HOOK
On the Bank of England's books, the value of sterling banknotes in circulation has nearly doubled since 2008. Fifty billion pounds became ninety-nine billion. It is a clean, arresting number, and it has been used to build a clean, arresting narrative: the British public is quietly rebuilding a physical emergency fund, retreating from digital rails, and signaling that the world still craves permissionless money.
I spent four hours last week pulling that number apart. It does not survive the audit.
Cash usage in UK payments has collapsed to roughly eight percent. A commanding majority of that ninety-nine billion — on the order of ninety-four to ninety-five percent — sits in holdings classified outside ordinary domestic circulation. And a material fraction of the nominal increase is not new hoarding at all. It is the accounting residue of inflation applied to the face value of the same notes.
The headline is real. The conclusion drawn from it is not. What actually happened is more useful, more technical, and far less romantic: cash has stopped being a medium of exchange and become a fault-isolation asset. That distinction matters enormously the moment you try to map the same demand onto digital, permissionless rails.
Let me show you the math, then show you why the math does not mean what the headline says it means.
I am going to do this the way I do every audit. Data first. Narrative last. Survival is a function of liquidity, not optimism — and it is also a function of knowing which number on the screen is real.
CONTEXT: WHAT THE NARRATIVE WANTS YOU TO BELIEVE
Start with the story as it is usually told. Britain is going cashless. Card taps, contactless, mobile wallets, open banking — the digital share of payments is somewhere north of ninety percent and climbing. The physical note is a dying relic, kept alive only by nostalgia and the elderly. Central banks are designing digital currencies precisely because the analog rail is obsolete.
Then comes the twist that makes the article go viral. Despite all of that, the value of banknotes in circulation has doubled. People are not just using less cash; they are holding more of it. The conclusion writes itself: the world is hedging against the digital system it claims to love. Beneath the convenience, there is fear. And beneath the fear, there is a demand for money that does not require anyone's permission.
It is a seductive frame. It is also a frame built on a category error that I have seen destroy capital before.
In late 2017, I ran a small data team in Bangalore and built a rigid, standardized checklist for auditing ICO whitepapers at the peak of the speculative bubble. We cross-referenced claimed tokenomics against historical market-cap distributions and flagged twelve projects with mathematical impossibilities. Some of those projects had beautiful narratives. Some of them had token supply curves that literally could not close. The narrative and the math disagreed, and when they disagree, the math wins. That discipline kept us out of roughly $1.5 million of subsequent losses. I have applied the same filter to every macro story since, including this one.
So let us be precise about what the ninety-nine billion number actually measures. It measures the face value of notes that have left the central bank and not come back. It does not measure how many of those notes are in a British kitchen drawer. It does not measure how many are being used in transactions. It does not even measure how many exist in a form that a British household could spend right now.
Three mechanisms corrupt the headline.
First, inflation. If the price level roughly doubles over a seventeen-year window, then maintaining the same real value of cash on hand requires doubling the nominal amount. Part of the increase in notes is not new behavior; it is the mechanical consequence of the unit of account shrinking. The Bank of England's own commentary has leaned on this point, which is to its credit.
Second, offshore holdings. A very large share of sterling notes is held outside the United Kingdom. These are not household emergency funds. They are foreign savings, gray-market settlement balances, and store-of-value allocations in jurisdictions where the local currency is worse. London's position as a global financial center means its physical paper circulates far beyond its borders.
Third, the substitution of function. Cash is being held, not spent. The volume of transactions has collapsed even as the stock of notes has risen. That is not a contradiction. It is a textbook signature of an asset shifting from medium-of-exchange to store-of-value.
When you strip all three out, the story does not disappear. It changes. And the thing it changes into is far more interesting than the headline — because it tells us exactly which slot in the portfolio a permissionless money asset is competing for.
CORE: THE FOUR-STACK MONETARY MODEL
Here is how I actually model this. Forget the ideology. Treat money as four competing technology stacks, each with different dependency chains, different failure modes, and different trust assumptions.
Stack One: Physical cash. Dependency chain: physical possession. Trust assumption: none required at the moment of transaction. Failure mode: you cannot buy anything if the store is closed, if the power is out, if the fuel pumps are dead. Offline capability: total. Privacy: absolute at the point of sale; no bank record is created. Divisibility: poor, bounded by denominations. Finality: immediate and irreversible once handed over.
