showing, and it is showing it quietly.
Core: the three-pool arithmetic
Follow the block templates. Every block a pool constructs is a small optimization problem: select the highest-fee transactions, respect the block weight limit, and broadcast a valid proof before a competitor finds one first. The subsidy portion is identical for everyone. The fee portion is where differentiation lives โ and differentiation is exactly what consolidates.
Concentration in mining is not, primarily, a hashrate story. It is a template-construction and payout-settlement story. The largest pools win on three axes, and each one compounds.
First, latency to the mempool. A pool with better network topology sees high-fee transactions marginally earlier. That advantage looks trivial when blocks arrive ten minutes apart on average โ but they arrive in clusters, and in the clusters, milliseconds decide who captures a congested block. Second, block template software. Optimized, greedy fee selection beats naive selection across thousands of blocks. It is measurable, and it does not decay. Third, payout mechanics. PPS, FPPS, and PPLNS each shift variance between the miner and the pool. Post-halving, with the subsidy halved and fees unreliable, miners have migrated toward pools that absorb variance โ and that migration has a direction, which is up the scale curve.
The result is that a handful of pools routinely control the majority of blocks in any given window. Occasionally the seven-day concentration print looks less alarming; then a single pool wins a streak and it prints a fresh high. The variance in these prints is itself the tell. A decentralized system should not produce concentration statistics that behave like a momentum indicator.
Now add the fee market. After the 2024 subsidy cut, fee revenue briefly spiked โ inscription traffic, ordinal minting, the usual congestion events. Each spike was real, and each spike faded. Fee revenue is not a floor. It is an event. It clusters, it clears, and it leaves. The chain does not sustain a high-fee regime for long, because the applications that generate congestion are themselves temporary.
There is a metric that captures all of this and gets almost no attention: hashprice, the revenue per unit of hashrate per day. It declined sharply post-halving and has not recovered to prior levels even as the spot price has bounced. Hashprice is the honest number. It aggregates subsidy, fees, and actual hardware efficiency into a single figure a miner can act on โ and its trajectory is a slow bleedback, not a recovery. Traders watch price. Miners watch hashprice. Only one of those groups has to pay for electricity.
So the miner's revenue stream is a fixed subsidy that halves on a schedule, plus a variable fee stream that behaves like sporadic weather. Against that, the cost side is almost entirely fixed: power contracts, ASIC depreciation, cooling, land, labor. If you have ever modeled a leveraged position with a fixed carry cost and a stochastic income, you already know how this resolves. The variable side does not rise to meet the fixed side. The fixed side drags the participant under, and the difficulty algorithm quietly redistributes the survivors' share upward.
The efficiency argument, and why it is incomplete
The standard rebuttal is efficiency. Newer ASICs โ sub-15 joule per terahash, then sub-10 โ let a miner produce the same hashrate for less power, so the "real" cost of security keeps falling. True, and irrelevant to the equilibrium question.
Efficiency improvements do not raise anyone's revenue. They lower the marginal cost of production, which under a fixed subsidy simply bids up total hashrate until the marginal miner is back at break-even. This is the classic trap of technological deflation inside a fixed-reward system: every joule saved is eventually spent on more machines, and the only durable beneficiaries are the ASIC manufacturers and the cheapest power producers. The network gets stronger. The median miner gets poorer. Those are not contradictory statements. They are the same sentence.
This is why I find the "miners will simply upgrade" argument frustrating. Based on my own simulation work โ back in the DeFi summer, stress-testing collateral cascades โ the honest model is always the one where efficiency is passed through to competition rather than retained as margin. That is what my analysis of loosened collateral factors showed before the last cycle's liquidations. Fixed subsidy, variable fee, open entry. Same failure mode, different asset class.
One more layer worth naming: Stratum V2 and the promise that miners, not pools, choose transaction selection. In theory it redistributes template authority downward. In practice, adoption is low โ because the largest pools have no commercial incentive to give it up, and the code does not enforce what the protocol merely permits. That is the difference between a specification and a system.
Contrarian: concentration is not a governance bug
The reflexive reading of three-pool dominance is that something is broken, that Bitcoin is centralizing, that the consensus is hollow. I want to be precise about what is and is not true.
It is true that concentration of template construction is real and consequential. The pools decide which transactions make it into blocks. They are not anonymous algorithms; they are entities, subject to jurisdiction, subpoena, and policy.
And that is where the argument usually stops โ with alarm. The more useful analysis is this: the concentration is an expected equilibrium of a fixed-subsidy, variable-fee system, not a governance failure. When the reward for a given unit of work shrinks every four years, the only participants who survive are those with scale, cheap power, and the ability to absorb variance. Small miners are not being pushed out by a villain. They are being priced out by the reward schedule itself โ the same schedule everyone agreed to at genesis.
The uncomfortable corollary is that the decentralization maximalists' favorite metric โ raw hashrate โ is precisely the metric that rises as concentration increases. A bigger hashrate number is fully compatible with fewer independent miners. The number that would actually tell you something is the count of distinct entities that could profitably mine an empty block at current difficulty. That number is not printed on any dashboard, because it is small and getting smaller.
The code doesn't negotiate. It halves on schedule, regardless of who is left holding the ASICs.
Takeaway: watch the fee share, not the hashrate
If you are trying to judge Bitcoin's post-halving health, stop staring at the hashrate chart. It will keep rising, because difficulty and sunk capital expenditure guarantee it. Look instead at the 90-day rolling fee share of total miner revenue, and at the number of pools that would individually survive a 40% hashrate reduction. Both are falling.
Two questions worth sitting with. When the 2028 halving cuts the subsidy to 1.5625 BTC โ and if the fee share is still in single digits โ which pools are still standing? And when only three remain, what exactly does "decentralized consensus" describe: the network, or the handful of data centers that produce its blocks?
The next halving is already written into the chain. The only open variable is whether the fee market will be there to catch what the subsidy drops. If you believe the last two cycles' fee data, you already know the answer is not "yes."