The money that left last week came back.
Per SoSoValue, US spot bitcoin ETFs saw net inflows of $433.03 million on 18 September (US Eastern) and $159 million on 17 September, roughly $592 million across the two days.
Against the two days we wrote about last week: $450.4 million out on 15 September and $295.9 million out on 16 September, $746.3 million in total. The direction reversed completely inside four trading days.
Price followed. Bitcoin jumped on 18 September and was still above $81,000 on Saturday.
Funding did not.
The inflow is extremely concentrated
Four outlets citing SoSoValue agree to within a decimal point:
| Item | Figure |
|---|---|
| 18 September net inflow | $433.03M |
| Fidelity FBTC | $310.72M (71.8%) |
| BlackRock IBIT | $108.44M |
| Those two combined | $419.16M, 96.8% of the day |
| Cumulative net inflow | $55.161B |
| Total net assets | $102.532B, 6.29% of BTC market cap |
Bitwise’s BITB took in $9.69 million; VanEck’s HODL and the ARK/21Shares ARKB were each under $3 million, and the rest were essentially flat. Grayscale’s GBTC has seen $27.84 billion of cumulative net outflow since conversion.
In practice, the money went into two funds.
Liquidations: four windows disagree, the direction does not
The cut-off times differ, so the four figures are some distance apart. We are listing the disagreement rather than splitting the difference:
| Source | 24h liquidations | Short share |
|---|---|---|
| CoinOTAG (as of 09-19 03:02 UTC, BTC perps only) | $283.09M | 96.75% |
| Gokhshtein (Friday, whole market) | $445M | BTC shorts $230M |
| CryptoTimes (as of 09-19 15:08 UTC) | $324.88M | shorts ~$256M |
| Gate / CoinGlass (four-hour window) | ~$250M (shorts only) | — |
The absolute values differ by a factor of two, and all four point the same way: what got cleared out was overwhelmingly shorts. CryptoTimes also gives 101,582 traders affected, with $87.32 million of ETH and $68.68 million of BTC liquidated.
By the usual reasoning, large spot buying plus a short squeeze plus a price break above a round number should push funding up.
Funding: four periods pinned at 0.010000%, then the fifth slipped
We re-pulled the settlements across the event window from Binance’s public futures endpoints (read 2026-09-20 14:11 UTC, latest settlement 09-20 08:00 UTC):
| Settlement (UTC) | BTCUSDT | ETHUSDT | SOLUSDT |
|---|---|---|---|
| 09-18 16:00 | 0.006853% | 0.010000% | 0.010000% |
| 09-19 00:00 | 0.006643% | 0.010000% | 0.010000% |
| 09-19 08:00 | 0.010000% | 0.010000% | 0.010000% |
| 09-19 16:00 | 0.010000% | 0.005416% | 0.010000% |
| 09-20 00:00 | 0.010000% | 0.010000% | 0.010000% |
| 09-20 08:00 | 0.010000% | 0.004128% | 0.010000% |
| Currently running | 0.008140% | 0.003275% | 0.010000% |
BTC printed exactly 0.010000% four periods in a row, and the period now underway has fallen back to 0.008140%.
Why “exactly 0.01%” is not good news
This is the whole point of the note.
Binance’s formula for USDT-margined contracts is:
funding rate = premium index + clamp(interest rate − premium index, −0.05%, +0.05%)BTCUSDT’s interest rate is 0.01% per 8 hours (the API’s interestRate field currently returns 0.0001). Unfolding that:
- When the premium index sits between −0.04% and +0.06%, the clamp does not bind and funding comes out exactly 0.01%.
- So a reading of exactly 0.010000% carries zero information — it says the premium was somewhere in that dead band, and nothing about how eager the longs were.
- A reading below 0.01% is the informative case: it requires the period-average premium index to have fallen through −0.04%, meaning the perpetual traded at a meaningful discount to spot.
Working the running period’s 0.008140% back through the formula puts the average premium index at roughly −0.0419%.
Prices at the moment of the read agree: BTCUSDT’s mark price was 80,593.70 against an index of 80,630.61, the mark sitting 0.0458% below the index. ETHUSDT was −0.0641% and SOLUSDT −0.0476%. All three perpetuals were trading at a discount.
The money came back, the rent did not
Put the two sides together and the conclusion is direct:
The ETFs buy spot, not perpetuals. The $592 million over two days went into spot custody accounts. That lifts the price; it does not directly lift the perpetual’s premium. The premium requires leveraged longs chasing, and this round did not have them.
More bluntly: the price was pushed up by spot buying and by shorts being forced to close. Neither of those pays funding — spot buyers were never in the perpetual market, and liquidated shorts are exiting positions, not building new leveraged longs.
Which is why our note two days ago said the same thing in different numbers: that time funding fell below baseline, this time it sat on baseline for four periods and then slid off. Two different readings of one situation.
What it means for a balanced two-leg position
Same as last time: nothing needed handling this round, and there was no good news either.
A spot long plus a futures short does not care whether BTC is 76k or 81k. It cares about one thing: did funding turn negative. It did not — BTC, ETH and SOL were positive across all four of the 09-19 to 09-20 periods.
But that is all. A half-billion-dollar ETF reversal plus a short squeeze bought BTC funding touching the baseline and then sliding off it. For a position whose only income is funding, that is a weak reading.
On an annual scale: BTCUSDT’s 788 periods year to date come to +2.091425%, or 2.910% annualised — below the lower bound of the federal funds target range over the same period.
What to watch next
No forecasts. Three observables:
- Whether a single BTC period genuinely clears 0.01%. That requires the premium index above +0.06%. It has not happened once across this year’s 788 periods, and the period that does it is the turn signal.
- Whether ETF inflows string together three days. Two funds accounted for 96.8% of the two-day total, and inflows that concentrated reverse easily.
- When the mark price gets back above the index. All three symbols are at a discount right now, and that moves before the funding reading does — inside the dead band funding flattens to 0.01% and whatever the premium is doing in there is invisible.
Funding approaching zero or turning negative in flat and bearish markets is something this approach genuinely runs into — the 0.787256% ETHUSDT line for 2022 on the track record page is how that happened, and SOLUSDT is at −0.960962% year to date. Funding did not turn negative this time; that does not mean it will not next time.
Sources: ETF flows all cite SoSoValue, via Bloomingbit, PANews, ChainCatcher and KuCoin (all on 18 September, US Eastern), with the previous two days’ outflows in our earlier note; the four liquidation windows come from CoinOTAG, Gokhshtein, Crypto Times and Gate, with different cut-off times; period-by-period funding, mark and index prices and interestRate from Binance’s public USDT-margined futures endpoints (read 2026-09-20 14:11 UTC, latest settlement 09-20 08:00 UTC); the funding formula is Binance’s published definition; yearly totals on the track record page, site snapshot as of 2026-08-25 13:42 UTC. Past performance does not indicate future results.