What Bought Bitcoin's Q3 2026 Three-Month Streak: $5.0B of US Spot ETF Inflows (July Just $173M, August $3.53B, September $1.31B to Sept 21), Strategy Net Zero BTC (846,000 on Both June 30 and Sept 20), CME Leveraged Funds Even Shorter, Stablecoins +$2.3B; Flows Are ~1% of the $558B Market-Cap Gain

etf

The parent card is wrong on its face: CoinDesk's 23 Sept piece (Omkar Godbole) is about a three-month WINNING streak (July, August and September 2026 all higher, last seen in 2012), not a daily losing streak, and the price it quotes is $85,450 with September up about 10%. Bitcoin's longest daily losing streak of 2026 was seven sessions ending 6 June (Coin Metrics daily closes), and the all-time record is 12 days ending 30 July 2014, so nothing about the last month is a losing-streak record either. What the streak statistic itself means is covered elsewhere; this card measures WHO actually bought the streak, month by month, from primary sources, and compares it with the only precedent, July-September 2012.

The scale of the move first. On Coin Metrics' daily series, Bitcoin's market capitalisation was $1.173 trillion at the 30 June close ($58,525, the trough of the 53% drawdown from the 6 Oct 2025 high of $124,824) and $1.732 trillion at the 22 Sept close ($86,205): roughly $558 billion of market value was added in 84 days. Monthly closes were +7.4% in July, +24.9% in August and +9.8% for September to date (CoinDesk quotes +4.8%, +25.2%, +10.9% on its own index; the difference is source and cut-off time, not substance).

US spot ETFs (The Block's per-issuer daily flow data, summed by month, last print 21 Sept): July net inflow was only $172.8 million, so the first green month of the streak happened with the ETF bid essentially absent, and IBIT (+$228M) was offset by GBTC (-$360M) and FBTC (-$210M). August was the flow month: +$3.53 billion, of which IBIT took $3.09 billion, FBTC $325 million and Grayscale's mini BTC $210 million. September to the 21st is +$1.31 billion, but it is lumpy: -$450 million on 15 Sept, -$296 million on 16 Sept, then +$433 million on 18 Sept and the single +$999 million day on 21 Sept that produced most of the month's total. The quarter's ETF take is about $5.0 billion, which at each month's average Coin Metrics price is roughly 2,706 BTC in July, 50,812 BTC in August and 16,577 BTC in September, about 70,094 BTC in total. For context, ETFs bled $2.41 billion in May and $4.51 billion in June, so Q3 has only just recovered June's outflow. CoinDesk's "over $5.5 billion since August" (SoSoValue) is the same picture on a different provider and date range.

Corporate treasuries, the other bid of 2024-25, contributed net nothing. Strategy's 8-K filed 6 July shows 846,000 BTC held at 30 June after selling 1,363 BTC on 29-30 June, then a further 2,225 BTC sold on 1-5 July (at an average $60,773) to fund preferred dividends, taking holdings to 843,775. Its 21 Sept 8-K shows 950 BTC bought on 14-20 Sept at $79,670 and aggregate holdings of exactly 846,000 BTC as of 20 Sept, with $174 million spent on STRC buybacks that week. Between those two filings it bought back the 2,225 it had sold (including 4,603 BTC in the week of 24-30 Aug at $80,318 funded by $369.7 million of MSTR ATM sales) and nothing more: the largest corporate holder starts and ends the streak at the same 846,000 BTC. Bitcointreasuries.net puts all public-company holdings at 1.273 million BTC ($108.8 billion) on 23 Sept, and the disclosed 2026 sellers (MARA, Riot and other miners funding AI capex) mean the non-Strategy cohort was a net seller over the period, so treasuries as a group were at best flat.

CME futures did not carry the move either. In the CFTC's Traders in Financial Futures data for CME Bitcoin (5 BTC contracts, code 133741), open interest went from 18,336 contracts on 30 June to 20,773 on 15 Sept (about 92,000 to 104,000 BTC notional), a 13% rise against a 47% price rise. Leveraged funds (hedge funds and CTAs) were net short 5,313 contracts on 30 June and net short 6,354 on 15 Sept (5,545 long vs 11,899 short), so the fast-money book got shorter as price rose, consistent with basis-trade hedging against ETF longs rather than directional buying, and it rules out a CME short squeeze as the driver. Asset managers went from net long 2,000 to net long 2,760 contracts, and dealers from net long 4,525 to net long 3,419. Total reportable traders: 111. Nothing in the futures book looks like new institutional length of a size that could move a $1.7 trillion asset.