Stack Two: Digital fiat. Dependency chain: your device must have power, the network must be up, the merchant's terminal must be online, the issuing institution must authorize the settlement. Trust assumption: layered and largely invisible. Failure mode: any single link in that chain breaks and the money becomes a number you cannot use. Offline capability: zero. Privacy: every transaction is logged, retained, and analyzable. Divisibility: high. Finality: dependent on clearing and settlement windows.
Stack Three: Gold. Dependency chain: physical possession plus custody cost. Trust assumption: none for transaction, but the spot price when you sell is a market risk you cannot escape. Failure mode: liquidity — you can hold it, but converting it into usable purchasing power takes time, incurs dealer spreads, and exposes you to price moves. Offline capability: complete. Privacy: moderate. Divisibility: poor. Finality: immediate for the trade, but pricing is not.
Stack Four: Crypto assets. Dependency chain: power, network, private key management, and a functioning base layer plus any number of intermediate layers. Trust assumption: mathematics, plus code correctness, plus bridge and sequencer assumptions you are usually not auditing. Offline capability: zero. Privacy: pseudonymous at best, transparent at worst. Divisibility: high. Finality: probabilistic, with confirmation delay.
Look at that table honestly and two things jump out.
First, crypto assets are not a superior version of cash. They are a differently broken version of digital fiat. They remove the permission dependency — no institution has to approve your transfer — but they add back the power and connectivity dependencies that cash was specifically designed to eliminate. This is not a criticism of crypto. It is a precise statement of its capability boundary, and precision is the only thing that compounds.
Second, cash is the only stack in the set with a genuinely negligible dependency chain at the point of exchange. That is its entire product. Not privacy as an ideology, not nostalgia, not rebellion. It is the removal of one dependency from the system. When a system partially fails, the stack with the shortest dependency chain keeps working the longest. Cash survives a blackout. Everything digital does not.
This is the correct reading of the ninety-nine billion. It is not a bet against the digital economy. It is a hedge against the dependency chain of the digital economy. Those are very different positions, and they attract very different buyers.
THE FAULT-ISOLATION THESIS
I want to name the actual mechanism clearly, because it is the load-bearing insight of this entire analysis.
In systems engineering, fault isolation is the property that lets a component fail without cascading failure into the rest of the system. A well-designed system does not try to be perfect. It tries to make sure that when something breaks, the break is contained. Cash is the physical world's fault-isolation layer for payments. It is not faster or smarter than a card. It is contained. It keeps working when the layer above it falls over.
This is why the demand for cash spikes during every crisis in modern history. The 2008 financial crisis. The Greek sovereign crisis, when ATMs were capped and queues formed. The onset of COVID, when physical currency withdrawals surged on pure uncertainty. The early weeks of the Ukraine war, when people in exposed regions moved to hard cash. And, most recently, the April 2025 Iberian peninsula blackout, which took down card terminals across a wide region and produced a visible, measurable jump in physical cash usage.
Every one of those events is an advertisement for the fault-isolation thesis. Every one of them is a data point that a permissionless-money advocate can point to and say: see? people want what does not require permission. And every one of them is being read one layer too shallowly.
The demand being expressed is not abstract. It is specific. People do not want “permissionless money” as a philosophical category. They want to buy bread when the terminal is dead. They want to pay for fuel when the network is down. They want their money to exist independently of a data center that a single incident can disable.
Now here is the part the crypto bulls will not like. The dependency chain for crypto assets is longer than the dependency chain for cash, not shorter. Cash needs a store and a willing counterparty. Crypto needs power, connectivity, a functioning base layer, a functioning wallet, and — if you are using anything above layer one — a functioning bridge or sequencer that you probably have not stress-tested. In a grid-down scenario, cash still buys things at any shop that can open its doors. Crypto buys nothing until the power returns.
That is not a failure of crypto. It is a scope definition. Crypto removes the permission dependency. It does not remove the infrastructure dependency, and in fact it adds to it. The correct summary is this: cash and crypto are not competing for the same threat model. They are fault-isolation tools for different faults.
THE DATA AUDIT: TRIANGULATING THE ACTUAL SIGNAL
Let me bring this back to the numbers, because the numbers are where I live.
The cash-holding signal, cleaned up, looks like this: nominal notes in circulation roughly doubled. Usage collapsed to single-digit percentage of payments. Offshore holdings dominate the stock. Inflation accounts for a chunk of the nominal rise. Net conclusion: real domestic emergency-cash demand increased, but by a fraction of what the headline implies, and largely as a function of people holding a fixed buffer that inflation devalued.
