The Seller Is Not the Owner
The Seller Is Not the Owner
On Sunday, bitcoin traded at $61,946 on Coinbase. On Monday it was back near $64,500, and it held that level while the Nasdaq fell 2.5 percent and the S&P lost nearly 2 percent. By Tuesday morning it was $64,310. That is a strange sequence if you believe bitcoin is simply a high-beta tech proxy: the proxy fell less than the thing it supposedly proxies, on a day when the proxy’s biggest single name lost 4 percent.
The interesting part of this drawdown is not the low. It is who did the selling, and why they had no choice. Three separate credit mechanisms fired inside ten days, and all three of them act on claims rather than on coins: a rate hike that futures now price for December, a $61 billion equity offering that pulled the marginal dollar out of everything else, and a leverage cascade that liquidated $1.78 billion of long positions between November 9 and November 15. None of those three events is an opinion about bitcoin’s monetary properties. All three of them set its price.
This is the paper era’s first real test, and it is worth being precise about what is being tested. Not whether bitcoin works. Whether a market whose marginal holder owns a ticker rather than a key can survive a funding shock without handing the coins to someone who does.
The first seller had a margin call
Start with the mechanics of the weekend. Bitcoin fell 6 percent on Saturday, a move that Nik Bhatia described in The Bitcoin Layer as leaving “so much technical damage it’s hard to see straight,” and which took out both $75,600 and $66,000 within what looked like a single session. Liquidations of leveraged longs came to roughly $1.78 billion over the November 9-15 week, according to Liqwall data cited by Marty Bent.
A liquidation is not a decision. It is an execution. Somebody who borrowed to be long gets closed out at the worst available price, and the sale is transmitted into the order book regardless of what the seller thinks about the asset. When $1.78 billion of that arrives inside a week, the resulting price is a record of forced selling, not of consensus.
The technical marker everyone is now quoting is the 200-week average. Wintermute told CoinDesk that losing it opens the $50,000s. I recomputed it rather than taking the claim: using daily closes from August 2017 through this morning, the 1,400-day mean sits at $68,007, and the last close was $64,385, about 5 percent below it. Bitcoin spent most of 2022 and early 2023 under that line; the last close below it came on 15 October 2023, and then nothing for 964 days. In 2026 the line has been crossed on 28 days: eleven in June, five in July, eight in August, and four in a row this weekend and today (own calculation, 17 November 2026, 09:31 CET).

Two things follow. The first is that this is a genuinely rare event, not a routine wobble below a moving line. The second is that it has happened before this year without ending the market, in June and again in mid-August, which is a caution against treating one line as destiny in either direction. What the line measures is momentum and positioning, and positioning was extreme: Bhatia’s own valuation score, a composite he publishes, sits at 19 out of 100, cheap but not as cheap as the summer low near $58,000 and nowhere near the November 2022 reading after FTX.
There is a contested subplot here that deserves to be flagged rather than repeated. Bhatia argues that Saturday’s dump had something to do with Binance absorbing the loss of roughly $6 billion, which he says the company admitted was not fully covered by its internal insurance fund. Binance’s chief executive has denied that the exchange sold crypto. I cannot verify either account, and a market maker does not publish its order book. Treat the Binance story as an unproven explanation that happens to fit the shape of the move, and treat the liquidation total as the verified part. That distinction is the whole difference between analysis and narrative.
The second seller had an allocation
The largest equity offering of the autumn prices on Wednesday. Anthropic prices its Nasdaq IPO on November 18 and begins trading the next day. The range is $70 to $78 a share, on 825 million Class A shares, for a valuation near $1.4 trillion at the midpoint. At the $74 midpoint that is a gross raise of about $61 billion (own calculation: 825 million shares at $74). Seven chip and platform companies, among them Google, Nvidia, Microsoft, AMD, Broadcom, Micron and TSMC, are committing a further $26 billion in a concurrent private placement.
