MASTHEAD IS LIVE MARKET DATA · FIGURES IN THIS REPORT ARE A DATED SNAPSHOT AS OF SEPTEMBER 8 2026
Fourth installment in our seasonal series. A season that ended almost exactly where it began, having gone a very long way in between, and what the round trip revealed about who is actually buying.
REPORT DATE SEPTEMBER 8 2026 · PRIOR INSTALLMENTS FALL 2025 · WINTER 2026 · SPRING 2026
Bitcoin ended the summer within a few percent of where our Spring report found it. Almost nothing else about the season was calm.
Between our Spring report and this one, Bitcoin fell to $57,718 on July 1, the low print of the season, then recovered to an intraday high of $81,480 on August 28. It sits at $77,932 as we publish, within about a point of where our May report found it. A holder who checked the price in May and again in September would conclude that nothing happened. That reading would be wrong.
The flows tell the real story. June produced the largest monthly outflow in the history of the US spot ETFs at $4.51 billion. July was nearly flat at $172 million of net inflow. August produced the strongest month of 2026 at $3.52 billion. Three consecutive months, three different regimes, in a product category that is supposed to represent patient allocation. The summer nets to roughly $818 million of outflow.
Underneath that, corporate treasury stress broadened. June tracker additions were 7,314 BTC, or 0.54× estimated new issuance, while several smaller balance-sheet holders reduced holdings or entered distress. We could not verify a clean June through August open-market purchase ratio excluding mined coins and tracker reclassifications, so we do not claim the sector was a net seller for the season. We claim only that the two-times absorption our Spring base case rested on is unsubstantiated.
We are cutting the 2026 base case again, to $85K–$110K. Bitcoin spent the summer inside our Spring bear range of $60K–$80K, closing there on 86 of 92 days and briefly breaking beneath it. When summer price action tracks the bear band rather than the base band, the base-case assumptions need rebuilding rather than defending.
SOURCE · Coinbase BTC-USD daily candles, granularity 86400s, June 1 to September 1 2026. Intraday extremes from that single series; publication price observed September 8 at 14:00 UTC. Exchange series differ by venue and timestamp convention.
Comparing the Spring 2026 forecasts against what the summer actually delivered.
| Spring 2026 call | Published May 19 | Summer actual | Verdict |
|---|---|---|---|
| 2026 EOY price (base) | $110K–$140K | $57,718 low · $81,480 high | Not in range at any point |
| 2026 EOY price (bear) | $60K–$80K | 86 of 92 closes inside the range | Tracked summer; year-end open |
| Hashrate clears 1.2 ZH/s (bull trigger) | 1.2 ZH/s | ~914 EH/s avg; 1.0 ZH/s spike | Did not occur |
| ETF outflows continue into Q3 (bear condition) | continued outflows | June −$4.51B, then Aug +$3.52B | Half right, then reversed |
| Treasuries absorb at ~2× mining rate (base condition) | ~2× mining output | June additions: 0.54× issuance; full-summer ratio unavailable | Base premise not substantiated |
| BITCOIN Act passes committee (base condition) | committee, not floor | Still in committee | Neither advanced nor died |
| Corporate demand is opportunistic, not mechanical | stated assumption | Several holders reduced positions or entered distress | Consistent with observed behavior |
This is the worst scorecard we have published, and the specific way it is bad matters more than the fact that it is. The band we assigned 25% probability is the one that described the summer, and the band we assigned 55% was not in contention on a single day of it. Year-end is still four months out, so none of these forecasts has expired; what has already failed is the reasoning that produced the weighting.
Our error was not the direction. It was that we treated the corporate bid as a floor without a measurement that could bear the weight. June's tracker additions came in at 0.54× issuance, and we could not verify a clean full-summer open-market ratio in either direction. That is not proof that the sector was a net seller. It is the absence of the proof our base case assumed.