Stablecoins: DefiLlama's aggregate USD-pegged supply was $309.6 billion on 30 June, $305.4 billion on 31 July (down $4.2 billion during the first green month), $308.8 billion on 31 Aug and $311.8 billion on 22 Sept. Net change over the quarter is +$2.3 billion (+0.7%), and the July dip means the streak began while dry powder in stablecoin form was shrinking. Spot volume (CoinGecko's aggregate across exchanges) fell during the rally: $953 billion in June, $819 billion in July, $821 billion in August, and $757 billion in the first 24 days of September ($31.5 billion a day vs $26.5 billion in August). Coin Metrics' reported spot volume shows the same shape ($301 billion June, $187 billion July, $216 billion August, $185 billion Sept to date). Realised volatility from daily closes was 31.5% annualised in July, 41.3% in August and 45.8% in September; Deribit's DVOL index sat between 35 and 39 for the whole quarter, ending 23 Sept at 37.6, so the streak was bought at low and stable implied vol. On-chain activity was flat: daily active addresses averaged 627,000 in July, 652,000 in August and 632,000 in September, and transaction counts rose modestly from 656,000 to 701,000 a day.

Put the identifiable flows together: about $5.0 billion of ETF inflow, zero net from Strategy, a net-selling treasury cohort, a shorter leveraged-funds book at CME and $2.3 billion more stablecoins, against $558 billion of added market value. Measured dollar buying accounts for around 1% of the capitalisation change. That is normal, not suspicious: Bitcoin's marginal price is set by a thin flow against a largely dormant float, and a 47% repricing needs only a modest excess of marginal demand when spot volume is falling and sellers (miners, treasuries needing cash) are exhausted after a 53% drawdown. But it also means the streak has almost no "sticky" ownership behind it. The August ETF flow was the one month with real absorption; July was a flow-less short-covering-free bounce off the June low, and September's positive number rests on a single $999 million day.

The 2012 comparison the headline invites is a comparison of different objects. In July-September 2012 Bitcoin's market cap went from $62 million (30 June close, $6.68) to $125 million (30 Sept, $12.39) on Coin Metrics data, a gain of $63 million, with daily active addresses of 35,000-44,000 and transaction counts of about 29,000-35,000 a day. Bitstamp, then the largest non-Gox venue, traded $0.4 million in July, $0.9 million in August and $0.6 million in September 2012 in USD terms; Coin Metrics' reported spot volume for the whole quarter is about $60 million. Realised vol was 56% in July, 173% in August (the Pirateat40 collapse) and 29% in September. There were no ETFs, no CME contract (launched Dec 2017), no stablecoins (Tether launched 2014) and no corporate holders. The 2012 October dip to $10.13 on 26 Oct and the 2,177% rise to $230.68 on 9 April 2013 happened in a market where a few million dollars was the entire order book. Today's streak was bought by roughly 70,000 BTC of ETF demand into a market trading $25-30 billion of spot a day; a similar percentage outcome would require an order of magnitude more flow than any quarter on record, and the actual flow composition of Q3 2026 (thin July, one big August, lumpy September, no corporate or futures length) argues for treating the streak as a low-volume repricing rather than an accumulation phase.

Sources (6)

https://www.coindesk.com/markets/2026/09/23/bitcoin-s-on-a-streak-it-hasn-t-hit-since-2012https://www.theblock.co/api/charts/chart/crypto-markets/bitcoin-etf/spot-bitcoin-etf-flowshttps://www.sec.gov/Archives/edgar/data/1050446/000119312526396093/mstr-20260914.htmhttps://www.sec.gov/Archives/edgar/data/1050446/000119312526295586/mstr-20260706.htmhttps://www.cftc.gov/dea/futures/financial_lf.htmhttps://stablecoins.llama.fi/stablecoincharts/all

AI Research

Key Takeaway

Q3 2026's three green months were bought by about $5.0B of US spot ETF inflow (almost all in August and on one $999M day in September), with Strategy net zero (846,000 BTC on both June 30 and Sept 20), CME leveraged funds net shorter and stablecoins up only $2.3B; identifiable flows are ~1% of the $558B market-cap gain, so the streak is a thin-volume repricing off the June low, not a measured accumulation phase, and 2012's precedent was a $63M market with no ETFs, futures or stablecoins.