Brad Gerstner of Altimeter told CNBC on Monday that he is placing a $5 billion order, five times his previous largest IPO commitment, and framed it as a rotation: “We are moving from an era over the past three and a half years that was focused on AI infrastructure to an era of intelligence and applications.” He confirmed that Altimeter cut semiconductor and infrastructure exposure to fund it.
Here is the arithmetic that the rotation narrative skips. Anthropic’s own amended S-1 discloses roughly $895.4 billion in total compute and infrastructure commitments. Against that, the offering raises $61 billion at the midpoint: the commitments are 14.7 times the raise (own calculation). For the nine months to September 30 the company spent $18.8 billion on compute and infrastructure, up 329 percent year over year, on revenue of $33.2 billion. That is 56.6 percent of revenue going to compute alone (own calculation), and it means the annualised revenue run-rate implied by those nine months, about $44 billion, is being valued at roughly 32 times sales.

None of that makes the IPO a fraud, and the steel-man version is easy to state: a company growing revenue at triple-digit rates, with government contracts and a locked-in customer base, can rationally trade at 32 times sales if the growth continues, and the $895 billion of commitments is a multi-year floor on capacity rather than a bill due in 2027. Sophisticated buyers with better information than I have are lining up. But it does mean the offering is not a rotation out of infrastructure. It is the second-largest call on public equity capital ever recorded, after SpaceX’s $75 billion in June, arriving in the same week that a rate hike is being priced.
The template was set in June. SpaceX went public on June 12 having raised $75 billion, then the largest IPO ever. Bitcoin fell 19 percent in the 50 days before the listing. It bottomed 18 days after, 25 percent below the pre-IPO level, and by day 100 it was 3.7 percent above where it started, which is 28 percent above the IPO-day close and 39 percent above the low, in the accounting that Strive’s Matt Cole published. Bitcoin is down 18 percent into Anthropic’s listing, over the same 50-day stretch. Two data points are not a pattern, and I would not trade on the overlay; but the mechanism it describes is real and it is not about bitcoin. When a $61 billion primary offering is priced in a week, portfolio managers do not fund it from cash they were holding for no reason. They sell what is liquid, and they sell it before the pricing, not after.
That is also why the AI story and the bitcoin story are the same story this month, told from two ends. Nvidia’s chief executive resigned last week; the board is reported to be investigating whether his conduct compromised company secrets rather than preparing his return, and Colette Kress is running the company as interim chief executive. Nvidia fell 4 percent on Monday, US chipmakers fell more than 5 percent, and the Nasdaq fell 2.5 percent. Interconnects’ Nathan Lambert made the connection that matters for anyone running open infrastructure: Jensen Huang was the most consequential financial patron of open-weight models, and that funding “will very likely dry up” while Nvidia stabilises. The capital consolidating around closed labs is the same capital that was underwriting the open alternative.
The third seller had a dollar short
The macro backdrop did the rest. Fed Chair Kevin Warsh spoke at a San Francisco Fed conference on Monday and, according to Bitcoin Magazine’s account of the session, offered no guidance on the near-term path of rates. Fed funds futures imply a quarter-point increase in December. I did not verify the CME pricing myself, and I flag that as reported rather than measured. The dollar index touched its highest level since April 2025 and stood at 103.54 this morning. The ten-year Treasury yield was 5.113 percent (own reading, 17 November, 09:31 CET). Gold futures were $3,869 an ounce.
A hike is the one macro variable that hits bitcoin through a channel no narrative can soften: it raises the cost of carrying every position in the system, including the ones that are long bitcoin with borrowed money. It also raises the discount rate applied to every long-duration cash flow in the AI complex. Both effects arrive at once because both are the same effect — the price of credit.
And Warsh, in his own prepared text, named the risk that the market is pricing. His speech laid out three stylised scenarios for what he calls super intelligence. In the first, productivity gains are shared and growth rises above trend with the economy near full employment. In the second, growth rises but automation displaces workers faster than it complements them. The third is the one to read twice:
“Or SI could disappoint. After significant capital investment and stock market gains, expected returns could turn down and equity valuations could fall, leading to an economic slowdown and a more notable rise in unemployment. In time, the economy would recover, but without the productivity gains.”
