US spot bitcoin ETFs recorded net outflows on two consecutive sessions, 15 and 16 September, totalling $746.3 million — $450.4 million on the 15th and $295.9 million on the 16th.
Over the same stretch, futures open interest rose rather than fell. The money left the spot side; the leverage did not. That combination is the point of this post.
The disagreements, stated first
Two data providers put almost the same number on the 15th, but describe the previous record differently:
| Source | Net outflow, 15 September | Says the last larger one was |
|---|---|---|
| Farside Investors | $450.4 million | 24 June |
| SoSoValue | $450.33 million | 25 June |
The amounts differ by under $100,000, which is close enough to corroborate. The reference dates differ by a day, so we list both rather than pick one.
By fund on the 16th (Farside): BlackRock’s IBIT shed $144.1 million, ARKB $84.4 million, Fidelity’s FBTC $52.7 million and Grayscale’s GBTC $18.2 million. Only Morgan Stanley’s MSBT took money in, at $3.5 million.
Open interest: two figures that will not reconcile
Coverage produced two open interest numbers, and they do not agree:
- One puts BTC futures open interest at $36.4 billion, above its $36.0 billion high band.
- The other has open interest rising from 676,000 BTC to 688,000 BTC.
At the roughly $76,000 price of the time, 688,000 BTC is about $52 billion, which does not reconcile with $36.4 billion — the two figures cover different sets of exchanges. So we take only the direction here: open interest is rising. Which absolute figure is right, we have no third source to adjudicate, so we do not adjudicate.
The one piece we could pull ourselves: BTC perpetual open interest on Hyperliquid stands at 35,642 BTC at a mark price of $77,617, about $2.77 billion (read at 2026-09-18 07:13 UTC from Hyperliquid’s public API). That is one venue and cannot stand in for the whole market, but at least we counted it ourselves.
The other widely quoted figure is perpetual cumulative volume delta at −$605.9 million. Rising open interest alongside negative CVD is usually read as traders actively adding shorts.
Funding: longs neither fled nor came back
This is the piece we can add. ETF flows tell you spot demand is retreating and open interest tells you positions are growing, but only the funding rate tells you what longs are willing to pay to hold.
Per-period settlements for the event window (2026-09-15 00:00 to 09-18 00:00 UTC, ten periods), from Binance’s public futures API:
- BTCUSDT: all ten positive, 0.060495% cumulative, high 0.009812%, low 0.002788%
- 30-day average 0.006816% per period, with just 1 negative period out of 90
Longs did not disappear. Had that $746 million of ETF outflow represented broad long capitulation, funding would already be negative. It is not.
But longs did not come back either. The highest of those ten periods did not reach 0.01%, barely above the 30-day average. A market that shook out more than seven hundred million dollars of institutional money in two days saw essentially no change in what leveraged longs were bidding — which is itself a reading. It says this episode played out in spot and institutional channels rather than in the retail leverage book.
ETH looks different: four of the same ten periods were negative, low of −0.008189% (details in the Fed hike post).
What it means for a balanced two-leg position
ETF flows do not enter our model. A balanced two-leg position does not care who is buying or selling. It cares about one thing: whether funding turned negative. BTC did not turn negative in any of these ten periods.
There is exactly one path by which ETF outflows would reach us: if sustained institutional selling depresses spot and the perpetual falls further still, the premium flips, and funding becomes shorts paying longs. That has not happened so far.
The thing to stay alert to is the other side of it: with open interest elevated and liquidity tightening, a cascade of forced liquidations makes the seconds in which both legs must move very expensive. That is execution risk rather than funding risk, and we wrote about it in what makes delta-neutral hard to run.
On this year’s scale
Weekly flows are extremely noisy. Binance BTCUSDT’s 781 periods this year total 2.030026%, which across 260 days annualises to 2.85%, with 209 of those periods negative.
That number says more than any single week of ETF flow: long demand has been flat all year. It did not deteriorate over these two days; it has been at this level throughout. For comparison, in the same week the Federal Reserve lifted the federal funds target range to 3.75%–4.00%.
What to watch
No forecasts, three observables:
- Whether the outflows transmit to funding. Judge on the per-period average, not single-period peaks. BTC’s 30-day figure is 0.006816%; below 0.003% is when it is genuinely thin.
- Rising open interest with falling funding is a combination worth noting. It implies new positions are predominantly short. If it goes that way, funding reaches zero first and turns negative after.
- ETH and SOL move before BTC. Across September’s 52 periods, BTC has 1 negative, ETH has 8 and SOL has 22.
Funding thinning toward zero or turning negative in flat and bear markets is something this approach genuinely runs into — ETHUSDT’s 0.787% line for 2022 on our track record page is exactly that, and we have not hidden it. It did not transmit this time. That does not mean it will not next time.
Sources: daily ETF net flows from Farside Investors (as cited by Forbes and CVJ.AI), with the SoSoValue figure via CryptoGazette, through 16 September 2026; the two open interest figures and the CVD figure come from those same reports, and since the two open interest values cannot be reconciled we take direction only; Hyperliquid open interest and mark price read directly from its public API (2026-09-18 07:13 UTC); per-period funding from Binance’s public futures API (same timestamp, latest settlement 2026-09-18 00:00 UTC); federal funds target range per the Federal Reserve FOMC statement of 16 September 2026; yearly totals on our track record page, on-site snapshot timestamp 2026-08-25 13:42 UTC.