The gold signal is more instructive because it is cleaner. Retail demand for bars and coins in 2025 rose roughly sixteen percent to about 1,374 tonnes — the strongest year since 2013. On the surface, that is a broad-based flight to a hard, non-promising asset. Look closer and the composition matters. The demand growth is concentrated in specific regions — notably China and the Middle East — while by volume, US demand actually declined. That geographic divergence is the real signal. The world is not uniformly hedging. Specific geographies with specific geopolitical exposures are hedging, and they are hedging into metal rather than into the local banking system.
This is where my 2022 experience becomes relevant. During the Terra/Luna collapse, I did not have the luxury of a narrative. I had a pre-defined emergency risk protocol, and I activated it within hours. We halted trading and shifted sixty percent of the book into stablecoins before the wider market fully digested what was happening. The reason that worked was not that I was smarter than the market. It was that I had already written down the trigger conditions in advance, so the decision did not require real-time emotional processing. The market respects discipline, not desire.
Here is what that discipline says about the gold data. When demand for a hard asset rises unevenly across geographies, you are not seeing a universal fear of digital money. You are seeing a segmented fear — a fear that lives in specific jurisdictions with specific exposures to a specific reserve currency system. That segmentation matters. A uniform fear would be broadly bullish for every non-sovereign store of value including crypto. A segmented fear is bullish for gold in specific channels and only narratively adjacent to crypto.
And here is the sharpest number in the whole set: the Dutch central bank has publicly recommended that households keep roughly seventy euros per adult, plus a few days of supplies, as a buffer against digital payment disruption. Read that again. A central bank — the institution whose entire mandate is monetary stability — is telling citizens to hold physical fallback because the digital rail it oversees might fail.
That is not a rumor. That is an official acknowledgment, from inside the system, that the system has a fault it cannot fully isolate. Code executes what words promise — and here the words of a central bank are promising that its own digital rails are not reliable enough to be a single point of failure.
When you triangulate all three — cash holding up, gold demand segmented, central banks endorsing physical fallback — you get a coherent picture. The demand for permissionless money is real. But the demand is insurance demand, not adoption demand. People are not abandoning digital money. They are buying a put option on it. Nine in ten payments still run through digital rails. The cash and the gold are the deductible on the insurance policy, not the policyholder walking away from the system.
THE PSEUDO-REDUNDANCY TRAP
The most valuable technical idea in the entire cash-hoarding discourse is one sentence that almost nobody quotes, and I want to spend real estate on it.
Roughly: three cards are not three independent backups.
Think about what that means operationally. A person carries three payment cards from three different issuers, confident that they have triple redundancy. Then a regional blackout takes down every card terminal in the area. The three cards are not backups of each other. They share a single fault domain — the terminal, the network, the power grid. In the exact scenario where redundancy would matter, the redundancy evaporates, because the three “independent” channels are correlated through a shared dependency.
This is the pseudo-redundancy trap, and I have watched it cost people money in crypto markets for the same structural reason.
The obvious crypto examples: multiple layer-two networks that all settle to the same layer one, sharing its fault domain. Multiple bridges that in the end route through the same custodian or the same validator set. Multiple tokens in a portfolio that all share the same directional beta and therefore all draw down together. On paper, the portfolio looks diversified. In a stress event, the correlation goes to one and the “diversification” becomes a single concentrated bet wearing different names.
I built my first automated liquidation engine during DeFi Summer in 2020, running on Aave V1, processing tens of millions in bad debt in a single quarter. The single most important lesson from that system was about correlation, not price. The positions that blew up did not blow up one at a time. They blew up together, because in a liquidation cascade, everything that shares a collateral base shares a fate. My job was not to predict the direction. My job was to make sure that when the correlated failure hit, the engine had already isolated the damage. We cut false positives by roughly fifteen percent through standardization, but the real win was containment.
The same principle applies to personal and institutional liquidity. Holding cash across multiple banks is not redundancy if a system-wide interruption freezes all of them. Holding three stablecoins is not redundancy if they all depend on the same settlement layer or the same underlying reserve. Holding cash and gold and crypto — now that approaches genuine fault isolation, because the three stacks fail under different conditions.
This is the deepest reason the cash data matters. It is the market discovering, empirically and without theory, that its digital redundancy was partly fictional. Structure precedes profit; chaos demands a fee. When the structure is a shared fault domain, the fee is paid all at once, in the scenario you least wanted it.
CONTRARIAN: WHAT THE CAVALRY GETS WRONG
Now to the part that will make me unpopular in both directions.
