The dismantling of the firewall
From roughly 2021 to 2024, American regulators did not ban banks from touching crypto. They did something more effective: they made it expensive. The centrepiece was SAB 121, the SEC staff bulletin requiring any entity safeguarding crypto for customers to book both an asset and a matching liability, which turned custody — normally an off-balance-sheet fee business — into something that consumed balance sheet and capital. Around it sat the Basel crypto standard, which assigns certain unbacked crypto exposures a 1,250% risk weight, meaning a bank must hold capital roughly equal to the exposure. That is a prohibition wearing a percentage sign. And around that sat the Federal Reserve's reluctance to grant master accounts, which kept crypto institutions off central-bank payment rails altogether.
The stated rationale was prudential rather than ideological: operational, custody, liquidity, market and legal risks that regulators judged atypically high, and a desire not to let large supervised institutions act as the transmission wire. Industry called it a blockade. Both descriptions are compatible.
The accounting piece went first. SAB 122 repealed SAB 121 effective 30 January 2025, removing the balance-sheet treatment that had made bank custody uneconomic. The Basel weight and the access constraints were not all erased at once, but the biggest single barrier was gone.
What came through the gap is a credit channel. JPMorgan launched institutional lending against bitcoin and ether collateral in March 2026 through its Kinexys unit, with coins held at third-party custodians including Fidelity Digital Assets and Coinbase Custody, at reported haircuts of 30% to 50% — an of roughly 50 to 70 cents on of collateral. We covered the institutional rollout on 20 August: a $1mn bitcoin position supports something like $500,000 to $700,000 of borrowing. The last sourced figure for the whole crypto-collateralised lending market is $73.6bn outstanding as of Q3 2025; we have not seen a credible mid-2026 total. Claims circulate that several other large US banks run similar facilities at 40–65% loan-to-value, but they come without named banks or launch dates, and we are not treating them as established.
The haircuts are conservative, and that matters. The risk is not that JPMorgan loses money on one loan. It is that a fast drawdown produces simultaneous margin calls across a client base holding the same two assets and selling into the same order book.
The stablecoin Treasury nexus
The GENIUS Act, signed on 18 July 2025, requires payment stablecoin issuers to hold 100% reserve backing in liquid assets such as US dollars or short-term Treasuries, with monthly disclosures. The White House fact sheet puts it in exactly those terms. This was sold as consumer protection, and it is. It is also an industrial policy: it legally conscripted a speculative asset class into being a structural buyer of US government debt.
The scale of that is the number worth remembering. Tether's most recent 2026 attestation reports $114.96bn of direct US Treasury bills, which would place it somewhere around the 17th to 18th largest holder of US government debt in the world, ahead of Germany, the UAE and Australia. Circle does not publish a comparable dollar figure; its reserves are described as roughly 80% short-dated Treasury bills held through a reserve fund structure. So a demand shock in stablecoins is now, mechanically, a demand shock in bills.
That transmission is not speculative. The BIS working paper Stablecoins and safe asset prices (Ahmed, 2025) measures the effect of dollar-backed stablecoin flows on short-term Treasury yields using daily data from January 2021 to March 2025. A Federal Reserve FEDS note summarising it reports the estimate that "a 2-standard deviation inflow into stablecoin lowers 3-month Treasury yields". Inflows push bill yields down. Outflows do the opposite, and there is no rule saying they have to be gradual. This is now a live input into the , not an adjacent curiosity.
One piece of good news, carefully sourced: Tether completed its first full independent financial audit, with KPMG issuing an unqualified opinion on statements showing reserves exceeding liabilities by $6.8bn at end-2025. We marked that down on 20 August while noting it reached us through a single outlet and we have not seen the statements ourselves.
The treasury-company flywheel in reverse
The digital-asset treasury company is a perpetual motion machine with one moving part: the premium. Issue equity above the value of the bitcoin behind each share, buy more bitcoin, watch bitcoin-per-share rise, which justifies the premium, which lets you issue again. built roughly $59bn of coin this way and dozens of imitators copied the template.
It only spins in one direction. Below net asset value, issuing stock to buy coin destroys value for existing holders, and the preferred dividends built up during the good years still have to be paid in dollars by a company whose only asset pays nothing. So the machine reverses. In the week to 16 August 2026, Strategy sold $333.7mn of common stock and bought no bitcoin with it; the proceeds went to repurchasing preferred and building cash. Over five weeks it sold about $2.1bn of common, repurchased roughly $347mn of STRC preferred, and sold $213.3mn of bitcoin, ending with a $4.8bn cash reserve. It had bought no bitcoin since mid-June. It had earlier retired $1.5bn of its 0% 2029 convertibles for about $1.38bn cash, cutting converts from $8.21bn to $6.71bn. We laid out the mechanism on 19 August and again on 18 August.
"Strategy's trading is pro-cyclical," Orbit Markets' Caroline Mauron told Bloomberg. "They are able to buy more when it goes up, and they are forced to sell when it goes down." The largest marginal buyer of bitcoin for six years is now a structural seller of it.
