Three published figures describe the same year and differ by a factor of more than a hundred. Each is correct. They measure different things.
For 2025, McKinsey put total stablecoin transaction volume at $35 trillion. Boston Consulting Group, working on the same calendar year, put gross transfer volume at about $62 trillion. Both then estimated how much of it was somebody paying somebody for something. McKinsey said $390 billion. BCG said $350 billion to $550 billion.
So the two firms disagree by $27 trillion about the size of the pipe and agree almost exactly about what comes out of it. That gap is the most useful fact in stablecoin measurement, and it explains why any stablecoin payment volume figure quoted without its method attached is unusable.
Why the raw number is so large
On-chain transfer volume counts every movement of a token between addresses. It does not ask why the token moved.
A single dollar can be counted many times in a day. It moves from a user to an exchange, from the exchange to a market maker, through an automated market maker pool, across a bridge, back to a treasury wallet. Each hop is volume. None of it is a payment.
Joe at Maven 11 Capital put the arithmetic in one sentence: “If you have $100,000 of USDC on Solana, you can do ~$136 million of ‘stablecoin volume’ for $1 in fees.” Cheap blockspace means the metric can be manufactured at almost no cost, which is a problem no amount of careful sampling fixes.
The categories doing the inflating are well understood. High-frequency trading bots cycling the same balance. Liquidity provision loops running funds repeatedly through the same contracts. Bridge routing, where one transfer becomes three. Exchange treasury rebalancing between internal wallets. Flash loans, which can move enormous sums with no capital committed at all.
None of that is fraud. It is what a public ledger records when the ledger is also a trading venue, a settlement system, and a database at the same time.
The middle layer, and what it removes
Between the raw figure and genuine payments sits a cleaned number that the industry has settled on calling adjusted volume.
Visa’s dashboard, built with Allium, publishes it. The stated criteria are to remove distortions from high-frequency trading, bots, bridge routing and internal exchange operations. The effect is large: one worked example cut a $3.9 trillion month to $817.5 billion, a reduction near 80 percent, and an earlier illustration from Allium cut a 30-day figure from $2.65 trillion to $265 billion.
The current readings are public. June 2026 came in at $1.79 trillion of adjusted transaction volume, up 63 percent on May and 125 percent on the same month a year earlier, with $10.2 trillion over the trailing twelve months. USDC accounted for $1.21 trillion of the June figure and USDT for $576 billion.
Here is the part that gets skipped. Adjusted volume is not payments volume. It strips out automation, and what remains still includes trading, collateral movement, exchange deposits and withdrawals, and every other legitimate use of a dollar token that has nothing to do with buying anything. Allium’s own framing describes the filter as a heuristic that removes inorganic data, and Visa has publicly asked the market what better methods would look like, which is a more honest position than most published metrics come with.
Treat adjusted volume as a measure of human-directed activity. That is a real and useful thing to count. It is not the payments number.
What the payments number actually contains
McKinsey’s estimate, produced with Artemis, is the narrow one. For 2025, it puts real-world payments at $390 billion, or 0.02 percent of global payments.
The composition matters more than the total. Business-to-business accounted for $226 billion, consumer-to-consumer $77 billion, consumer-to-business $76 billion, and business-to-consumer $11 billion. Most of what qualifies as a stablecoin payment today is one company paying another, usually across a border, usually because the alternative was a correspondent bank and three days.
The consumer-facing categories are small. About $76 billion of people buying things from businesses, and $11 billion moving the other way. Anyone building a retail stablecoin product is building into the thinnest part of the chart, which is an argument about opportunity or about difficulty depending on the week.
McKinsey’s description of the remainder is blunt: most of these transactions are trading, internal transfers and automated processes, not real-world payments.
The second study, and why its agreement matters
BCG published its own attempt in January, and the method is worth understanding because it is the more transparent of the two.
The paper opens by naming the problem directly. Published estimates of stablecoin payments run from tens of billions to more than a trillion dollars, and the reason is that they rely on different definitions and different data. Raw blockchain metrics, the authors write, materially overstate meaningful economic activity, because they capture settlement mechanics instead of use cases.
The paper works in three steps. Strip non-economic activity, meaning bots, protocol mechanics, bridges, and automated market maker routing. Then infer use cases from wallet behavior, using transaction size, frequency, flow symmetry, counterparty diversity and temporal consistency. Then classify anything ambiguous as trading instead of payments, so the error runs toward understatement.
That approach, set out in a paper on what sits behind the numbers, produced roughly $4.2 trillion of economic activity for 2025 out of about $62 trillion gross, and real economy payments of $350 billion to $550 billion inside that. The breakdown by share ran about 40 percent business to business, 25 percent consumer to consumer, 25 percent consumer to business, and 10 percent business to consumer, with every segment growing 50 percent a year or faster.