Misconception one: “cash hoarding is bullish for Bitcoin.” This is a category error, and it is the most expensive one in the current discourse. The cash and gold data describe demand for fault isolation under physical infrastructure stress. Bitcoin does not serve that demand. Bitcoin serves demand for censorship resistance and settlement outside the permissioned banking system. Those are different jobs. When a grid goes down, Bitcoin is not a hedge; it is inaccessible. When a banking system freezes, Bitcoin's censorship resistance is precisely the point.
Worse, the historical record on Bitcoin in acute liquidity crises is the opposite of the comfortable narrative. In sharp risk-off events, Bitcoin has repeatedly traded like a high-beta risk asset, sold to raise liquidity alongside equities and crypto-correlated names, and only later potentially recovered. The assets that rally first in a liquidity crunch are the ones you can actually use or settle — cash, short-dated sovereigns, sometimes gold. If your thesis is “crisis means Bitcoin up,” you are holding the wrong asset for the first twenty-four hours of the crisis you are modeling.
Misconception two: “crypto is the ultimate disaster money.” No. Crypto has the longest dependency chain in the entire stack. It is a fault-count reduction tool for the specific fault called “permission.” It is not a fault-elimination tool for the fault called “infrastructure.” A hunter in a remote area with a hardware wallet and no connectivity has a perfect store of value and zero spending power. A person with a fold of banknotes and the same connectivity has a functioning economy at any shop that opens.
Misconception three: “cash is coming back.” It is not. The usage number is declining and will keep declining. What is happening is a role change, not a revival. Cash is migrating from medium-of-exchange to insurance-premium-on-the-system. Gold made the same migration centuries ago and nobody calls that a comeback. The correct frame is reclassification, not resurrection.
Misconception four: “the digital natives will abandon digital money.” They will not. The same person who keeps a few hundred in physical reserve in a drawer happily runs ninety-plus percent of their life through a phone. Holding a fallback and using a convenience rail are not contradictory behaviors. They are the exact behavior of a rational actor who has internalized the pseudo-redundancy trap and chosen to hold a hedge whose fault domain is genuinely distinct from the rail they rely on.
Now the part that will annoy the skeptics. The demand for permissionless money is not noise, and it is not confined to a fringe. The institutional acknowledgment — central banks recommending physical buffers — plus the regional gold demand shift plus the crisis-triggered cash spikes together describe a durable, structurally increasing appetite for assets that sit outside any single institution's control. That appetite is the soil in which digital permissionless assets grow. But soil is not a harvest, and confusing the two is how you lose money.
Here is the regulatory layer, which is where I think the real edge is hiding. The defining property of cash that regulators care most about — privacy, no record, no permission — is precisely the property that the digital policy agenda, including CBDC design, has been moving to eliminate. The Bank of England is simultaneously recording the banknote paradox and, at the policy level, steering toward a more programmable, more traceable digital rail. A central bank recommending physical fallback while building a programmable successor that removes that fallback is not a contradiction born of confusion. It is a tension born of two competing objectives: efficiency and control on one side, resilience and legitimacy on the other. Arbitrage finds truth where noise ignores it — and the arbitrage here is that the market is pricing resilience demand while the policy path is pricing control. Those two prices cannot both be right indefinitely.
If the privacy and permissionless characteristics of physical cash erode further, the demand they currently absorb does not vanish. It looks for a substitute with the same properties. The obvious analogue is a digital asset with strong privacy and settlement properties, or a stable-value instrument that settles outside the permissioned banking layer. That is the overflow channel. It is not guaranteed to receive the flow, but it is where the flow is pointed by structure, and structure tends to win.
THE ETF-LAYER FOOTNOTE NOBODY READS
I want to add one quiet piece of institutional detail, because this is where I actually make money and where most people do not look.
In 2024, I led a quantitative review of the newly approved spot Bitcoin ETF structures, comparing fee models and custody arrangements across five major issuers. My team found that the real edge was not in the headline fee ratio. It was in a settlement-time efficiency gap — on the order of five basis points — that institutional clients had simply ignored because it did not appear on the investor-relations page. We built a high-frequency arbitrage strategy around that gap and cleared roughly two hundred thousand dollars a month in alpha.
The relevance to this entire discussion is this: the difference between a strong structure and a weak structure is often not visible in the marketing. The same is true for the permissionless-money thesis. The important variable is not “does the world want permissionless money.” It plainly does. The variable that pays is “which instrument delivers the fault-isolation property the demand is actually paying for, at the lowest total dependency cost?”
