Institutional Bitcoin Demand Falls to 23.8, Third Week Down
Are institutions buying bitcoin right now?
Yes, but the evidence for it is about to get much thinner. bryptoflows' composite institutional bitcoin demand score reads +23.8 on a -100 to +100 scale as of the September 17 close, down from +28.9 a week ago and +55.2 two weeks ago. The model still classifies that as moderate accumulation. All eight components were computable on this run — the missing-components list is empty and weight coverage is 1.0 — so this is a complete composite and directly comparable to both earlier readings.
The number itself is the least interesting thing about it. What matters is that 93% of the flow currently propping up this score leaves the measurement window within five trading sessions, on data that already exists. The live breakdown is on the institutional signal dashboard.
Where the institutional bitcoin demand score sits, component by component
The composite fuses eight components, each normalised to -100..100, weighted, and summed. Here is the full contribution breakdown in points of the final 23.8, as of the September 17 close:
- Supply absorption, 20% weight, normalised +84.0, contributing +16.8. ETFs absorbed 24,367 BTC over the 20-session window against 9,000 BTC issued — a ratio of 2.71x.
- ETF flow momentum, 24% weight, normalised +46.9, contributing +11.2. Net flow of $1,935.4M over the same 20 sessions.
- Treasury capacity, 6% weight, normalised +95.8, contributing +5.7. Strategy's $44.34B of undrawn disclosed financing against a position the model marks at $66.05B.
- Treasury accumulation, 14% weight, normalised +18.0, contributing +2.5. 4,603 BTC acquired across the last four filings, on holdings of 845,050 BTC.
- Flow persistence, 10% weight, normalised +24.5, contributing +2.4. A positive flow streak exactly one session long.
- Price confirmation, 6% weight, normalised -0.5, contributing -0.03. Its 20-observation price window nets out to -0.08% — as close to nothing as this component gets.
- Flow concentration, 6% weight, normalised -53.8, contributing -3.2. A product HHI of 0.788 against an even-split value of 0.083.
- Flow breadth, 14% weight, normalised -83.3, contributing -11.7. One of twelve products positive.
Absorption and momentum together contribute +28.0 of the 23.8 — more than the entire score, with the remaining six components netting negative. That concentration of explanatory weight in two trailing windows is the structural fact this column has flagged for three weeks running, and it is why the roll-off below matters more than any single session's flows.
Ninety-three percent of the score's support rolls off in five sessions
The composite's 20-session window currently runs August 20 to September 17. The five oldest sessions in it — August 20, 21, 24, 25 and 26 — carry $1,797.9M of the window's $1,935.4M net flow and 23,445 of its 24,367 absorbed BTC. That is 92.9% of the momentum reading and 96.2% of the absorption reading sitting in five days that are five sessions from expiry.
Recomputing the two components from the model's own published normalisation — a calculation that reproduces the live figures to the decimal — gives the sensitivity:
- If the next five sessions were flat, the 20-session net flow falls from $1,935.4M to $137.5M, and momentum drops from +46.9 to +3.6, costing about 10.4 points of composite.
- Absorption falls from 2.71x to 0.10x — well below the parity that marks its neutral point — taking the component from +84.0 to about -56.6 and costing roughly 28.1 points.
- Holding every other component still, the composite lands near -15, having crossed zero.
Flows will not be exactly flat, and this is a sensitivity rather than a forecast: one strong session resets a meaningful part of it. But it establishes which way the mechanical pressure runs, and the asymmetry is severe. Last Friday's signal piece put this scenario ten sessions out. It is now five, because the late-August block has aged to the front of the queue. Tuesday's supply piece tracked the same decay from the absorption side, where the ratio has already fallen from 5.24x on September 14 to 2.71x today.
The shorter windows say the erosion is underway rather than hypothetical. Net flow over the last three sessions is -$586.8M and over five sessions -$440.1M, while the 10-session figure is +$15.8M — effectively zero. Week-to-date stands at -$426.9M and the year-to-date total is -$1,797.5M. The 20-session number is the only flow window still reading clearly positive.
The biggest drag came on a day that was net positive
Flow breadth is the composite's largest negative contributor at -11.7 points, and the reading behind it deserves a closer look than the number invites.
On September 17, ETF flows were net positive at +$159.5M. BlackRock's IBIT took in $183.7M. Fidelity's FBTC gave back $16.6M and VanEck's HODL $7.6M. The other nine products reported exactly zero. Breadth counts positives against all reported products, so one of twelve scores -83.3, near the component's floor — on a session where more money arrived than left.
This is the same construction quirk this column flagged on September 11, and it cuts both ways. A day where eleven funds are genuinely selling and a day where nine funds simply report no creations or redemptions produce a near-identical breadth score, and they are not the same market. The honest reading is that the US spot complex is not broadly bid right now — that much is true, and Monday's flow piece documented how quiet the whole shelf has become — but the component overstates how negative September 17 specifically was.