The parent card is wrong on its face: CoinDesk's 23 Sept piece (Omkar Godbole) is about a three-month WINNING streak (July, August and September 2026 all higher, last seen in 2012), not a daily losing streak, and the price it quotes is $85,450 with September up about 10%. Bitcoin's longest daily losing streak of 2026 was seven sessions ending 6 June (Coin Metrics daily closes), and the all-time record is 12 days ending 30 July 2014, so nothing about the last month is a losing-streak record either. What the streak statistic itself means is covered elsewhere; this card measures WHO actually bought the streak, month by month, from primary sources, and compares it with the only precedent, July-September 2012.

The scale of the move first. On Coin Metrics' daily series, Bitcoin's market capitalisation was $1.173 trillion at the 30 June close ($58,525, the trough of the 53% drawdown from the 6 Oct 2025 high of $124,824) and $1.732 trillion at the 22 Sept close ($86,205): roughly $558 billion of market value was added in 84 days. Monthly closes were +7.4% in July, +24.9% in August and +9.8% for September to date (CoinDesk quotes +4.8%, +25.2%, +10.9% on its own index; the difference is source and cut-off time, not substance).

US spot ETFs (The Block's per-issuer daily flow data, summed by month, last print 21 Sept): July net inflow was only $172.8 million, so the first green month of the streak happened with the ETF bid essentially absent, and IBIT (+$228M) was offset by GBTC (-$360M) and FBTC (-$210M). August was the flow month: +$3.53 billion, of which IBIT took $3.09 billion, FBTC $325 million and Grayscale's mini BTC $210 million. September to the 21st is +$1.31 billion, but it is lumpy: -$450 million on 15 Sept, -$296 million on 16 Sept, then +$433 million on 18 Sept and the single +$999 million day on 21 Sept that produced most of the month's total. The quarter's ETF take is about $5.0 billion, which at each month's average Coin Metrics price is roughly 2,706 BTC in July, 50,812 BTC in August and 16,577 BTC in September, about 70,094 BTC in total. For context, ETFs bled $2.41 billion in May and $4.51 billion in June, so Q3 has only just recovered June's outflow. CoinDesk's "over $5.5 billion since August" (SoSoValue) is the same picture on a different provider and date range.

Corporate treasuries, the other bid of 2024-25, contributed net nothing. Strategy's 8-K filed 6 July shows 846,000 BTC held at 30 June after selling 1,363 BTC on 29-30 June, then a further 2,225 BTC sold on 1-5 July (at an average $60,773) to fund preferred dividends, taking holdings to 843,775. Its 21 Sept 8-K shows 950 BTC bought on 14-20 Sept at $79,670 and aggregate holdings of exactly 846,000 BTC as of 20 Sept, with $174 million spent on STRC buybacks that week. Between those two filings it bought back the 2,225 it had sold (including 4,603 BTC in the week of 24-30 Aug at $80,318 funded by $369.7 million of MSTR ATM sales) and nothing more: the largest corporate holder starts and ends the streak at the same 846,000 BTC. Bitcointreasuries.net puts all public-company holdings at 1.273 million BTC ($108.8 billion) on 23 Sept, and the disclosed 2026 sellers (MARA, Riot and other miners funding AI capex) mean the non-Strategy cohort was a net seller over the period, so treasuries as a group were at best flat.

CME futures did not carry the move either. In the CFTC's Traders in Financial Futures data for CME Bitcoin (5 BTC contracts, code 133741), open interest went from 18,336 contracts on 30 June to 20,773 on 15 Sept (about 92,000 to 104,000 BTC notional), a 13% rise against a 47% price rise. Leveraged funds (hedge funds and CTAs) were net short 5,313 contracts on 30 June and net short 6,354 on 15 Sept (5,545 long vs 11,899 short), so the fast-money book got shorter as price rose, consistent with basis-trade hedging against ETF longs rather than directional buying, and it rules out a CME short squeeze as the driver. Asset managers went from net long 2,000 to net long 2,760 contracts, and dealers from net long 4,525 to net long 3,419. Total reportable traders: 111. Nothing in the futures book looks like new institutional length of a size that could move a $1.7 trillion asset.