That is a central banker describing, in neutral language, the top of an investment boom.
The bill the Fed already sees
Two hours later, Governor Michael Barr gave the same scenario a balance sheet. Speaking in Milan, he summarised the Federal Reserve’s Financial Stability Report, published the previous Friday, which is devoted this edition to the AI investment boom and to the leverage inside the debt structures financing it. The numbers are his:
On-balance-sheet debt of the seven largest AI firms “has been growing significantly in 2026 and is expected to exceed $800 billion before year’s end.” Their off-balance-sheet obligations (long-term contracts, lease commitments, residual value guarantees) have grown even faster, and the report estimates “the potential maximum nominal value” of those at more than $3 trillion. Hyperscalers and large GPU producers hold over $100 billion of cumulative equity stakes in model labs and cloud startups, a structure Barr calls circular financing: the supplier invests in the customer, the customer spends the money renting compute from the supplier. And the equity premium, the extra return investors demand for owning stocks instead of risk-free bonds, is at a low level relative to history, though not as low as in the dot-com bubble.
Barr’s own reading is measured, and it is not a crash call: with debt still limited relative to the size of the buildout, he is “less concerned about direct losses” and more concerned about hedge-fund leverage and fire sales if sentiment turns. He compares the AI buildout to the railroads of the 1870s, the highways of the 1950s and the fiber of the 1990s, booms that left useful capacity behind, some of which also blew up their financiers. He even supplies the phrase for what happens to the losers: “dark fiber.”
Strip the institutional politeness and this is the Austrian story with an American accent. A cluster of investment is financed increasingly by credit rather than by retained earnings; the collateral is capital goods whose value depends on a demand curve nobody has observed yet; the guarantees are off-balance-sheet, which means the risk is real but the location of it is unknown; and the whole structure is valued off a discount rate that is now being revised upward. That is not a prediction of collapse. It is a description of why the marginal cost of capital rises exactly when the marginal project needs it most, and why a boom of this shape ends with a repricing rather than a decision.
Bitcoin is outside that structure in every mechanical sense that matters: no debt, no leases, no residual-value guarantees, no supplier holding equity in it, no board, no roadmap. It has no obligations at all. And that is precisely why it gets sold first.
You sell what you can sell
In a liquidity event, the portfolio manager does not sell the worst asset. He sells the one that can be sold today, at three in the morning, in size, without a lock-up, a gate, a board approval, or a counterparty asking questions. In an equity book stuffed with private positions, pre-IPO stakes and data-center obligations, bitcoin is the only instrument that clears instantly and settles in minutes. So it is the first thing converted to dollars, and the conversion says nothing about the asset’s fundamentals. It says the portfolio needed dollars.
This is the inversion that the “risk asset” framing misses. Bitcoin behaves like the most liquid collateral in the system, not like the riskiest holding in it. On a weekend when equities are closed and the IPO is days away, the only place to raise cash is the market that never closes. The seller is not the owner, and the owner is not the seller.
The uncomfortable counterweight is that in this cycle the marginal holder really is weaker than before. Cory Klippsten of Swan, whose firm sells the alternative, so read him as an interested party, makes the argument that we are in a “paper Bitcoin era”: the spot ETFs launched in January 2024, most new buyers now start with price exposure in a brokerage account, and by his estimate two-thirds to three-quarters of Swan’s new customers over the past year already held an ETF before they came for real coins. His conclusion is that the journey from exposure to conviction has doubled, from two or three years to four to six.

He is describing a real structural change, and he is optimistic about it. I am less sure the optimism is warranted, for a reason his own numbers imply. If the marginal buyer arrives holding a claim that can be sold with one click, then the marginal seller in the next drawdown is also holding a claim that can be sold with one click, and he has not yet built the conviction that would make him hold through a 30 percent decline. Each leg down therefore meets a slightly less committed holder than the last, while the coins behind those claims sit in a handful of custodians. That is not a bearish forecast. It is a description of who sets the price now.
