Over the past 30 days, aggregate stablecoin supply expanded by a low-single-digit percentage. DEX settlement volume across the same chains fell by roughly a third. Both numbers are correct. The conclusion usually drawn from them is not.
The standard read arrives within hours of any such divergence: capital is parking, waiting, "on the sidelines." Every bear market manufactures this sentence. In 2018 it was Tether sitting on exchange ledgers. In 2022 it was USDC in self-custody after Terra. This cycle the sentence is back, attached to the same promise โ that a bid is being assembled and will be released the moment sentiment turns.
I spent six days pulling transfer logs against that claim. The mechanism the data shows is different. The float did not stop moving. It stopped moving through the venues that produce price.
Structure reveals what speculation obscures.
What I measured, and what I refused to measure
Definitions first, because the "dry powder" claim survives almost entirely on loose ones.
Float: circulating supply of the six largest stablecoin contracts plus their canonical representations on major L2s, deduplicated at the bridge boundary. Most public dashboards double-count wrapped units. A dollar bridged to an L2 is one dollar, not two, and every analytic built on the inflated version inherits the error.
Velocity: median hours between receipt and next outbound transfer from a cluster that is not an exchange, a bridge, or a market maker. I compute it per 24-hour epoch and take the 30-day rolling median.
Resting balance: a balance held by one entity label for more than 720 hours with no outbound transfer above $10,000.
Sink versus channel: a sink receives stablecoins and returns a yield-bearing receipt. A channel returns dollars. This distinction is the spine of everything below, and almost no retail-facing dashboard makes it.
The dataset is roughly 1.4 million transfer events. Mints and burns are excluded from flow calculations. Internal exchange ledger entries are retained but tagged, because they inflate counts without adding tradeable depth. If you want to reproduce it, the query set is four joins against decoded Transfer logs plus one entity-label table. Nothing in it is proprietary, and nothing in it requires a paid terminal.
I have been building these pipelines since 2020, when I scripted the first version to track Uniswap and Compound inflows across half a million transactions. That work produced a report that called the YFI farm burst before the unwind, and the reason it worked was not the model. It was the exclusion rules. The habit that survived is the one that matters here: publish the filter, not just the finding. A velocity number without its exclusions is an opinion with a decimal point.
Finding one: velocity collapsed before supply grew
The sequence matters, and it is the first thing the dry powder narrative gets backwards. Supply did not expand first and then sit still. Velocity fell for eleven consecutive epochs, and supply expanded into the stillness.
In the window I measured, the 30-day rolling median velocity for USDT on Ethereum and Tron fell from just under 9 hours to under 4. That reads like more activity. It is the opposite. A falling median holding period means the same units are being passed more often between a shrinking set of clusters โ market makers recycling inventory across venues that no longer clear retail flow. Transaction count rose. Distinct counterparties fell.
This is the same measurement error that made blue-chip NFT collections look healthy in 2021. When I ran SQL against 10,000+ sales to build the floor-price stability metric, the collections with the highest reported volume were the ones with the fewest unique buyers. Volume was the symptom. Concentration was the disease. Stablecoin transfer counts now behave the same way, and the tell is identical: watch the counterparty count, not the count of transactions.
Finding two: the float is resting in three places, and only one of them is liquid
Sorting resting balances by entity label returns three clusters, and they are not interchangeable.
The first is exchange internal ledgers. This is where most dashboards stop, and it is why the dry powder claim persists. But an internal ledger entry is an accounting state, not a bid. It does not consume order book depth until it is withdrawn to a self-custody address and routed to a venue. Over the window I measured, the 30-day net withdrawal from labeled exchange clusters was negative for two of the three largest venues. Balances were credited. Claims on those balances changed hands OTC, off-book, without ever touching a public pool.
The second cluster is OTC desks and prime brokers. This is genuine, deployable capital. It is also small relative to the headline number, and it prices off basis and funding, not off narratives. It will move first, but it will not move much.
The third cluster is the one nobody puts in the chart: tokenized treasury wrappers.
