On 14 August, Bitcoin++ posted a talk Peter Todd had given in Toronto on 23 July, titled Tail Emissions and Demurrage. Within two days the twenty one million supply cap was being relitigated across the industry.
That argument is about the block subsidy. In the same week, two numbers landed that say the block subsidy is the one part of bitcoin's security budget behaving exactly as it was designed to.
Transaction fees fell to 0.69 percent of miner revenue, a ten year low. And the largest listed miners reported quarters in which most of their revenue no longer came from mining at all.
Both of those are the security budget. Neither of them is the subsidy.
Hashpower got more expensive.
1 The Layer Nobody Is Arguing About
The case for a tail emission rests on a projection. When the subsidy approaches zero, fees have to carry security. That is not a 2140 question. It is a measurable series, and it is going the wrong way.
Fees are 0.69 percent of miner revenue. In April they touched 0.52 percent. They have been under 1 percent for almost a year. Glassnode's Rafael Schultze-Kraft put the comparison in the only terms that matter: the last time the fee share sat this low, bitcoin traded below $400.
So the fee market is not merely failing to grow into its role. Measured as a share of what pays for the chain, it is smaller than it was ten years ago.
This matters more than a ratio suggests. A fee share of 0.69 percent means block space demand is contributing almost nothing to the cost of attacking bitcoin. Every argument about what happens when the subsidy runs out is, at present, an argument about a revenue line that does not exist.
2 What AI Actually Changed
The second thing that landed was a pair of income statements.
Core Scientific reported second quarter colocation revenue of $136.7 million against $27.5 million from bitcoin mining, so colocation was 83 percent of sales, up from 67 percent in the first quarter. TeraWulf reported $31.9 million of high performance computing lease revenue against $12.8 million from mining, 71 percent of the total, up from 62 percent.
These are not companies leaving the industry. They are the same companies, on the same sites, with a different tenant.
In aggregate, TheEnergyMag estimates the listed miner cohort's realized hashrate fell from 368.3 exahash per second in the fourth quarter of 2025 to 319.0 in the second quarter of 2026, a 13.4 percent decline in six months. Excluding Bitdeer, which grew 44 percent over the same window to 63.0 exahash, the rest of the cohort fell 21.2 percent, from 324.6 to 255.9 exahash.
The reading in circulation is that bitcoin's security budget problem has arrived twenty years early. That reading has the mechanism wrong, and the correction is the point of this paper.
The security budget is the block subsidy plus fees, multiplied by the price of bitcoin. Demand for AI compute touches none of those three terms. Not the subsidy, which is fixed by the halving schedule. Not fees, which are set by demand for block space. Not the price.
What it touches is the cost of the input. Power and datacenter shells now have a second bidder, and that bidder pays more per megawatt than mining does at a hashprice near thirty two dollars per petahash per day. The same budget, denominated in exactly the same dollars, buys less hashrate than it did a year ago.
That is a narrower claim than the one being made, and it survives better. It requires no forecast about 2140. The repricing already happened, on two income statements, this quarter.
3 Framework · The Three Layers
Bitcoin's security is paid for in three layers. They are set by different things, they are moving in different directions, and only one of them has a constituency.
| Layer | What sets it | Direction in 2026 | Who is arguing about it |
|---|---|---|---|
| Block subsidy | The halving schedule, fixed in 2009 | Exactly on schedule | Everyone, this month |
| Fee market | Demand for block space | 0.69 percent of miner revenue, a ten year low | Almost nobody |
| Hashpower opportunity cost | The best alternative use of the power and the shell | Newly contested, and rising | Nobody |
Read down the last column. The only layer with a constituency is the only layer that is working. A tail emission, whatever its merits, is a proposal to enlarge the first row. The two layers that moved this year are the two nobody has proposed anything about.
There is a version of this that connects to ground we have already covered. Our first and fifth papers described treasury companies selling bitcoin to service a perpetual coupon, with no operating cash flow and one use for the asset on the balance sheet. Miners carry a cost of production and are structural sellers for a similar reason. The difference is what they hold. A treasury company has one use for its bitcoin. A miner has two uses for its datacenter. That optionality is why the bid for hashpower is now set outside bitcoin, and it is not available to anyone whose only asset is the coin.
4 What Eight August Already Answered
The objection to any emission change is usually stated as politics. The twenty one million cap is a social contract, and the industry will not move it. That is true, and it is also unnecessary to argue, because the experiment ran this month.
On 8 August the bitcoin blockchain split into two chains for roughly eighteen hours over BIP-110. The mechanism of the failure is worth stating precisely. OCEAN had added two endpoints on 8 July to signal for the proposal. A Stratum configuration flaw meant the pool's default settings routed unassigned hashpower to what became the minority chain. Around 0.3 bitcoin of miner rewards were affected, OCEAN pledged rebates within seventy two hours, and it restored its default endpoint to the chain without BIP-110 while leaving both available by choice.
BIP-110 is now a stalled minority chain.
Six days later the tail emission talk was posted and the industry began arguing about whether the cap could ever move. It had just been tested. The answer took eighteen hours and cost 0.3 bitcoin. How it died is the other half of the lesson. Not in a vote and not in debate, but in a pool's default configuration. Consensus change in bitcoin is decided by where unassigned hashpower points when nobody has made a choice.
5 The Other Side
Four arguments cut against this paper, and the third is the one that would break it.
Difficulty adjustment is working. The retarget on 8 August raised difficulty by 0.99 percent to 127.48 trillion, the following adjustment was estimated near 128.06 trillion, and the seven day average network hashrate was around 920 exahash in the middle of August. Whatever happened across the last two quarters, it is not continuing at the same rate. A network that reprices its own cost of production every two weeks is not obviously fragile.