Stablecoins: DefiLlama's aggregate USD-pegged supply was $309.6 billion on 30 June, $305.4 billion on 31 July (down $4.2 billion during the first green month), $308.8 billion on 31 Aug and $311.8 billion on 22 Sept. Net change over the quarter is +$2.3 billion (+0.7%), and the July dip means the streak began while dry powder in stablecoin form was shrinking. Spot volume (CoinGecko's aggregate across exchanges) fell during the rally: $953 billion in June, $819 billion in July, $821 billion in August, and $757 billion in the first 24 days of September ($31.5 billion a day vs $26.5 billion in August). Coin Metrics' reported spot volume shows the same shape ($301 billion June, $187 billion July, $216 billion August, $185 billion Sept to date). Realised volatility from daily closes was 31.5% annualised in July, 41.3% in August and 45.8% in September; Deribit's DVOL index sat between 35 and 39 for the whole quarter, ending 23 Sept at 37.6, so the streak was bought at low and stable implied vol. On-chain activity was flat: daily active addresses averaged 627,000 in July, 652,000 in August and 632,000 in September, and transaction counts rose modestly from 656,000 to 701,000 a day.

Put the identifiable flows together: about $5.0 billion of ETF inflow, zero net from Strategy, a net-selling treasury cohort, a shorter leveraged-funds book at CME and $2.3 billion more stablecoins, against $558 billion of added market value. Measured dollar buying accounts for around 1% of the capitalisation change. That is normal, not suspicious: Bitcoin's marginal price is set by a thin flow against a largely dormant float, and a 47% repricing needs only a modest excess of marginal demand when spot volume is falling and sellers (miners, treasuries needing cash) are exhausted after a 53% drawdown. But it also means the streak has almost no "sticky" ownership behind it. The August ETF flow was the one month with real absorption; July was a flow-less short-covering-free bounce off the June low, and September's positive number rests on a single $999 million day.

The 2012 comparison the headline invites is a comparison of different objects. In July-September 2012 Bitcoin's market cap went from $62 million (30 June close, $6.68) to $125 million (30 Sept, $12.39) on Coin Metrics data, a gain of $63 million, with daily active addresses of 35,000-44,000 and transaction counts of about 29,000-35,000 a day. Bitstamp, then the largest non-Gox venue, traded $0.4 million in July, $0.9 million in August and $0.6 million in September 2012 in USD terms; Coin Metrics' reported spot volume for the whole quarter is about $60 million. Realised vol was 56% in July, 173% in August (the Pirateat40 collapse) and 29% in September. There were no ETFs, no CME contract (launched Dec 2017), no stablecoins (Tether launched 2014) and no corporate holders. The 2012 October dip to $10.13 on 26 Oct and the 2,177% rise to $230.68 on 9 April 2013 happened in a market where a few million dollars was the entire order book. Today's streak was bought by roughly 70,000 BTC of ETF demand into a market trading $25-30 billion of spot a day; a similar percentage outcome would require an order of magnitude more flow than any quarter on record, and the actual flow composition of Q3 2026 (thin July, one big August, lumpy September, no corporate or futures length) argues for treating the streak as a low-volume repricing rather than an accumulation phase.

Related Feeds

The 'Two More Tests' After Friday's $15.6B Deribit Expiry Are Scheduled Prints, Not Chart Levels: Durable Goods 12:30 UTC, UMich Final (Prelim 47.8, 1-Yr Inflation 4.6%) 14:00, CME Sept Futures Cash-Settle 15:00 on ~$8.9B COT OI; Last 4 Quarterly Expiries Fell 4/4 on D-1 and Rose 4/4 by D+14