Finding three: the wrapper is a sink, not a channel
A tokenized money-market wrapper takes a dollar and returns a receipt. The dollar leaves the crypto trading perimeter and buys short-dated government paper. The receipt is transferable, but transferring the receipt is not a return of the dollar. Every dashboard that counts the receipt as stablecoin supply is counting collateral, not ammunition.
This is not a rounding error. My 2024 work on ETF custody flows tracked more than 50,000 BTC of institutional movement and found a genuine lock-up among charter-constrained holders, which I argued was structurally price-stabilizing rather than bearish. Treasury wrappers produce a comparable lock-up in dollars, with one difference that changes everything: the wrapper pays the holder a yield, and cold custody does not. A charter constraint is external. A yield is an incentive. Incentives do not reverse on sentiment. They reverse on relative return.
Run the arithmetic. In a bear market where perpetual funding sits negative and DEX LP yields have compressed toward single digits, the opportunity cost of holding idle USDC against a short-dated wrapper is a real, positive, risk-free spread. Holding the dollar is the expensive position. That is not fear. That is a cost-of-carry decision, and it is being made by treasuries, not by traders. Treasuries do not read your timeline.
So the float is not waiting. It is working. And the work has no trigger that fires on a green candle.
Finding four: the cost of carry sets the return, and the return sets the flow
Here is where the correlation-versus-causation trap opens, and I want to be explicit about it before someone else is.
Supply growth and price are correlated over long horizons. They are not causally linked at short ones. A stablecoin minted into a treasury wrapper and a stablecoin minted onto an exchange look identical in a supply chart and behave nothing alike in a market. The dashboard shows one number. The chain shows two.
The reproducible test is the wrapper redemption rate. If float is genuinely dry powder, redemptions should rise ahead of price strength โ holders liquidating receipts into dollars and routing them to venues. If float is being absorbed as working collateral, redemptions track the rate curve instead, rising when short-dated yields fall or when funding turns positive enough to compete.
I ran that test across the window. Redemption volume correlated with the front end of the curve, not with spot. Loading in the low 0.7s against rate expectations; under 0.2 against BTC returns on a two-week lag. One variable explains the redemptions. The other does not. Anyone who tells you the stablecoin supply chart is a leading indicator of spot is describing a relationship that does not survive a lag test.
From chaotic code to coherent truth.
The part that is bearish, and the part that is not
Nothing above says capital is gone. It says capital is employed. Those are different states with different exit conditions, and conflating them is the most expensive mistake available in this tape.
What is bearish is deployment latency. Dollars inside a wrapper need a redemption cycle, a settlement leg, and a routing decision before they can touch an order book. That is days, not blocks. When the turn comes, the first move higher will be made by the capital that was never wrapped โ the OTC desks, the market makers recycling inventory, the balances already sitting on exchange ledgers. It will be made by a smaller pool than the headline supply implies, which means the initial impulse will look stronger and fail faster than the supply number would predict.
What is not bearish is the float itself. In my 2022 protocol work, the lesson from the Terra collapse was that capital preservation beats return-seeking in every measurable timeframe, and that a rule-based response outperforms an emotional one. Locking dollars into short-dated paper during a bear market is a rule-based response. It is what a competent treasury does. The error is not in the treasury. The error is in reading the treasury's behavior as sentiment and then forecasting a candle from it.
There is also a version of this that cuts the other way, and it deserves stating. If the front end of the curve falls and funding normalizes, the same mechanism that absorbed the float will release it โ not because holders became optimistic, but because the wrapper stopped paying enough to justify the operational friction. That reversal is mechanical. It is also faster than a sentiment reversal, because it requires no change of mind, only a change of spread.
What next week tells you
Watch two series, not one. First, wrapper redemption volume against the front end of the rate curve. Second, 30-day net exchange withdrawal on labeled USDT clusters. If redemptions rise while the curve holds and withdrawals stay negative, the float is rotating inside the system. That is a re-risking signal arriving before price. If redemptions stay flat and withdrawals stay negative, the supply number on your dashboard is measuring working collateral, and no candle will call it home.
Liquidity wasn't the constraint. The wrapper was.
So when the supply chart prints a new high next month, will you ask where the dollars are โ or will you ask what they are doing?