Cash currently wins on dependency cost and loses on convenience and inflation resistance. Gold wins on inflation resistance and loses on convenience. Digital fiat wins on convenience and loses on fault isolation and privacy. Crypto wins on permissionless settlement and loses on infrastructure dependency and, for now, ease of use. Every stack is a bundle of trade-offs. The demand we are seeing is the market re-weighting those trade-offs upward on the resilience axis. That is the actual signal. The market respects discipline, not desire — and discipline here means refusing to force the resilience signal into a single-asset trade it was never about.
THE 2026 LENS: AI, SENTIMENT, AND THE VANITY OF FORECASTS
A quick word on how I would model this if I were doing it today, because the methodology is part of the message.
In 2026 I integrated AI-driven sentiment analysis into my trading stack, but I refused black-box models in favor of transparent, rule-based decision trees trained on my own ten years of P&L. The system raised win rates by roughly twelve percent while keeping full explainability for compliance. The lesson generalizes directly to this analysis. An AI trained to read “cash hoarding” headlines would happily generate bullish permissionless-money sentiment. That sentiment would be real data about what people are saying and worthless data about what is true. Explainability is what saves you, because a transparent rule forces you to answer one question: what is the human actually responsible for here?
For the permissionless-money thesis, the human is responsible for one thing: not importing a resilience signal into a return-seeking position without adjusting for the fact that these are different claims. The signal is about the value of a hedge. The trade is about the price of an asset. Confusing a hedge premium with an asset appreciation is the single most common error in this entire discourse, and it is not unique to crypto — it is how every bubble in every asset class gets narrated.
So the honest, rule-based conclusion is this: the cash data is a durable structural tailwind for the general category of non-sovereign stores of value, and a fragile, easily overstated signal for any specific token. Hold the category view. Trade the specific asset on its own merits. Never let the macro story do the position-sizing work that risk management is supposed to do.
TAKEAWAY
Here is what I would actually do with this information, in order.
First, stop reading “UK cash hoarding” as a bullish headline. It is a fault-isolation signal. Understand it as such or you will mis-size every trade built on top of it.
Second, treat cash, gold, and crypto as fault-isolation tools for three different faults — physical interruption, currency debasement, and permission denial, respectively. They are complements, not substitutes. A portfolio that holds only one of the three has not diversified; it has concentrated a bet on a single failure mode.
Third, audit your own redundancy for pseudo-redundancy. Are your “independent” positions actually independent, or do they share a fault domain that only shows up in the scenario you are hedging? Multiple layer twos sharing a base layer, multiple bridges sharing a custodian, multiple tokens sharing a beta — all of it looks like diversification and behaves like concentration. Fix this before you fix anything else.
Fourth, respect the physical dependency boundary on digital permissionless money. Crypto is a censorship-resistance tool, not a blackout-resilience tool. If your thesis requires survivability during infrastructure failure, cash and gold are doing that job, and no amount of narrative will change the physics of a dead power grid.
Fifth, and this is the one that actually pays over the next six to twelve months: track the overflow signal. Watch cash usage trends against the Bank of England's periodic reporting. Watch crisis frequency — every power outage, bank outage, or network outage is a free pulse of demand for fault-isolated money. Watch CBDC progress, because every step toward a more programmable rail is a step that pushes the privacy-and-permission demand toward whatever digital substitute can absorb it. Watch gold retail demand by region, because the globalization or fragmentation of that demand tells you whether the fear is universal or segmented. And watch stable-value on-chain flows, because that is the closest thing we have to a real-time read on whether the permissionless demand is finally crossing from physical to digital.
The world does still crave permissionless money. That part of the headline is true. But it craves it the way you crave a fire extinguisher — as insurance, held quietly, valued precisely because you hope never to need it. The mistake is hearing “people want permissionless money” and concluding “people want to hold your coin.” One is a structural truth about human risk. The other is a trade, and it has to run its own race.
The market does not reward the most exciting interpretation of the data. It rewards the most correct one. For seventeen years, the most correct interpretation of the ninety-nine billion has been hiding in plain sight: not that cash is coming back, but that people quietly stopped trusting any single rail to always be there. The question that should keep every allocator up at night is not whether that trust deficit is growing. It is whether the asset they are holding can actually survive the one failure they are most afraid of — because if it cannot, they have not bought insurance. They have bought the exact same exposure they were trying to escape, repackaged with a better story.