We flagged in Spring that our bear cases had been more accurate than our base cases across both prior reports, and we widened the bear weight accordingly. That was the right instinct and it was still not enough, because we widened the weight without questioning the assumption underneath the base case. We are widening it again below, and this time we are removing the corporate bid from the base case as a load-bearing input.
Through a 21-month price low, the thing that was supposed to break did not.
Seven-day simple moving average, reading 922 EH/s on our own node in early September. It never reached the 1.2 ZH/s we set as a bull trigger in Spring.
Down 1.31% at block 963,648, after an increase to 127.48T on August 8. Ordinary oscillation, not a distress signal.
The hashrate level our Spring bull case required. It was not reached at any point in the season, and the gap to it widened rather than closed.
Bitcoin spent the summer in the low end of our bear range, touching $57,718, and the network did not seize. Difficulty adjusted in both directions on schedule. Hashrate declined modestly rather than collapsing. Several listed miners announced material AI and high-performance computing projects over the same period, though we cannot quantify what share of the hashrate decline those conversions account for.
We said in Spring that healthy network metrics remove a category of downside risk without being a price catalyst. The summer tested exactly that claim under a much harsher price than we expected, and the claim survived. There was no self-reinforcing miner-sell spiral, even at $57,718.
SOURCES · Hashrate is a 7-day simple moving average; end-August ~915 EH/s, with a ~922 EH/s reading on Carroll Park Capital's own full node in early September (private observation, dated calculation available on request). Summer high ~998.8 EH/s, rounded to 1.0 ZH/s in the scorecard. Difficulty 125.81T at block 963,648, August 23 2026, −1.31%, read from the same node and cross-checked against Luxor Hashrate Index.
Hashprice pressured marginal fleets while several listed miners accelerated announced AI/HPC diversification.
Hashprice sat near $31.90 per PH/s per day on August 11, having traded around $29 in July. Hashprice spent much of June and July below $35 per PH/s per day, pressuring less-efficient and higher-cost fleets. We are deliberately not drawing a universal shutdown line from that number: actual profitability depends on machine efficiency, power contracts, facility overhead, curtailment revenue and hosting arrangements, and it varies widely between operators.
The sector did produce one genuine insolvency in the window. Poolin Technology and two US affiliates filed for Chapter 11 on July 22 in the District of New Jersey. The causes include unresolved wallet liabilities dating to 2022 rather than summer hashprice alone, so we would not read it as a clean read on current mining economics.
Several listed miners announced or expanded AI and high-performance computing projects; the converted capacity, realized economics, and effect on Bitcoin hashrate cannot yet be quantified. We could not verify a current weighted-average public-miner cash-cost estimate to publication standard, and we would note that cash cost, electricity-only breakeven and all-in cost are not interchangeable measures, so figures quoted elsewhere are frequently not comparable.
What we are watching: MARA sold 23,093 BTC in H1, of which 20,880 BTC occurred in Q1; its Q2 filing reports 2,213 BTC sold during April to June. Without month-level disclosure, June to August MARA sales cannot be quantified.
SOURCES · Hashprice observations via Luxor Hashrate Index, August 11 2026. No current weighted-average public-miner cash-cost estimate met our sourcing standard, so none is quoted. Poolin Technology Pte. Ltd. and two US affiliates, Chapter 11, District of New Jersey, case 3:26-bk-18325, filed July 22 2026 — claims agent docket. MARA figures from its Q2 2026 Form 10-Q (period ended June 30 2026).
Corporate treasury stress broadened this summer. June tracker additions were 7,314 BTC, 0.54× estimated new issuance, and a clean June to August open-market purchase ratio, excluding mined coins and tracker reclassifications, could not be verified.
