Twenty-eight million on the day the price broke

Bitcoin traded as low as $82,941 on Thursday and was near $84,800 late in the session, having given back the $84,000 it recovered on Tuesday for the first time since 31 January. The Treasury's par yield curve settled the same afternoon at 5.18 per cent for the ten-year, 5.53 for the twenty and 5.47 for the thirty. This desk wrote this morning that the long end had stopped standing still, on Wednesday's close of 5.11. It moved another seven basis points while we were publishing.

The nineteen-year description survives the extra move and gets stronger. On the Federal Reserve's daily ten-year series back to 1962, the last reading at or above 5.18 per cent was 6 July 2007. The highest print of the entire post-crisis era before this month was 4.98 per cent, in October 2023.

That is one half of Thursday. The other was a set of fund-flow numbers that appeared to say the opposite. Spot bitcoin exchange-traded funds took in about $4.6bn between 19 August and 22 September on Farside Investors' daily table — enough to turn 2026's cumulative flow positive for the first time. On our sum of those rows the crossover came on Tuesday 22 September, when a $715m day carried the year from minus $365m to plus $349m. Farside's site refused automated requests to this desk and the table was read through a reader proxy: the figures are Farside's, the arithmetic ours.

The arithmetic is also where the story turns. The rebound is not a wall of money meeting a falling price. It is a four-day decay: $999m on Monday, the largest single day of the year, then $715m on Tuesday, $347m on Wednesday and $28m on Thursday, with BlackRock's fund flat. The bid did not fight the sell-off. It stopped on the same day.

The table also shows concentration. Of the roughly $4.97bn taken in since 19 August, some $3.62bn — 73 per cent — went into the iShares fund alone, while across the year Grayscale's original vehicle lost $2.59bn and Fidelity's $1.10bn. The net figure that has just turned positive is a small residue of two large and opposite migrations.

Now the part that matters more than either. The usual framing is that bitcoin has failed a hedging test the bond market set for it. The data supports something worse. Of the roughly hundred basis points the ten-year has added this year, some eighty-three are real yield and about eight are inflation expectations: the inflation-protected ten-year has gone from 1.93 per cent at the end of December to 2.76 on Wednesday, its highest since November 2008, while the breakeven crept from 2.25 to 2.33. This is a discount-rate move, not a debasement move, and a discount-rate move reprices anything whose entire value is terminal.

Which is why gold is down too. Bullion sat near $4,290 an ounce on Thursday, about 1.7 per cent lower on the year and 22 per cent below its January peak, falling alongside bitcoin on Wednesday and Thursday both. There is no version of this week in which the metal passed a test the token failed. Both are zero-coupon stores of value marked against a real rate at an eighteen-year high. Equities with earnings are not: the Nasdaq set a record on Tuesday and is up about 16 per cent on the year.

That comparison is where the long-duration argument needs care. On daily moves bitcoin trades like a high-beta technology stock: on this desk's own calculation, correlation with the Nasdaq of about 0.43 across 2026 and 0.72 in September alone, and a correlation with daily changes in the ten-year real yield of minus 0.35 this month against minus 0.15 for the year. Day to day it is a duration instrument, and an increasingly obedient one. On levels it is nothing of the kind: down about 4 per cent in 2026 and a third below its October 2025 high of $126,080, while the index it tracks sits 1 per cent off a record. That is a leveraged proxy for risk appetite, not an asset behaving like the ones it moves with.

What is producing the creations, then? Not leverage: perpetual funding on the offshore venues printed flat on Thursday, which is not what a crowded long looks like. The forward curve is more suggestive, with annualised basis running from roughly 4.4 per cent at five weeks to 5.1 per cent at a year — essentially level with the ten-year Treasury. A cash-and-carry buyer clears a thin positive spread that widens with tenor, consistent with part of the inflow being delta-neutral rather than a view on the price. Consistent is not proved, and the exchange whose data would settle it blocks automated access.

The bull reading, made by Bitcoin.com on Thursday, is that $80,000 to $85,000 is now support underwritten by fund demand, the funds' average cost being around $82,000 — a figure this desk could not verify. But an average cost three per cent below spot describes how little cushion those holders have, not how much. And the flows were not defending anything on Thursday. They were $28m.

The honest summary is that price and flow have been out of phase in both directions all month. A week ago the analytics firms were writing about spot strength against fund outflows. On Tuesday the flows turned positive and the price was at its high. By Thursday the price had broken and the flows had gone. Each leg has been the contrary indicator for the other, which is another way of saying neither tells you much about the asset — while the thing that has been telling you something all year, the real yield on a ten-year Treasury, went up again.