Supply and Scarcity · Published

Bitcoin Supply Shock Stalls: ETFs Hand Back 3,949 BTC

How much bitcoin did ETFs absorb against new supply this week?

They did not absorb any. Across the five reporting sessions from September 8 through September 14, US spot bitcoin ETFs returned a net 3,949 BTC to the market while the network minted 2,250 — an absorption ratio of -1.76x. Over the same stretch the headline twenty-session reading, the one the site's own chart draws, still says 5.24x. Both numbers are correct. The bitcoin supply shock visible in the one-month series is a rear-view artifact of one extraordinary week in August, and that week is about to roll out of the window.

This is the arithmetic, unchanged: each session's net ETF flow is converted to bitcoin at that day's close, then divided by the 450 BTC the network mints daily. Above 1.0 the funds took more coins off the market than miners created; below zero, redemptions handed coins back. bryptoflows builds the series from Farside Investors flow data and Hyperliquid closes. The rolling absorption chart lives on the institutional signal dashboard and the underlying product-level flows on the ETF flow dashboard. Everything below is computed from that dataset, current through the September 14 session.

The last five sessions returned coins, they did not take them

Session by session, in coins rather than dollars:

Four of five sessions removed nothing and gave coins back. Monday's $159.9M inflow was the only session on the right side of zero, and it recovered only 55 percent of Thursday's return. Monday's flow analysis described the dollar side of this as a complex that had very nearly stopped trading; measured in coins, the same period was not dormant at all. It was net supply-returning, and the two descriptions sit together because gross activity collapsed while the small amount that did trade leaned consistently one way.

Why the move through parity did not print

Last week's piece argued that the trailing sixty-session window would rise through 1:1 on arithmetic alone, as June's redemptions aged out of the back of it. That argument was half right, and the half that failed is the more instructive one.

The mechanism was real. On September 4 the sixty-session window held 20,729 BTC, a ratio of 0.768x, and its five oldest sessions — June 11 through 17 — carried a net -1,099 BTC between them. Rolling those five off does lift the window: to 21,829 BTC, or 0.808x, had nothing whatsoever arrived to replace them.

Something did arrive, and it was negative. The window now reads 17,879 BTC, or 0.662x — lower than it was a week and a half ago, and lower than the do-nothing counterfactual by exactly the 3,949 coins the last five sessions handed back. The roll-off tailwind was worth about four hundredths of a ratio point. The incoming week was worth fifteen against it. A trailing window only rises mechanically when the front of it is quiet, and the front of it was not quiet; it was busy in the wrong direction.

That reading now sits at the 36th percentile of the 627 rolling sixty-session observations recorded since January 2024 — below the median, and a long way from the parity crossing that looked close a week ago. The honest summary is that the calendar effect described last week was correctly identified and too small to matter at the magnitudes that actually turned up.

The bitcoin supply shock now faces its roll-off in reverse

The twenty-session window is where this gets sharp, because the same mechanism that was a modest tailwind for the quarter is about to become a severe headwind for the month.

That window covers August 17 through September 14 and holds 47,120 BTC against 9,000 mined. But 27,228 BTC of it — 57.8 percent — came from five consecutive sessions, August 17 through 21, when the complex absorbed between 2,926 and 8,306 coins a day. Those five sessions roll off over the next five. Holding net flow at exactly zero from here, the ratio decays like this:

To hold the twenty-session reading at today's 5.24x, the ETF complex has to replace all 27,228 of those coins inside five sessions — roughly $2.13B of net inflow at Monday's $78,182 close. It managed -$302.8M over the five sessions just completed. A fall from 5.24x toward the low twos over the next fortnight is not a forecast about demand; it is what the recorded data already implies unless something changes materially.

