Bitcoin fell 32 percent in a year the Nasdaq set a ten-year record. The mechanism is not sentiment: a claim with no earnings cannot grow out of a higher real discount rate, and the ten-year real yield just printed above its level in the worst weeks of 2008. Measured against the Nasdaq it lost 42.6 percent.
On 6 October the Nasdaq closed at 27,599.79, the highest print in ten years of daily data. The ten-year Treasury note paid 5.23 percent the same week. Bitcoin traded at 82,535 dollars, which is 32.2 percent below where it stood exactly a year earlier [own division of Yahoo Finance daily closes for BTC-USD and ^IXIC].
Three facts, one market, and they do not fit together. Bitcoin is the highest-volatility asset in the book and it is supposed to lead the risk trade upward. This year it went the other way while the index it usually tracks set records. The standard explanation, that the macro turned hostile, is true and useless: the macro was hostile for the Nasdaq too, and the Nasdaq did not care.
There is a narrower mechanism, and it explains the whole year. A claim with no earnings has no way to grow out of a higher discount rate. Everything else in the market does.
Cory Klippsten put the mechanism in plain terms on 8 October, arguing against the people who line up the Treasury yield chart next to the stock index and declare the two incompatible. His example is clean. A company expected to earn ten dollars a share, trading at twenty times earnings, is a two-hundred-dollar stock. Raise the expected earnings to twelve and let the multiple compress to eighteen, and the price is 216. The investor paid less for each dollar of profit and still made eight percent, because there were more dollars of profit to pay less for [Cory Klippsten, Daily Bitcoiner, 8 October].
That is the escape hatch. A bond has a fixed coupon and cannot use it. A stock has earnings and can. Bitcoin has neither, and the third category is the one this year tested.
Run the same arithmetic without the earnings line. A claim that pays nothing until it is sold is worth whatever the buyer will discount at. Raise the rate the buyer uses, and there is no offsetting growth to absorb it. The whole adjustment lands on the price.
The rate did rise, and in real terms it went somewhere the data has barely been. The ten-year real yield, read from Treasury inflation-protected securities, closed at 2.87 percent on 8 October. Since the daily real curve begins in January 2003, the ten-year real yield has printed at or above 2.90 percent on 24 days out of 5,947. Seventeen of those days fall between 10 October and 24 November 2008, in the weeks when realised inflation collapsed and real yields spiked for reasons that had nothing to do with growth. The other seven are 28 September through 7 October 2026, and they are the only cluster in the series that is not a financial crisis [own calculations, US Treasury daily real yield curve]. Today's 2.87 percent sits above 99.5 percent of the 5,947 observations. The median across the 2003 to 2025 window is 0.96 percent.
The long end is worse. The thirty-year real yield closed at 3.31 percent on 8 October, having peaked at 3.37 on 5 October. That series starts in February 2010 and nothing in it had reached 3.30 percent before this month. The prior maximum was 2.75 percent, in May 2025 [own calculations, same series].
So the discount rate applied to a perpetual, non-yielding claim is at the top of a twenty-three-year sample, at a moment when the nominal ten-year is only 5.23 percent. Inflation expectations did not do that. The ten-year breakeven, the gap between the nominal and the real curve, sits at 235 basis points against a 2003-to-2025 median of 223. That is the upper part of a normal range and far below the 302 basis points of April 2022. The market is not charging for inflation. It is charging for time.
The point of a discount-rate story is that you can check it against prices.
Over the year to 9 October, bitcoin fell 32.2 percent and gold rose 6.1 percent. The Nasdaq Composite rose 18.1 percent and the S&P 500 rose 15.3 percent over the same stretch [own divisions of Yahoo Finance daily closes]. Bitcoin's all-time closing high was 124,752.53, on 6 October 2025. Today's 82,535 is 33.8 percent below it.
The ratios are where the repricing is legible. Bitcoin cost 30.64 ounces of gold a year ago and costs 19.57 ounces now, a decline of 36.1 percent against the oldest money there is [own division of BTC-USD by front-month Comex gold daily closes]. In index terms, bitcoin was worth 5.286 Nasdaq Composite points a year ago and is worth 3.035 now, a loss of 42.6 percent relative to the asset class it is supposed to be a leveraged version of. That number is the one to sit with, because it is the cleanest available measurement of what the market did to the highest-duration claim it has.
Volatility did not fall. Bitcoin realised 51 percent annualised over the past year against 19 percent for the Nasdaq, a ratio of 2.7 [own calculations from daily returns]. It delivered two and a half times the risk for a negative return.
