How much of Polygon is Polymarket? We counted every transaction.

There is a version of Polygon that exists in marketing decks. A general purpose payments chain, cheap and fast, carrying stablecoins for people who find Ethereum too expensive. That chain was real about a year ago. In August 2025 the two busiest contracts on Polygon were USDC and USDT, each handling roughly 4.7 percent of transactions, and the money moved the way money moves: lots of small transfers, spread thin.

We went back through twelve months of Polygon blocks to see what the chain looks like now. It does not look like that.

In July 2026, the single busiest contract on Polygon was Polymarket's CTF Exchange V2, which took 25.3 percent of all transactions on the network and burned 39.5 percent of its gas. The second busiest was Polymarket's other exchange, the one that handles multi outcome markets, at 6.1 percent. Two contracts, one company, nearly a third of the chain.

Count everything Polymarket touches and the number goes higher. Over the thirty days to 9 August we sampled 300 random Polygon blocks and read every transaction receipt in them, which is the only way to see both where a transaction was sent and every contract it woke up along the way. Polymarket appears in 38.7 percent of Polygon transactions and consumes 62.2 percent of its gas.

Polygon has one large tenant, and the tenant is a betting site.

The climb took eight months

A year ago Polymarket was a rounding error on Polygon: 2.4 million transactions in August 2025, about 2 percent of the chain. Then it started compounding. By December it was 17 percent. By February, 41. It peaked in March 2026 at 44.6 percent of all Polygon transactions and 69.9 percent of the network's gas, which means that in March, for every three units of computation Polygon performed, two of them were settling somebody's bet on an election or a football match or the price of bitcoin at 12:35 in the morning.

It has cooled since. July came in at 34.4 percent of transactions, the first ten days of August at 31.6. Whether that is Polymarket shrinking or the rest of Polygon recovering matters, so we checked.

It is Polymarket shrinking. Polygon's non Polymarket transaction count barely moved all year.

The growth figures are where this gets awkward for Polygon. It processed 120 million transactions in August 2025 and 197.9 million in July 2026, growth of 64.9 percent, the kind of figure that goes in an annual review. Take Polymarket out of both ends and the chain grew 10.5 percent. Polymarket accounts for 84.2 percent of Polygon's transaction growth over the year and 74.7 percent of its gas growth.

A chain that grew two thirds mostly rented out space to one application.

Gas is the honest metric

Transaction counts flatter cheap activity. A wallet approving a token spends almost nothing; settling a matched order book against a conditional token contract is expensive work. Gas measures the work, and on gas Polymarket's footprint is far larger than its transaction share: 62.2 percent against 38.7.

The gap has been there the whole year and it widens at the peaks. In March, Polymarket was 44.6 percent of transactions and 69.9 percent of gas. Whatever else Polygon is doing, Polymarket is doing the heavy part of it.

There is a practical reading of this for anyone running infrastructure on the chain. Polygon's block space is priced by demand, and the demand is one company's order book. When Polymarket is busy, everybody else's transactions get more expensive, and when Polymarket cools, the chain's gas market cools with it.

About three thousand wallets are doing all of this

Here is where the obvious conclusion breaks, and it is the part we got wrong before we looked properly.

If Polymarket is 39 percent of Polygon's transactions, you would expect it to be a large share of Polygon's users. It is not. On 7 August, Polymarket's contracts received 1,787,815 transactions from 3,004 distinct sending addresses. Polygon that day saw 415,464 distinct senders in total. Polymarket was 0.72 percent of them.

Widen the window and the share falls rather than rises. Over thirty days, Polymarket's senders number 12,566 against the chain's 4.28 million, which is 0.29 percent. Transactions scaled by a factor of thirty four over that period. Senders scaled by four.

The explanation is architectural. Polymarket runs a central limit order book where users sign orders off chain and a small fleet of operator addresses submits the matched trades. The average Polymarket sending address made 595 transactions on 7 August. No human does that. These are relayers, working in shifts.

So the finding cuts both ways. Polygon's throughput is Polymarket. Polygon's address count is almost untouched by it. Anyone quoting active addresses as evidence of Polygon's health is measuring something Polymarket barely participates in, and anyone quoting transaction counts is mostly quoting Polymarket.

This also means we cannot tell you how many people use Polymarket. The senders are infrastructure. The traders are behind them, invisible at this layer.

What is left when you take Polymarket out

We ranked every destination contract on Polygon for July 2026 to see what the rest of the chain is made of. After Polymarket, the largest source of activity is not a payments app or a game or a DeFi protocol anyone would recognise.

It is a yield farm called Longinus. Seven addresses running stake, unstake and claimInterest against the LGNS and sLGNS tokens produced 42,247,905 transactions in July, 21.4 percent of the entire chain. They consumed 13.5 percent of its gas, which tells you the transactions are small and repetitive.

Polymarket and Longinus together are 56 percent of Polygon's transactions on the conservative count, before the indirect calls are added.

The USDC transfers are still there, at 4.4 percent. USDT at 1.9. The stablecoin payments chain still exists. It is just no longer the main thing happening.

The money is real

Activity metrics can be gamed and often are, so we checked the money separately.