What ownership looked like this week
While the ticker fell, the filings told a different story, and the contrast is the point.
Wells Fargo disclosed 49.58 million shares of BlackRock’s IBIT, worth about $2.35 billion as of September 30, more than double its second-quarter position of 24.4 million shares. Roughly 85 percent of it sits in the bank’s brokerage arms as client holdings. The BTC-equivalent exposure, about 28,000 coins, is within 7 percent of the 30,221 bitcoin that Tesla and SpaceX hold directly on their own balance sheets. Goldman Sachs reported about $1.28 billion of spot bitcoin ETF exposure at quarter-end, Millennium about $1.21 billion. That is the distribution machine working exactly as designed, and it is also the point at which the largest single IBIT position reported by any US bank belongs to clients who own a share of a trust, not a coin.
Strategy bought 485 bitcoin last week for $33.4 million at an average of $68,887, lifting its stack to 850,849 coins bought for $64.20 billion, an average of $75,452 including fees. At this morning’s price that stack is about $9.4 billion underwater (own calculation). But the same filing shows the company spent $228.1 million buying back its own variable-rate preferred, STRC, 6.8 times what it spent on bitcoin, funded by selling 2.09 million common shares for $244.7 million in net proceeds. Over three weeks the split is $624.6 million to STRC buybacks against $180.8 million to bitcoin, and $332.2 million remains in that buyback programme. Strategy also faces an estimated $2.8 billion of index-driven selling when MSCI removes it at the close on November 30, per JPMorgan’s estimate cited by Bitcoin Magazine.
Read that as a capital-allocation statement rather than a betrayal. The preferred stock carries a 12 percent dividend. A company whose equity trades below the value of its coins cannot issue equity accretively, so defending the preferred, whose coupon is a hard cash obligation, takes priority over buying more of the asset with no coupon at all. That is what a treasury company looks like when the reflexive loop inverts. It is not the end of the model; it is the model becoming a normal, leveraged balance sheet, subject to the same arithmetic as everyone else.
And then there is the smallest number of the week, which may be the most instructive for anyone who thinks self-custody ends counterparty risk. An attacker drained about $6.4 million in USDC and a yield-bearing token from a Safe multisig, a wallet that normally requires several owners to sign. No owner signed. The funds left through Zodiac’s Roles Modifier, an add-on that lets owners grant limited permissions to other addresses. Zodiac says its contracts had no flaw and that a role member’s key, probably compromised, sent a transaction that a custom policy permitted but almost certainly never intended. Safe said the module had been installed outside its own app.
Three gradations of the same thing: a brokerage line that is a claim on a trust, a corporate treasury whose coins are pledged against preferred dividends, and an on-chain wallet whose owners delegated signing power to a key they did not control. None of them is fraud. Each of them is a decision to hold the exposure without holding the thing, and each of them worked fine until the week it didn’t.
The register
Meanwhile the part of the system that has no counterparty did what it does. Blocks kept arriving, difficulty kept adjusting, and the value settled on the chain kept rising through the selloff: James Check’s transfer-volume measure put the 30-day average at $9.57 billion per day on November 15, up from $8.34 billion on October 11, with a peak of $9.98 billion on November 11. Roughly ten billion dollars a day moved across a network that had just fallen 26 percent from its September high of $87,396. Measured from the all-time high of $126,200 in October 2025, the drawdown was 49 percent (own calculation from daily closes).
That divergence, settlement volume up and price down, is the whole argument in one line. The register is not repricing. The claims on the register are.