The thin card says Bitcoin's rally faces two tests after Friday's expiry: 'post-expiry repositioning' and 'resistance near key technical levels'. Its own source (CryptoSlate, 23 Sept 2026) says something different and more checkable: the tests are a fixed clock of scheduled events on Friday 25 Sept: US durable goods at 12:30 UTC, the University of Michigan final September sentiment at 14:00 UTC, and CME's September Bitcoin futures cash settlement at 15:00 UTC, all landing after Deribit's 08:00 UTC expiry. None of them is a chart level. This card measures each one. 1. The Deribit leg, from the live order book rather than a vendor screenshot. Deribit's public API (get_book_summary_by_currency, pulled 23 Sept ~12:25 UTC) shows 183,039 BTC of open interest in the 25SEP26 expiry, 36.7% of all 498,900 BTC of BTC options OI on the venue. At the Deribit index of $85,431 that is $15.64B notional, which is where the '$16B' headline comes from. Calls are 107,239 BTC ($9.16B, 58.6%) and puts 75,799 BTC ($6.48B), a raw put/call ratio of 0.71. The 0.52 put/call figure the article quotes comes from ByKaranteli's gamma model, not the exchange aggregate; the two are not the same statistic. The largest single strike is $70,000 (8,792 calls, 8,133 puts), followed by $90,000 (7,286 calls, 559 puts), $100,000 (6,950 calls), $85,000 (6,734 calls) and $80,000 (6,230 calls); the largest put strike is $60,000 (5,571 puts). Minimising aggregate option payoff across the live strikes gives a max-pain strike of $76,000, roughly 11% below spot. The important point the headline misses: 63% of Deribit OI does not expire Friday. The 30OCT26 book already holds 105,172 BTC with a put/call ratio of 0.31 (80,418 calls vs 24,754 puts) and 25DEC26 holds 118,371 BTC at 0.60. Whatever dealer gamma survives Friday is more call-skewed than what expires, so the 'hedges disappear and the market is naked' framing overstates the reset. 2. The CME leg is smaller than it sounds and cannot deliver anything. CME Bitcoin futures are cash-settled against the CME CF Bitcoin Reference Rate (BRR), which CF Benchmarks printed at $86,037.52 at 15:00 GMT on 22 Sept; the September contract terminates on the last Friday of the month at 16:00 London, i.e. 15:00 UTC on 25 Sept. The CFTC Traders in Financial Futures report as of 15 Sept 2026 shows total CME Bitcoin futures open interest of 20,773 contracts across all months (5 BTC each, 103,865 BTC, about $8.9B at $85.4K), down 310 contracts on the week, with 111 reporting traders. The September contract is only a fraction of that, since most positions roll to October and December in the week before termination. Positioning is the familiar basis-trade shape: leveraged funds are 5,545 long vs 11,899 short (net short 6,354 contracts, 31,770 BTC), asset managers 4,528 long vs 1,768 short, dealers 6,587 long vs 3,168 short. Micro Bitcoin futures add 37,455 contracts at 0.1 BTC (3,746 BTC). A cash settlement of a net-short-by-hedgers book against a 1-hour London TWAP is a roll event, not a liquidity event. 3. The macro prints are second-tier, and the one that matters comes next week. The Fed raised the funds target by 25bp to 3.75-4.00% on 16 Sept 2026 (12-0 vote), so inflation expectations are the live variable. The University of Michigan preliminary September reading (published 11 Sept) was sentiment 47.8, down 7.5% on the month and 13.2% on the year, with year-ahead inflation expectations jumping to 4.6% from 4.0% (highest since June) and long-run expectations 3.4% from 3.3%; the final print is scheduled for Friday 25 Sept at 10:00 ET (14:00 UTC). Final UMich revisions are usually within a point of the preliminary. The Census Bureau's Advance Durable Goods report for August is the only Census release that day, at 08:30 ET (12:30 UTC). The heavyweight, August PCE, is not on Friday at all: BEA's release schedule has Personal Income and Outlays for August on Wednesday 30 Sept at 08:30 ET, the same day as quarter-end. If a rate-sensitive repricing is going to hit Bitcoin, the calendar says 30 Sept, not 25 Sept. 4. Base rates for quarterly expiry weeks, from CoinGecko daily closes (00:00 UTC snapshots, so 'D' is the price at the start of expiry day). Sept 26 2025: -3.6% on the day before expiry, +0.4% on expiry day, +10.3% by D+7, +11.4% by D+14. Dec 26 2025: -0.5%, +0.1%, +1.8%, +4.4%. Mar 27 2026: -3.6%, -3.5%, -2.7%, +3.8%. Jun 26 2026: -2.1%, +0.5%, +3.0%, +5.8%. Four for four, the day before quarterly expiry closed lower (average -2.5%); three of four expiry days were flat within 0.5%; and four for four Bitcoin was higher two weeks later (average +6.4%). The pattern that repeats is pre-expiry pressure toward max pain followed by a post-expiry drift higher, which is the opposite of the 'expiry then crash' story. Four observations is a small sample and Sept 2025's +11% came off a different regime (BTC was $109K then; the 365-day range is $58,566 on 1 Jul 2026 to $124,740 on 7 Oct 2025). What the article gets right: the sequencing. Deribit's 07:30-08:00 UTC TWAP settlement, then durable goods, then UMich, then CME, is an accurate clock, and the equity-side precedent (IBIT's roughly $5B 18 Sept expiry, its largest, with max pain at $40 per Crypto Briefing) did pass without incident. What it gets wrong is proportion: it stacks three 'tests' of very different weight into one dramatic day while the actual rate-sensitive catalyst (PCE) and quarter-end both fall five days later. The peer-reviewed reversal effect it cites (Finance Research Letters 2026, Deribit expiries 2021-2023) is an intraday two-hour phenomenon strongest under negative gamma, and the article's own model places the market in positive gamma at $86K. Correction to the parent card: the 'two more tests' are not 'post-expiry repositioning and resistance near key technical levels'; per the cited article they are the 12:30/14:00 UTC US data releases and the 15:00 UTC CME futures settlement. Spot at the time of writing: CoinGecko $85,395 (23 Sept 2026 daily), Deribit index $85,431.