The disagreement worth sitting with is in the share price. The stock rose 12.64% on 18 August to its period high, and was up 15.7% over five days by 20 August, even as the company sold coin. Its shares are down about 73% over a year against bitcoin's 45%. We wrote on 19 August that we lean towards reading the rallies as pure beta rather than as a verdict on capital management, because nothing about a 12% day reverses that arithmetic.
Meanwhile the index providers are circling. MSCI did not adopt the crypto-specific exclusion it floated — it dropped that proposal — but subsequently proposed a broader "non-operating companies" screen, which in simulation removes Strategy, Metaplanet and Yellow Cake from the MSCI ACWI IMI. Strategy's float-adjusted market capitalisation in that index was put at about $23.93bn. JPMorgan estimated roughly $2.8bn of passive outflows if MSCI acted alone and about $8.8bn if other index providers adopted similar screens; another cited estimate puts the Strategy-specific impact at $1.8bn to $2.0bn. Passive money does not negotiate.
Why it matters to this crash
The plumbing has changed, and the plumbing is the point. Franklin Templeton has received what it says is the first US regulatory clearance for a tokenised product to sit inside conventional funds, and plans to put its tokenised money-market fund into its own ETFs and mutual funds, as a holding or as collateral, as early as the fourth quarter. We covered it on 20 August. Sandy Kaul's pitch is the honest one — managers can "manage more precisely, capture more of the yield, better and more tightly manage how much cash liquidity they have to hold". The flip side is stated in the same sentence: investors in otherwise conventional funds could end up holding tokenised assets without having sought them out. An instrument designed to be movable at any hour is an instrument that can leave at any hour.
So the exposure now arrives three ways. Through bank credit, where a regulated deposit-taker's book has a contractual link to the coin price. Through fund plumbing, where a settlement technology becomes a liquidity dependency. And through the bill market, where a stablecoin redemption is a Treasury sale.
What has actually happened so far is milder than that sounds, and we should say so. Over the twenty days to 19 August, bitcoin was down 0.6% at $64,333 while Coinbase fell 16.9%, Galaxy Digital 19.6% and Strategy 9.2%. The losses have landed in listed equity — the wrappers and the intermediaries — not in funding markets. When the wrapper falls three times as far as the thing inside it, the market is pricing the financing structure rather than the asset. That is the orderly version. The channels described above are what makes the disorderly version available.
What would make this dangerous
The regulatory template for the bad outcome is not crypto at all. It is the Reserve Primary Fund in September 2008: a write-down of Lehman commercial paper took the net asset value to $0.97, and within four days investors had pulled $23bn and requested $60bn against roughly $62bn of assets. Federal Reserve history records over $400bn withdrawn from prime money market funds in the subsequent run; other accounts put the sector-wide figure nearer $300bn. The ECB's description of the mechanism is the one to hold onto: "widespread redemptions, fire sales and a freeze in short-term funding markets." A product treated as cash-like takes a small loss and everything that resembles it gets redeemed at once.
The things that would turn this from a structure into an event:
A redemption wave at a major stablecoin issuer large enough to force bill sales. The BIS finding runs both directions; watch short-dated Treasury yields on days of large stablecoin outflows. As of our 18 August read, aggregate stablecoin market capitalisation was $14.5bn below its May level. That is drift, not a run. A run looks different, and it looks fast.
Speed. The March 2023 failures are the benchmark for how fast this can go. depositors withdrew $42bn in 24 hours before it was closed on 10 March 2023. Signature depositors submitted $23.3bn of outbound wire requests on 10 March, of which $2.2bn could not be completed that day, and the bridge bank processed $19bn on 13 March. Silvergate, whose SEN network moved dollars between crypto institutions 24/7/365, lost about 68% of deposits — roughly $8.1bn — in a single quarter, sold securities at a $718mn loss, borrowed $4.3bn from the FHLB, and announced voluntary liquidation on 8 March 2023. The FDIC's staff study of deposit flows at the three failed banks, released 14 May 2026, characterises these as among the fastest runs in US history, with wire transfers the dominant channel.
Those runs were still constrained by wire cut-offs and banking hours. — chartered this year, Thiel- and Luckey-backed, raising at an $8bn valuation — plans to use blockchain rails so clients can money at any time, for a depositor base drawn from one industry that reads the same group chats. The last physical brake on a run is being removed on purpose, as a feature.
Collateral that has to be valued at three in the morning. Once a tokenised fund share is pledged, somebody must be able to seize and sell it in a stress, on a chain, at a price. Nobody has watched that happen in a bad market.
Simultaneous margin calls. A 30–50% survives a halving of the collateral. It does not prevent every borrower in a bank's crypto book being called on the same afternoon and selling into the same order book.
An index screen adopted rather than proposed. If MSCI's non-operating-company test goes through, billions of passive dollars become a forced seller of the equity of a company that is already a forced seller of the coin.
And the direction is not fixed. In March 2023 the contagion ran the other way: Circle could not access $3.3bn of USDC reserves held at SVB, and the stablecoin briefly depegged because a traditional bank failed. The pipes that carried it then were narrow and improvised. They are now wide, legal and load-bearing.