| Company | Summer action | Position | Note |
|---|---|---|---|
| Strategy (MSTR) | Net add | 845,050 BTC | Reportedly a net balance change of roughly +1,344 BTC from June 1 to August 30 per filing summaries, not a fully reconciled purchase figure. Reportedly paused ten weeks, sold a token 32 BTC in June, then resumed via share issuance, and briefly traded below the value of its own holdings. |
| Metaplanet | Added | 43,000 BTC | Reportedly bought 2,823 BTC on July 2, and is reported to be targeting 100,000 by year-end. |
| Twenty One Capital | Turmoil | ~43,500 BTC | The three-way proposal with Strike and Elektron was reportedly abandoned, though two-way talks with Elektron reportedly continued. Its chief executive resigned July 20. The reported $1.27B first-half loss was predominantly fair-value accounting rather than operating cash burn. |
| MARA Holdings | Sold | 35,577 BTC | Sold 23,093 BTC across H1, of which 20,880 BTC fell in Q1. The Q2 filing reports 2,213 BTC sold across April to June, which is not the summer window. Month-level disclosure does not exist, so June to August sales cannot be quantified. |
| Smaller treasury companies | Distress | — | Satsuma reportedly approved a full liquidation and delisting. Nakamoto reportedly sold roughly 600 BTC in June and reduced debt. Sequans and Smarter Web reportedly reduced holdings. All four rest on secondary reporting we could not reconcile to primary filings, and we could not establish a common debt-repayment purpose. |
The summer showed that treasury-company demand depends on financing conditions. Several smaller vehicles reduced holdings or entered distress as valuations and financing conditions tightened, but public evidence does not establish one common motive or funding constraint across them.
Strategy is the instructive contrast. It paused for ten weeks and sold a token 32 coins in June, then resumed accumulating by issuing shares rather than by liquidating its position. We have not established the financing options available to the smaller vehicles, so we draw no comparison about their access to capital. The observation we will make is narrower: this demand source tracks financing conditions rather than stated conviction.
What we are deliberately not claiming: that the sector was a net seller across the summer. Aggregate holdings remain large, at roughly 1,271,444 BTC across 198 public companies as of September 8, but that is a stock figure and not purchase flow. The constant-universe monthly flow files needed to compute a clean July and August ratio were not available, and a stock comparison across those dates mixes changing company coverage, reclassifications and retained mined coins. We can substantiate that the two-times absorption premise is unsupported. We cannot substantiate its opposite.
SOURCES · June net additions and the 0.54× issuance ratio from the BitcoinTreasuries June 2026 report (PDF); public-company stock snapshot of 1,271,444 BTC across 198 companies observed at bitcointreasuries.net on September 8 2026 — a stock figure, not purchase flow, on a changing company universe. MARA detail from its Q2 2026 Form 10-Q. Constant-universe monthly flow files for July and August were not available. EVIDENCE CLASS BY ROW · Strategy balance and net change from company filing summaries; MARA from the linked Q2 Form 10-Q; Twenty One's merger status, leadership change and fair-value loss characterization from company disclosure. Metaplanet, Satsuma, Nakamoto, Sequans and Smarter Web rest on secondary reporting that we could not reconcile to primary filings, and are stated as reported rather than as verified fact.
Record out, then nearly nothing, then the best month of the year. The dispersion is the finding.
Largest monthly outflow since the funds launched in January 2024.
Smallest positive monthly net inflow to date. Net creation and redemption activity was nearly balanced.
Strongest month of 2026, alongside a roughly 25% price gain.
Two separate measures, which are often conflated. Year to date, the category sits at roughly $1.77 billion of net outflow, improved from about $5.29 billion before August. Cumulative net flows since the January 2024 inception stood at $54.847 billion on August 31. The three summer months together net to roughly $818 million of outflow. So the category remains deeply positive since launch and modestly negative for the calendar year, and those two facts are not in tension.
In Spring we asked whether May's redemptions were tactical positioning or a thesis change, and said we were watching whether they spread beyond the largest fund. They did spread, in June, and the honest reading of that month is that it was worse than tactical. What followed in August is the part we would not have predicted: the same channel came back at its strongest pace of 2026 within eight weeks, which argues the June exit was about rates and risk appetite rather than about Bitcoin.