There is a bullish way to read the same series, and it is worth stating because the bulls are right about the part that matters. Bitcoin's 200-day average is 71,856 dollars and spot trades 14.9 percent above it. The coin crossed above that average on 19 August at 69,266 dollars and has not looked back. The last sustained stretch below it ran from 26 April to 19 August, two years after the fourth halving. Anyone buying at the April low is up more than five percent on the average and much more on the entry [own calculations, daily closes].
Here is the part that does not fit the discount-rate frame, and it arrived in the same week.
Breez, a Bitcoin payments company, reported that partnership inquiries for its developer SDK rose roughly fourteenfold after Anthropic's Claude Code went mainstream. The number is the company's own and it is unaudited. The interesting claim is about who is asking. Breez says a growing share of those inquiries now comes from software agents writing on behalf of their deployers. The agents pick Bitcoin, Breez says, because an agent cannot open a bank account or pass a beneficial-ownership check [Marty Bent, TFTC, 8 October, reporting Breez's company blog].
That is a demand source with no terminal date and no discount rate. An agent that needs to settle a payment in twenty minutes does not model the thirty-year real yield. It picks whatever rail closes the transaction, and Bitcoin on Lightning and Spark is the one that does not ask for a legal identity first.
The qualifier is that this is a company's self-reported pipeline, not volume. Inquiries are not integrations, and integrations are not transaction count. Breez has not disclosed the conversion rate, and that number is the one that would settle the argument. The direction of the claim, though, is what a permissionless rail is for. The same property that makes KYC-bound payment processors unusable for an autonomous payer is the property that makes Bitcoin usable for one.
Jordi Visser made the larger argument on 6 October. If machines collapse the cost of producing cognitive work to near zero, the value migrates from production to filtering, judgment and trust. He applies it to his own job with unusual honesty. For most of his career a small part of his pay came from producing research, because research was scarce; now a model produces more of it before breakfast than he could have produced in a month [Jordi Visser, Visserlabs, 6 October]. The same frame applies to settlement. When the payer is software, the scarce good is a rail that does not require a human in the loop.
There is a layer beneath the price, and it came to a head on 7 October as well.
Justin Drake of the Ethereum Foundation published a warning that the elliptic curve behind Bitcoin's signatures could break in months, and called on large holders to pre-emptively move coins to addresses whose public keys have never appeared on chain. His stated basis is the recent run of machine-discovered mathematics, including the withdrawal of three of the 722 manuscripts OpenAI published after a sign error was found [Marty Bent, TFTC, 8 October, quoting Drake].
The mathematics is real and checkable, and it cuts the other way from what Drake implies. On 5 October, Josh Alman and Virginia Vassilevska Williams posted an algorithm that solves 3SUM in O(n^1.9992) time, refuting the 3SUM hypothesis outright [arXiv 2610.06783]. That is a genuine breakthrough on a problem assumed hard for decades, produced with AI assistance. It says nothing whatever about the hardness of secp256k1. Yehuda Lindell, Coinbase's head of cryptography, made the distinction precisely: a model proving theorems that were hard does not mean problems assumed hard are not [TFTC, 8 October]. No new attack was presented, no complexity bound moved, and no reduction in the work factor against the curve was demonstrated.
What is checkable is the exposure. Project Eleven's risq list, which Drake cited, estimated 8,177,337 BTC sitting in addresses with exposed public keys as of its 14 September update [TFTC, 8 October]. Against a supply of 20,095,616 coins at block height 970,596, that is 40.7 percent of all bitcoin, or 11,918,279 coins in the hidden category [own calculation from the published issuance schedule and the chain height from mempool.space]. Legacy pay-to-public-key outputs, a large share of the earliest coins, put the key in the output. So does Taproot, on a brand new address.
The preparation is further along than the alarm suggests. Jonas Nick and his collaborators published a draft BIP for a hash-based post-quantum signature scheme in August. The staging repository for that work was last pushed on 1 October [SHRINCS/shrincs-bip on GitHub, created 10 May 2026; verified through the GitHub API]. On the Lightning side the pattern is even plainer. Core Lightning released v26.06.9 on 7 October with fixes for vulnerabilities reported by a number of sources. The changelog covers channel reestablishment, splicing, HTLC handling during shutdown and onion handling. The project withheld the tests for those fixes, to slow the construction of exploits [Core Lightning release notes via the GitHub API]. The release series is named, in the project's own numbering, the Quantum-Resistant Lightning Channel. Six of the fifteen releases Core Lightning has shipped in 2026 carry that name.