Now put the two studies side by side. Their gross figures differ by nearly two to one, which tells you that even the raw number depends on which chains, which tokens, and which contract addresses a researcher decided to include. Their payment estimates overlap. McKinsey’s $390 billion sits inside BCG’s range.
Two teams, two methods, two readings of the pipe that differ by nearly half, and the same answer about the water. That convergence is the strongest evidence available that the payments figure is approximately right, and it is more persuasive than either study alone.
The reason the bottom converges while the top does not is that payments have a behavioral signature. A supplier invoice looks like a supplier invoice: a round-ish sum, a stable counterparty, a monthly rhythm, money that arrives and stays. Bot activity has no such shape, and how much of it exists depends entirely on where you draw the boundary of the dataset. Counting the recognizable thing turns out to be easier than counting the noise around it.
The comparison that keeps getting made
Every few months a headline announces that stablecoins have overtaken Visa or Mastercard by transaction volume.
The claim compares a raw ledger total against a card network’s cleared volume. One counts every hop a token makes, including automated ones, at a cost approaching zero. The other counts purchases made by people at merchants, each one authorized, cleared, and settled, each one with a merchant and a price behind it.
Using the figures above, the honest version of that comparison is not close. Set stablecoin real-economy payments for 2025 against any major card network’s annual volume, and the card network is larger by more than an order of magnitude. Set raw transfer volume against it and stablecoins win easily, which tells you nothing except that the two metrics were never the same unit.
This is the specific error the site intends not to make, and the reason for writing the rest of this piece down.
How to read a company’s own volume claim
The same discipline applies one level down, where a processor, wallet or card program announces the volume it moved. These claims are usually accurate and frequently misunderstood, and four questions settle most of them.
Which layer is being counted? A processor that reports every token movement its infrastructure touched is reporting activity, and a processor reporting the value of customer transactions it settled is reporting payments. Both are honest. They are not comparable, and the press release rarely says which one it is.
Is the figure cumulative or annual? Cumulative volume since launch is a legitimate number that grows forever and never shrinks, which makes it the favored metric of any company whose recent months were slower than its earlier ones. A figure with no period attached is almost always cumulative.
Are both legs counted? When a business converts dollars to a token, sends it, and the recipient converts back, a generous method counts the movement more than once. Ask whether the number describes value delivered or transfers observed.
And does it include minting and redemption? Issuance and burn are movements on chain, and whether a dataset treats them as transfers changes the total materially. Most credible providers exclude them. Not all providers say so.
None of these questions is hostile. They are the same questions any analyst would ask of a payments company reporting total payment volume, and stablecoin firms should expect them as the sector matures into one.
Read more: What stablecoin card spend adds up to in 2026 • How Visa and Mastercard settle in stablecoins, side by side
A working definition for this site
Three layers, named every time, with the source and the period attached. That is the standard used in every volume figure published here from now on.
Raw transfer volume is every on-chain movement of a stablecoin between addresses. It is useful for measuring chain activity and blockspace demand. It is never described here as payments.
Adjusted volume is raw volume with automation removed, using a named provider’s filter. It is useful for tracking human-directed activity and the relative weight of chains and tokens. It is described here as activity, never as payments.
Real-economy payments are the value of transfers that pay for goods, services, wages or settlement obligations, estimated from wallet behavior by a named researcher. It is the only layer described here as payment volume, and it always carries the estimate’s range and the year it covers.
Two further rules follow. A stablecoin figure is never compared against a card network, an automated clearing house, or a real-time payment system unless both sides are measuring the same layer, which in practice means both sides are measuring payments. And any figure is dated, because a trailing twelve-month number and a calendar year number are not interchangeable, even when they are close.
Supply is a separate question with better data behind it, and our stablecoin supply coverage handles that side. The point of separating them is that the industry measures its balance sheet well and its cash flows badly, and conflating the two produces most of the bad analysis in this market.
What would make the measurement better
Four changes would move this from estimation toward counting.
Published methodology from every provider quoting an adjusted figure, at the level of detail BCG used, so two analysts can reproduce a number instead of citing it. Visa’s open question about better methods is the right instinct, and it needs answers more than it needs agreement.
Labeled addresses at scale. Much of the classification problem disappears when a payment processor, an exchange, or a card program is identifiable on-chain. The infrastructure for that exists, and the incentive to use it does not, since nobody wants their flows readable by competitors.
A settled treatment of internal transfers. When a business moves its own balance between its own wallets to fund a payroll run, most methods correctly exclude it, and the payroll that follows is often invisible for the same reason. That is a known undercount with no agreed fix.
And disclosure from the companies actually moving the money. Payment processors, card issuers and remittance firms know their own volumes exactly. Few publish them, which is why the best available figures still come from inference instead of addition.
Until those arrive, the number to use is the payments layer, quoted with its range, its year, and the name of whoever produced it. Everything above that layer measures something real. It just does not measure what the headline says it does.