Meanwhile the sixty-session window is set up the opposite way. Its next five sessions to roll off carry -23,580 BTC, so five flat sessions would lift it from 0.662x to 1.54x, and ten flat sessions to 2.14x. The one-month reading is about to fall hard and the three-month reading is about to rise hard, on the same data, from the same demand, purely because of what is aging out of each. Anyone quoting either number in two weeks should be made to say which window it is and how much of the move was the calendar.

This matters beyond the chart. On the composite institutional score, supply absorption is currently normalised at 99.5 out of 100 and contributes 19.9 points of the 44.6 total — close to half the reading, from the single component with the most calendar risk in it. Last Friday's signal piece covered the breadth collapse underneath that score; the point here is that the component holding it up is the one with a known decay schedule.

How much bitcoin do ETFs buy over a full cycle?

The rolling windows swing violently. The cumulative figure does not, and it is the more durable expression of bitcoin scarcity from this channel.

Since January 2024, net ETF creations converted at each day's close total 659,067 BTC. At 450 a day that is 1,465 days of production — just over four years' worth — held by a product category thirty-two months old, and about 3.14 percent of the 21 million coins that will ever exist. Life to date, absorbed against issued, the ratio is 1.93x.

The annual breakdown shows how uneven the accumulation has been:

That last line is the one to keep in view whenever a one-month reading above 5x appears. Across 175 sessions this year, the US spot ETF complex has been a net supplier of bitcoin to the market, not a net absorber, by more than twenty-six thousand coins. September has genuinely improved on that — nine sessions, +5,491 BTC, a ratio of 1.36x — but improving on a deficit is not the same as erasing it. Bringing 2026 back to flat from here needs roughly 26,300 more coins.

Bitcoin daily issuance already halved once inside this series

The denominator is the only fixed part of the calculation, and this dataset happens to straddle the step. The series opens at 900 BTC per day and drops to 450 on April 22, 2024, so both regimes are measured on identical methodology:

The counterintuitive result is that the higher ratio belongs to the era with the larger denominator. That was the launch quarter, when demand was genuinely exceptional, and it is a useful corrective to the assumption that a halving mechanically produces a higher absorption reading. It does — for identical demand. The same 450-era coins measured against a 900-a-day denominator would read 0.80x instead of 1.60x, and today's 5.24x would read 2.62x. The next halving will double every ratio on this page overnight without one additional dollar arriving. That is a denominator story, and it will be widely reported as a demand story.

Strategy is not filling the gap

The second channel that can remove coins faster than miners create them has been inactive. On bryptoflows' Strategy dataset, the three most recent weekly 8-K filings — August 24, September 8 and September 14 — all report a direction of none: zero bitcoin bought, zero sold. The last purchase was the August 31 filing, 4,603 BTC at an average $80,318, and before that the company was a net seller, releasing 1,638 coins on August 3 and 1,690 on August 10. Holdings stand at 845,050 BTC.

So across the same five sessions in which ETFs returned 3,949 coins, the largest corporate holder added none. Thursday's treasury coverage owns that subject properly; the point here is only that nothing in the corporate channel offset what the fund channel gave back.

What the absorption ratio does not count

Four limits matter, and one of them is doing real work this week.

The fifth caution is the one this week supplies for free. Over the same twenty sessions in which absorption read 5.24x — the 80th percentile of 667 readings since January 2024 — bitcoin's price fell 2.7 percent. The strongest one-month supply absorption figure in months coincided with a lower price. Whatever the ratio measures, it is not a mechanical input to price, and this window is a clean demonstration of why it should not be read as one.

The summary is straightforward. The bitcoin supply shock in the headline chart is real but backward-looking: five sessions in August account for most of it, those sessions are rolling out, and the most recent week of actual demand ran negative. The three-month picture will improve over the next fortnight and the one-month picture will deteriorate, largely for calendar reasons in both directions. The cumulative number — 659,067 coins, four years of issuance, 3.14 percent of the eventual supply — is the one that has not moved much, and it remains the most honest single statement of what this channel has done to bitcoin scarcity.