Falling hashrate is also not a falling budget. The dollar budget is set by price, not by exahash. Fewer machines competing for the same reward means higher margin for the machines that remain, which is the self correction functioning as designed. Our claim is the narrower one: the budget buys less hashrate than it did, not that the budget shrank.
The strongest counterargument is reversibility. Core Scientific and TeraWulf did not sell their sites. They changed tenant. If hashprice rose far enough, that capacity could return, which would make the converted fleet a latent security reserve rather than a loss. The reason we do not think this fully defuses the point is that colocation and compute revenue arrives under multi year leases. A lease is a contractual lock rather than an economic preference, and a leased datacenter cannot be flipped back to mining because hashprice improved. That is checkable, and it is the single fact that decides whether any of this is permanent. It also cuts in an uncomfortable direction. A large fleet of convertible datacenters is a security reserve and an attack provisioning pool at the same time.
And the sample is thin. Two quarters, two income statements, and a cohort that is roughly a third of the network. Working from the same figures, hashrate outside the listed cohort fell from about 702.7 to 638.0 exahash across the same six months, a 9.2 percent decline against the listed cohort's 13.4 percent. Everyone shed. The listed miners shed faster, and within that cohort Bitdeer absorbed part of what the others gave up. Some of this is reallocation rather than disappearance, and two quarters is not a trend.
We would separate the two claims here accordingly. That the security budget is being repriced through the cost of hashpower rather than the size of the subsidy is an analytical claim, and it holds on two quarters or on eight. How far it has been repriced, and whether it is permanent, rests on a sample we would not defend hard.
6 What Would Settle It
- Whether the fee share breaks back above 1 percent and stays there. It has been below for almost a year, and every argument about the era after the subsidy is an argument about this line.
- The duration of the colocation and compute leases at Core Scientific and TeraWulf. Contract length, not hashprice, decides whether the converted capacity is recoverable.
- Whether a third and a fourth listed miner cross fifty percent of revenue from something other than mining. Two firms is a case. Four is a category.
- The difficulty trajectory through the third quarter. It turned slightly positive in August, and another quarter of decline in the listed cohort against a flat network would mean the hashpower is being replaced by operators who have no alternative use for the power.
- Whether any consensus change clears signalling after BIP-110. The next attempt tells us whether 8 August was an anomaly or the template.
If the fee share stays where it is and the leases turn out to be long, then the security budget was never a 2140 problem, and the layer that needs an answer is the one nobody has proposed anything about. If fees recover, or the leases are short, this was a cyclical quarter with an unusually good story attached to it.
7 Sources
Figures verified against the linked reporting. Retrieval 19 August 2026.
- Cointelegraph, Bitcoin miners earn under 0.7% of revenue from fees in new 10-year low, 12 August 2026. The 0.69 percent fee share, the 0.52 percent April reading, the run of almost a year below 1 percent, the comparison to 2016 and to a sub $400 bitcoin price, and the Glassnode and Checkonchain data behind them.
- Bitcoin.com News, Public miners shed 21% of bitcoin hashrate as AI revenue accelerates, 16 August 2026. TheEnergyMag cohort estimates of 368.3, 344.4 and 319.0 exahash, the 13.4 percent six month decline, network quarterly averages of 1,071, 993 and 957 exahash, Bitdeer's 44 percent growth to 63.0 exahash, the 324.6 to 255.9 exahash series excluding Bitdeer, Core Scientific's $136.7 million of colocation against $27.5 million of mining at 83 percent of sales from 67 percent, and TeraWulf's $31.9 million of lease revenue against $12.8 million at 71 percent from 62 percent.
- Bitcoin.com News, OCEAN pledges BTC refunds after chain split error, 10 August 2026. The 8 August split, its roughly eighteen hour duration, the two endpoints added on 8 July to signal for BIP-110, the Stratum configuration flaw, the 0.3 bitcoin affected, the seventy two hour rebate pledge, and BIP-110 as a stalled minority chain.
- Bitcoin.com News, Peter Todd's tail emissions pitch sparks bitcoin inflation debate, 16 August 2026. The 23 July Toronto talk posted by Bitcoin++ on 14 August, Todd's proposal of a fixed number of new coins per block and his cited annual loss rate near 0.1 percent, Adam Back's position, and Eli Ben-Sasson's argument from lost keys.
- Hashrate Index, Hashrate Index roundup, 17 August 2026. The 8 August retarget to 127.48 trillion at plus 0.99 percent, the following estimate near 128.06 trillion, the 920 exahash seven day average, and hashprice near thirty two dollars per petahash per day.
- OX85, The death spiral of the digital asset treasury, July 2026, and What does the wrapper cost?, August 2026. The cost of carry framing and the comparison of single use collateral.
On sourcing, and on what we did not use. Positions attributed to named individuals are close paraphrase of the linked reporting, not direct quotation. Several claims circulating with this story are excluded. A widely repeated line describing a tail emission as a dangerous trap dressed in technical logic is not Adam Back's wording; his recorded comment concerns how inadvisable proposals are sold through narratives that are simple but false, which is a different argument. The figure of 0.21 bitcoin per block at 0.05 percent annual inflation originates with a participant in the debate rather than with Todd, whose own cited number was an annual coin loss rate near 0.1 percent. A further named participant appearing in material we reviewed does not appear in the cited reporting and is excluded. A difficulty peak of 156 trillion in November 2025, and a figure of more than 32,000 bitcoin sold by listed miners in the first quarter, could not be verified against a primary source and are not used. The series for hashrate outside the listed cohort in section five is our own arithmetic on the cited cohort and network figures, not a published series.
This paper is published for informational purposes and does not constitute investment, legal or tax advice. It is not a recommendation regarding any security or digital asset. Figures are as of the dates stated and may have changed.