Strategy's Three September 8-Ks: 950 BTC Bought for $75.7M vs $489.6M of STRC Bought Back, Zero ATM Sales, USD Cash Down $560M - and the '1.17x mNAV Premium' Is an Enterprise-Value Figure; on Market Cap Alone MSTR Trades at 0.88x Its Coins

Strategy filed three Monday 8-Ks in September (Sep 8, Sep 14, Sep 21). Across the three weeks it bought 950 BTC for $75.7 million (average $79,670) and repurchased 5,002,590 shares of its STRC preferred for $489.6 million: $176.3M (1,810,885 sh, Aug 31-Sep 7), $139.3M (1,420,467 sh, Sep 8-13) and $174.0M (1,771,238 sh, Sep 14-20). It sold zero shares under any at-the-market program in all three weeks. Holdings are 846,000 BTC at an aggregate cost of $63.80 billion, average $75,416, so at roughly $86,300 the stack is worth about $73.0 billion, a paper gain near $9.2 billion. What changed versus August: the Aug 31 8-K (Aug 24-30) showed the old machine briefly running: 4,531,421 MSTR shares sold via ATM for $602.8M net, 4,603 BTC bought for $369.7M at $80,318, plus $151.8M of STRC buybacks. In September the ATM went silent and every dollar came out of the balance sheet instead. USD Cash fell from $1.61B (Aug 30) to $1.44B (Sep 7) to $1.30B (Sep 13) to $1.05B (Sep 20), a $560M drain in three weeks, while the USD Reserve slipped from $5.10B to $5.04B. On Sep 8 the board doubled the digital credit securities repurchase authorization from $1.0B to $2.0B; $875.1M remained after Sep 20, meaning $1.12B has been used since the program began. The $1.0B MSTR common buyback authorization is untouched. The headlines got two things wrong. First, 'Strategy is buying bitcoin again' (Bloomberg, Fortune, Sep 21) describes one week in which $75.7M went into BTC and $174.0M went into STRC, funded from cash. For the month, STRC buybacks outran bitcoin purchases 6.5 to 1. This is still a preferred-defence programme with a bitcoin garnish, not a return to accumulation. The buybacks are working on their own terms: STRC closed at $97.07 on Sep 17 against $100 par, up from under $75 in late June, with the dividend held at 12% annualized. Second, the widely quoted 'mNAV at 1.17x, highest since June 22' (mnav.com, CoinGape) is an enterprise-value style multiple that counts the roughly $21.0B senior stack of converts and STRK/STRF/STRD/STRC preferreds ahead of the coins. BitcoinTreasuries shows the same company at 1.13x on an EV basis but 0.88x basic and 0.89x diluted when you simply divide the $64.5B market cap by $73.1B of bitcoin. The Block put enterprise mNAV at 1.1 after the Sep 14 filing. The two framings answer different questions, but the equity-only one explains September's behaviour: when common trades below the value of the coins per share, issuing stock to buy bitcoin is dilutive per BTC, so the ATM stays shut and cash does the work. MSTR closing at $167-168 on Sep 21, its best since May 15, narrows the discount but has not closed it. What to watch: the Sep 28 8-K for whether the ATM reopens now that MSTR has rallied 16% in a week; whether STRC prints at or above $100, which would end the case for further buybacks; USD Cash, which at $1.05B funds only about six more weeks at September's combined $190M weekly pace before Strategy must either issue or stop; and the basic mNAV crossing back above 1.0x, the true signal that equity-funded accumulation can resume.