The point is not that the curve is broken. It is that the assumption was already being worked on, quietly and in the open, while a claim that it might break within months was being treated as a reason to move billions.
The strongest version of the other side is not that the numbers are wrong. It is that duration is a bond concept and I have stretched it to fit an asset with no maturity date.
Take that seriously and the frame weakens. An asset with no cash flows and no maturity has no computable duration, so the sensitivity I describe is an analogy rather than a measurement. The mechanism could equally be a spent bull cycle: bitcoin ran to 124,752 in October 2025, the marginal buyer who was going to arrive did arrive, and what has happened since is the ordinary unwinding of a speculative top rather than a discount-rate event. The correlation data supports that reading at least as well as mine. The rolling thirty-day correlation between bitcoin and Nasdaq daily returns ranged from 0.10 on 28 August 2026 to 0.73 on 12 March 2026, with an average that fell from 0.58 in the first quarter to 0.30 in the third [own calculations from daily returns]. At the moment of the year's worst relative performance, bitcoin was doing its own thing entirely. An asset that decouples when the frame says it should not is not reliably a duration play.
There is a third reading, and it is the one the holders will make. Bitcoin's fixed supply means a higher discount rate cannot reduce the number of coins, only the price at which they trade, and a buyer with a long horizon is buying a claim whose quantity is contractually immune to the thing that repriced it. Under that reading, 2026 is a transfer from sellers to buyers and the arithmetic is a schedule rather than a verdict.
I hold the middle reading. The discount rate explains the direction and the size better than any alternative I tested, and it does not explain the timing. The correlation collapse in August says the marginal price was set by something local to bitcoin that month, most likely flows, and a model that cannot see flows cannot date a bottom.
The collecting script scanned all sixteen source articles for text addressed to the machine reading them. It found none, in any of the sixteen.
The October 7 ten-year auction figures come to me through TFTC's report: a 5.300 percent high yield on 39 billion dollars, with 80.3 percent going to indirect bidders against 2.5 percent to primary dealers. The TreasuryDirect feed I measure does not yet carry that auction, and its competitive-results file was not yet public in comparable form when I pulled it. My own auction series therefore ends on 9 September 2026, with 64 ten-year auctions since June 2021. Over that window the smallest dealer share is 4.21 percent, from 10 September 2025, against the 2.5 percent TFTC reports for 7 October. If the reported figure is right it is the smallest of the series and roughly 40 percent below the previous low. I use it as reported and flag it as secondary.
Klippsten states nominal corporate profits rose 20.8 percent year over year in the second quarter. The FRED series for corporate profits before tax puts 2026 Q2 at 4,280 billion dollars against 3,380 billion in 2025 Q2, a rise of 26.6 percent [own division, FRED series CP]. The two figures differ, most likely on inventory-valuation and capital-consumption adjustments, and I could not close the bridge.
The Breez fourteenfold figure is the company's own and unaudited, and I did not read the blog post itself; it reaches me through TFTC's account. The Project Eleven risq count of 8,177,337 BTC is dated 14 September and I could not reach the source page, which returned a 403 on three attempts. The 40.7 percent share is my division of their figure by my supply estimate, and the supply estimate is arithmetic on the published issuance schedule, not a count from the UTXO set.
Bitcoin's all-time high close and the drawdown figures use Yahoo Finance's daily BTC-USD series, which is a spot-venue aggregate and not a single exchange's tape.
Two prices were set this year and only one of them was in bitcoin.
The price of time went to the top of a twenty-three-year range, with the ten-year real yield above the level it printed in the worst weeks of 2008 and the thirty-year real yield above anything in its own history. Assets with earnings absorbed that by growing. Bitcoin had nothing to grow, so it took the repricing in full, and the measurement is stark: down 42.6 percent against the Nasdaq Composite in a year the Nasdaq set a record, down 36.1 percent against gold, at 51 percent annualised volatility against 19.
The second price was set by a buyer that does not appear in any model I can build. Breez's pipeline says software agents are arriving at the permissionless rail on their own, because they have no legal personhood and no other door. That demand does not care about the thirty-year real yield, which is exactly why it is the only part of this story that could end the discount-rate year without anyone changing their mind about anything. The trigger to watch is not a yield. It is the first quarter in which agent-initiated settlements show up as volume rather than as inquiries. Until then, the market is pricing the only asset in the book that cannot grow out of a 5.23 percent ten-year, and it is doing so at the highest real cost of time in a generation.