Two paychecks, not one
Every time a miner successfully produces a block, they're paid twice, from two different sources. The first is the subsidy: newly created bitcoin, minted by the protocol itself, the exact figure tracked chapter to chapter in this book — 3.125 BTC per block right now. The second is fees: small payments attached by whoever wants their transaction included in that block, collected entirely from other users, not created by the protocol at all.
Chapter 5's table only shows the first paycheck, because only the first one follows a fixed, predictable schedule anyone can compute in advance. The second paycheck follows a market instead — and understanding why miners keep showing up requires looking at both paychecks together, not just the one that happens to be easy to put in a table.
The subsidy was always meant to hand off
This isn't an emergent problem the network stumbled into — it's the design working as intended. Bitcoin's original whitepaper describes the subsidy explicitly as a bootstrapping mechanism: a way to distribute coins and pay for security in the network's early years, before enough people were transacting on it for fees alone to matter. The expectation, stated from the start, was that fees would gradually take over as the subsidy wound down, not that the subsidy would fund mining forever.
Using this book's era numbering, era 33 still allows one satoshi of subsidy per block; era 34 begins at block 6,930,000 with zero subsidy. Integer-satoshi truncation fixes that block boundary, not a calendar date: ~2140 is an estimate. Chapter 5's schedule shows the final nonzero era and the first zero-subsidy era separately. Whether fees then support sufficient mining security is a separate, genuinely open question — covered honestly further down, not guaranteed by the end of issuance.
Fees: an auction for scarce space
A Bitcoin block has a hard ceiling on how much transaction data it can hold — roughly 4 million weight units under the current rules, translating in practice to a few thousand typical transactions per block. That ceiling is what turns fees into an actual market rather than a flat toll: when more people want their transaction in the next block than there's room for, they compete by attaching higher fees, and miners — rationally maximizing their own revenue — assemble each block by picking the highest-fee transactions waiting in the queue first.
This is the same kind of self-correcting mechanism Chapter 4 described for difficulty, applied to a market instead of a mining rule: no central authority sets the fee, it's simply what clears the auction each block.
That auction isn't hypothetical. On May 7, 2023 (block 788,695), fees inside a single block totaled 6.701 BTC — more than the 6.25 BTC subsidy paid at the time, the first block since 2017 where the fee paycheck alone outearned the subsidy paycheck. The trigger was the newly launched Ordinals and BRC-20 protocols: people inscribing and trading tokens directly on-chain flooded the mempool past 500,000 pending transactions and pushed priority fees above 650 sat/vB, with the average transaction fee briefly touching roughly $31. Demand cooled within a week or two and fees fell back to ordinary levels — a real instance of the auction above clearing at a genuinely higher price, not proof that fees alone can sustain a security budget this size indefinitely, but a concrete example that they can, for a stretch, rival it.
The security budget: miner revenue, not an attack price
Subsidy and transaction fees both reward miners for producing valid blocks. Here, security budget means the combined subsidy and fees paid to miners over a stated period, measured in bitcoin or converted to another currency. It describes miner revenue before costs, not miner profit, and it is not a quoted price for an attack.
Rewriting transaction history requires competing proof of work, not simply out-earning miners. An attacker trying to replace a past block must redo its work and the work of the blocks after it, then catch up with the honest chain. The whitepaper's double-spend model considers relative hash power and how many blocks the attacker is behind; it does not equate total miner revenue with an attacker's costs and potential payoff. Its incentive discussion says rewards may help encourage honesty, rather than establishing a universal profitability threshold.
Both revenue sources therefore matter to mining incentives, but adding them together does not establish that two networks with the same total revenue are equally expensive to attack. That comparison would need an explicit attack model, including the attacker's resources, costs, objective and time horizon. This chapter does not estimate an attack price or guarantee that future fees will provide sufficient security.
Sources: the Bitcoin whitepaper, sections 4, 6 and 11, and the Bitcoin Developer Guide on proof of work and block rewards.
The subsidy was never the point. Paying for security was the point — the subsidy was just the first, temporary way to fund it.
Why unprofitable miners don't break anything
Mining is a competitive, low-margin business by design, and individual miners regularly become unprofitable — when electricity prices spike locally, when a mining ban forces relocation, when older hardware simply can't compete with newer, more efficient machines. Chapter 4 already covered one real example: the July 2021 mining exodus from China, which knocked a huge fraction of global hash rate offline almost overnight.
The network does not need every miner to stay profitable, but difficulty adjustment is neither an instant response nor a profitability guarantee. Bitcoin mainnet retargets every 2,016 blocks — approximately two weeks at the target pace of ten minutes per block, not after a guaranteed elapsed fortnight. If miners switch off and hash rate falls, blocks can take longer to arrive, lengthening the wait for the next retarget. The adjustment uses the observed timestamps of the first and last blocks in the preceding interval: the target adjustment multiplier is clamped between one-quarter and four, subject to the proof-of-work target ceiling. A single adjustment therefore need not fully compensate for a hash-rate change. A higher target means lower difficulty, which can improve expected revenue per unit of hash, but profitability still depends on the block subsidy and transaction fees, bitcoin price, hardware efficiency, electricity and other operating costs. Lower difficulty alone neither guarantees that remaining miners become profitable nor explains why hash rate later recovers.
Primary sources for the retarget rules: Bitcoin Core v29.0 proof-of-work adjustment code, mainnet parameters, and difficulty interval definition.
The honest tension this book won't resolve for you
In the spirit of Chapter 2's both-sides treatment: whether fees alone can eventually fund a security budget as large as today's subsidy-heavy one is a real, actively debated question among people who study Bitcoin's economics, not a settled fact either direction here can responsibly claim. One side of the debate expects transaction volume and per-transaction value to grow enough, over the many decades before era 34, that fee revenue scales up to replace the shrinking subsidy. The other side worries that block space is capped by design, so fee revenue can't simply grow without limit the way subsidy-in-dollars theoretically could if the price rose — and that a security budget built entirely on a capped, competitive auction might end up smaller, and more volatile, than one anchored by a predictable subsidy.
Nothing in this chapter's arithmetic settles that question, because it isn't an arithmetic question — it depends on how much of the world actually ends up transacting on the base chain versus on layers built on top of it (a later chapter covers one such layer). What the arithmetic does settle is the mechanism: two revenue sources, one shrinking on a fixed public schedule, the other set by an open auction for scarce space, both paid for the same underlying job of making the chain expensive to attack.