Concentration says the structural half of this more cleanly. A product HHI of 0.788 against an even-split value of 0.083 means flow is bunched into effectively one vehicle. Across the full 689-session history, IBIT accounts for 57.9% of gross flow and 116.8% of cumulative net flow, a figure above 100% because Grayscale's GBTC has bled -$27.8B and drags the denominator down. Institutional bitcoin buying through this channel is, increasingly, one fund's order book.
Why the components suddenly agree more than they did
The model's agreement figure — which measures how much the eight components corroborate one another, not a confidence interval around the score — rose to 43.7 from 34.2 on September 15 and 34.3 on September 10. It was 61.0 on September 3.
Rising agreement alongside a falling score is not the reassuring combination it sounds like. The components are converging because the trailing engine is finally catching down to what the acute components have been saying for two weeks, not because the acute components improved. Price confirmation is the exception and the clearest single move: it read -31.7 on a -5.9% 20-day return when Wednesday's macro brief covered it, and now reads -0.5 on a window that nets to roughly flat. Bitcoin traded at $78,305 on Hyperliquid as this was written, up 2.3% on the day and 13.0% over 30 days.
So the composite's price input has gone from a real drag to a neutral one, while its two largest inputs are beginning a decline that has not finished. That is the shape of the week: the score fell 5.1 points despite its most improved component, because the components that carry the weight moved against it.
What the composite cannot see
It sees US spot ETFs and one company's treasury. Direct institutional purchases, offshore vehicles, structured products, futures basis positioning, and every corporate treasury other than Strategy's sit outside the dataset entirely. Deutsche Bank announced on September 16 that it plans to launch regulated digital-asset custody for institutional and corporate clients in Germany, covering bitcoin among other assets, still subject to BaFin approval (Bitcoin.com). That is a real bitcoin institutional adoption datapoint and this composite will never register it, because plumbing that lowers the cost of holding bitcoin only shows up here if and when it converts into US ETF creations.
Its treasury components run on a weekly filing cadence. The treasury data is current through Strategy's September 14 8-K, which disclosed no purchase — holdings flat at 845,050 BTC, average cost $75,412, as Thursday's treasury piece covered in detail. The accumulation component's four-filing window still contains the August 31 purchase of 4,603 BTC, which is the entire source of its +2.5 contribution. That purchase drops out of the window two filings from now. If Strategy keeps directing cash to buybacks rather than bitcoin, this component decays to zero on its own schedule, independent of anything ETFs do. Full filing history is on the MSTR treasury page.
Treasury capacity measures permission, not intent. The +5.7 points it contributes reflect $44.34B of undrawn financing disclosed as of August 31 — capacity that could fund purchases, not evidence that any are planned. It has been a positive contributor through weeks in which Strategy bought nothing.
The macro channel is invisible to it. The Fed raised its target range 25 basis points to 3.75%-4% on September 16 in a 12-0 vote, its first hike since July 2023, with 16 of 18 participants projecting at least one more (CNBC). None of that enters the composite except through the flow and price channels it can actually measure — which is precisely why the score can look stable through a repricing and then move sharply once flows respond.
What would move the bitcoin demand indicator next week
In rough order of how much each could shift the score:
- Whether flows replace the late-August block. Roughly $360M per session over five sessions would hold the 20-session window where it is. Anything materially below that mechanically lowers the score regardless of sentiment, and this is by far the largest lever.
- Whether breadth broadens. Two or three funds printing positive on the same day moves breadth from -83.3 toward neutral and is worth several points. It also distinguishes a genuinely bid market from an IBIT-only one, which matters more for the bitcoin ETF outlook than the headline dollar total.
- Strategy's Monday 8-K. A purchase would lift treasury accumulation and, more importantly, break a seven-week run of capital going to buybacks instead. No purchase leaves the component drifting toward zero.
- Flow persistence. The current positive streak is one session against a component that saturates near four. A short run of consistent inflows is worth a few points on its own.
- Follow-through on the Fed repricing. The hike is now known. What is not known is whether the September 17 inflow was a genuine reset or a single session inside a downtrend.
What this score is not
The composite is a directional proxy for institutional bid strength. It is not a price forecast, there is no mechanical link between the score and future returns, and nothing here is investment advice. A reading of +23.8 says the measurable institutional channel is still net accumulating, on a window whose support is concentrated and expiring — not that bitcoin will rise or fall.
Two limits are worth restating plainly. The composite reads twelve US ETF products and one balance sheet, so it is a measure of one channel rather than of institutional demand in general. And its two heaviest components are 20-session trailing windows, which means the score is designed to lag: it will confirm a turn well after the turn, in both directions. That is the correct trade-off for a demand indicator, but it makes the roll-off arithmetic above more informative than the current level.
All ETF flow, absorption, treasury and composite figures are read directly from bryptoflows' own API, current through the September 17 close and Strategy's September 14 filing, with the spot price quoted live. Every macro and corporate item is sourced to the reporting linked inline and retrieved September 18, 2026. Current readings stay live on the overview and institutional signal dashboards.