SOURCE · SoSoValue US spot bitcoin ETF dashboard, inception-cumulative observed August 31 2026; monthly net-flow figures are provider-reported, revision-prone, and reflect that provider's fund universe.
A quiet quarter for enactments and a busy one for rulemaking.
No new statute expanding sovereign Bitcoin holdings became law this summer, and no US state enacted a reserve law in the window. On sovereign buying we found no independently verified new open-market purchase during June through August. El Salvador's reported additions remain disputed against its IMF representations, and wallet transfers between addresses are not purchases.
Our base case continues to assume no Reserve legislation passes in 2026. We were wrong in Spring to assume even committee passage, so we are now treating the entire legislative track as optionality rather than as a projected input.
SOURCES · BITCOIN Act S.954 and H.R.2032, status observed September 8 2026. Regulation Crypto Assets, Federal Register, published August 21 2026. H.R.8957 status observed September 8 2026.
Probability weighting: Bear 35% / Base 50% / Bull 15%. Bear weight raised again. With under four months left in the year, the ranges compress toward spot. Long-term thesis unchanged.
Macro risk-off persists, the Federal Reserve holds or hikes into year-end, geopolitical stress keeps energy elevated, and the treasury-company unwind continues to add supply.Spring '26 was $60K–$80K
August's flow recovery holds without accelerating, the Fed cuts once, no Reserve legislation passes, and the forced selling in the treasury sector finishes working through.Spring '26 was $110K–$140K
ETF flows sustain the August pace through the fourth quarter, the market-structure bill clears the Senate, and a rate-cut cycle resumes in earnest.Spring '26 was $175K–$220K
Weighted midpoint: ~$93,000 year-end 2026. Prior reports: Fall 2025 was $188K, Winter 2026 implied $155K, Spring 2026 was $118K. This is the third consecutive cut. We publish the direction of our revisions rather than quietly re-basing, because the pattern of the revisions is itself information about the model.
METHODOLOGY NOTE · Our Spring 2026 report published a weighted midpoint of $118,000. Weighting the midpoints of its own published ranges at 25 / 55 / 20 produces $125,750, so that figure was either derived differently or stated in error. We are recording the discrepancy rather than restating it after the fact. The Summer figure above is the probability-weighted average of the three range midpoints, which computes to $92,750.
The thesis is intact on the components that are about Bitcoin. The component that was about balance sheets is the one that failed.
The structural case we set out in Fall 2025 rested on four legs: supply-side scarcity, institutional absorption, sovereign accumulation, and illiquid-float compression. After this summer, three of those are unchanged and one needs restating.
Scarcity is a property of the protocol and did not move. Institutional absorption through the ETF channel is intact, and August demonstrated it can return at scale within weeks of a record exit. Sovereign accumulation has not advanced at all this year, which was already our base assumption.
Float compression was partly attributed to leveraged corporate vehicles. This summer supplied clear examples of stress and selling, while June tracker additions fell to 0.54× issuance; a clean June to August open-market ratio could not be verified. The corporate bid is therefore removed from the model as a dependable floor, not because aggregate seasonal net selling was proved.
Price is currently $77,932.
We would rather write this section than the one where the base case quietly resets and nobody mentions the last three. Our forecasts retained a bullish directional bias, while realized timing and magnitude repeatedly undershot them. That is now a four-report pattern rather than a coincidence, and we treat it as a structural bias in how we model adoption speed rather than as bad luck.
Long-term projections from Fall 2025 are unchanged: weighted $476K by 2030 and $1.15M by 2035. Nothing this summer touched the supply schedule, the settlement network, or the direction of institutional access. What it touched was the assumption that a particular class of buyer would be there at every price, and that assumption is now out of the model.
Next report: Winter 2027.
This report has been prepared by Carroll Park Capital to provide investors with a directional outlook on Bitcoin's potential price trajectory and on-chain state. This analysis is a hypothetical exercise based on publicly available market data, historical patterns, and probability-weighted scenario modeling. This is not financial advice. All projections are speculative. Consult a qualified financial advisor before making any investment decisions.