What I could not verify
Four things in this piece rest on secondhand reporting, and the reader should know which. The December hike expectation comes from Bitcoin Magazine’s reading of fed funds futures; I did not pull the CME curve myself, so treat the direction as reported and the magnitude as unconfirmed. The Binance insurance-fund story is a market participant’s judgment about an exchange’s internal book, and the exchange denies the selling; I could not substantiate either side. The Anthropic figures come from filings quoted at length by TFTC, not from a prospectus I downloaded. And the Wells Fargo position is a Form 13F snapshot of September 30, filed this week, which by construction tells us nothing about what those clients did during the selloff it is being used to explain. TFTC notes that IBIT saw more than $1 billion of outflows after that quarter ended, which points the other way.
What decides the next leg
The honest part: I do not know where the low is, and neither does anyone quoting a chart. Wintermute’s $50,000s are a plausible target if the ETF bid keeps retreating. Bitcoin Magazine’s MVRV line at 1.1, around $59,000, is the level at which, historically, the marginal seller runs out of coins to lose. Bhatia’s valuation score of 19 says cheap, not capitulation. The 2018 template says bottoms come in December; the 2022 template says November. The SpaceX overlay says the bottom may already be behind us, 18 days after an offering that has not priced yet.
What I can say with confidence is what to watch, and it is not the price. Watch whether the coins that left the ETFs come back on-chain when the claims get expensive, because that is the tell for whether the paper era produces owners or only renters. Watch the December meeting and the ten-year at 5.1 percent, because a hike is a tax on every levered position in the system, bitcoin’s included. Watch the Anthropic pricing on Wednesday, not because an AI lab determines bitcoin’s value, but because $61 billion raised in one week is $61 billion that has to come from somewhere, and the somewhere is every liquid asset in every portfolio that wants an allocation.
The rest is noise dressed as signal. A price set by margin calls, rebalancing and a rising discount rate is a price worth recording, not a verdict worth internalising. The coins did not move because anyone changed their mind about them. They moved because the people holding claims needed dollars, and bitcoin was the only thing in the book that would turn into dollars on a Sunday.
Sources
- Federal Reserve — Governor Michael S. Barr, “Some Potential Financial Stability Implications of Artificial Intelligence”, 16 November 2026
- Federal Reserve — Chairman Kevin Warsh, “Opening Remarks”, 16 November 2026
- TFTC / Marty Bent — “Did You Think It Was Going to Be Easy?”, 16 November 2026
- TFTC / Marty Bent — “Wells Fargo’s IBIT Position Doubles to $2.35B”, 17 November 2026
- TFTC / Marty Bent — “Altimeter Bets $5B on Anthropic IPO as AI Capex Keeps Climbing”, 16 November 2026
- CoinDesk — “Bitcoin could slide into the $50,000s as Fed rate hike bets build, Wintermute says”, 17 November 2026
- Bitcoin Magazine — “Bitcoin Price Holds Near $65,000 as Nvidia Turmoil Drags Nasdaq Down 2.5%”, 16 November 2026
- Bitcoin Magazine — “Why Bitcoin’s June Low May Not Be the Bottom — $58K the Key Test”, 16 November 2026
- The Bitcoin Layer / Nik Bhatia — “Winter Remains”, 16 November 2026
- Daily Bitcoiner / Cory Klippsten — “The Paper Bitcoin Era, and Why I’m Still Bullish”, 16 November 2026
- Unchained — “Strategy Buys 485 Bitcoin for $33.4 Million, Lifting Holdings to 850,849 BTC”, 16 November 2026
- Unchained — “Hacker Drains $6.4 Million From a Multisig Through Permissions Its Owners Set”, 16 November 2026
- Interconnects / Nathan Lambert — “Open-source AI without Jensen”, 14 November 2026
- Import AI / Jack Clark — “Import AI 481: AI sub-contractors; a wet lab for AI; and subterfuge during development”, 16 November 2026
- Farside Investors — Bitcoin ETF flows, table read 17 November 2026 (own summation)
- Binance — BTCUSDT daily candles, June 2020 to 17 November 2026 (own calculation of the 200-week mean and drawdowns)
- The Block — “Grayscale says crypto pullback offers compelling entry points as bitcoin falls to $64,000”, 17 November 2026
Write a comment