Polymarket settled $3.68 billion of notional in July 2026. Every decentralised exchange on Polygon put together settled $7.78 billion. One prediction market is now running at about half the volume of the chain's entire trading sector, up from a ninth a year ago. As a share of the two combined, Polymarket is 32.1 percent, which lands close to its transaction share of 34.4 percent. When two unrelated measurements land in the same place, it usually means neither is an artifact.

That $3.68 billion figure comes with a caution we will come back to, because getting it wrong by a factor of two is the easiest mistake in this dataset.

One stablecoin belongs to Polymarket now

Polygon carries four dollar tokens that matter: native USDC issued by Circle, the older bridged USDC.e, USDT, and DAI. Together they moved $86.3 billion in July 2026.

Polymarket touched $32.7 billion of it, 37.9 percent. But that number is misleading in an interesting way, because the exposure is not spread across the four. It sits almost entirely on one.

Of all USDC.e moved on Polygon in July, 70.2 percent passed through a Polymarket contract. Of all USDC.e transfers, 94.8 percent did. Native USDC, by contrast, is 0.0007 percent Polymarket. USDT is zero to four decimal places. DAI is zero outright.

A year ago USDC.e was 4.3 percent Polymarket.

Two things happened at once. Polymarket's USDC.e flow grew twelvefold, from $2.7 billion to $32.7 billion. And everybody else left: non Polymarket USDC.e volume fell from $59.2 billion to $13.9 billion while native USDC volume more than doubled, up 139.5 percent to $29.7 billion. That second move has nothing to do with Polymarket. Circle has been pushing users off the bridged token onto the native one for years, and on Polygon they went.

Polymarket did not go. The bridged stablecoin Polygon inherited from its bridge era is now, by volume, mostly one prediction market's settlement rail, and by transfer count almost entirely so.

There is a fifth token, and it is Polymarket's own. When the exchange moved to V2 it introduced pUSD, a collateral token it issues itself, which moved $28.9 billion across 359.6 million transfers in July. Every one of those is Polymarket by construction.

We have deliberately not added that $28.9 billion to anything. pUSD is minted against USDC.e and redeemed back into it: the collateral token, the two CTF adapters and the conditional token contract account for $30.0 billion of the $32.7 billion of Polymarket USDC.e flow. The pUSD and the USDC.e are the same dollars wearing two hats, one leg wrapping and the other unwrapping. Add them and you would report $61.6 billion of Polymarket stablecoin volume, roughly double the real figure. It is the same mistake as counting both sides of a matched trade, in a different costume.

How we counted, and the two ways to get it wrong

The perimeter is twenty contracts, taken from Polymarket's own repositories rather than from memory: the two V2 exchanges, the two dead V1 exchanges, the conditional token contract, the collateral token, the proxy and Safe factories, the relay hub, the fee modules and the oracle adapters. Every one was checked to carry bytecode on chain 137.

One thing that check caught: Polymarket has migrated to a second generation of its exchange, and V1 is effectively dead. In an identical 2,000 block window the V2 exchange emitted 160,450 events and the V1 exchange emitted 224. The old neg risk exchange emitted none. V2 also settles in its own collateral token, pUSD, rather than the bridged USDC that Polymarket used for years. Any analysis of Polymarket keyed to USDC.e is measuring a contract nobody trades on any more.

The first trap is counting events. A single Polymarket trade lights up several contracts at once. On 7 August, Polymarket's perimeter emitted 19,267,067 event logs across 2,046,797 transactions. Count the logs and you overstate the answer by 9.4 times. Everything here is counted once per transaction hash, whatever else happens inside it.

The second trap is volume, and it is worth walking through because it is invisible unless you open a transaction.

Polymarket's exchange emits an OrderFilled event for every order in a match, one for the taker and one for each maker on the other side. Take transaction 0x7aba72c5, which is one person selling five shares. It contains three OrderFilled events: two makers buying the "Down" side for $2.40 between them, and the taker's fill of $2.60 on the "Up" side. Add up the collateral in all three and you get $5.00, which is not the size of the trade. It is five shares at a dollar of collateral each, both sides of the same bet, counted twice.

So the raw sum has to be halved. We did not want to take that on faith, so we tested it two ways: against DefiLlama's daily figures, where the halved number lands at 0.99 for every day we could check, and against Polymarket's decoded trades in Bitquery's own Polygon database for the whole of July, which produced $3,657,258,354 against DefiLlama's $3,681,073,336. That is 0.9935. The halving is right.

Two independent pipelines were run end to end, one on raw block receipts pulled from public Polygon nodes and one on Bitquery's indexed tables. Where they measure the same thing they agree: direct attribution comes out at 32.0 percent of transactions from the sample and 32.87 percent from the exact count, inside the sampling interval.

What we are not saying

We measured activity, not revenue, and Polygon does not bill Polymarket the way a landlord bills a tenant. A chain being busy on behalf of one application is not automatically fragile; Polymarket chose Polygon for real reasons and has stayed through a full exchange rebuild.

But the concentration is now a fact about Polygon rather than a fact about Polymarket. If the prediction market moved to its own chain, as several large applications eventually do, Polygon's transaction count would fall by roughly a third and its gas consumption by close to two thirds overnight. The March peak, when Polymarket ran at 69.9 percent of network gas, is the clearest picture of what that dependency looks like at full stretch.

The chain in the marketing deck is still down there. It is now the minority tenant in its own building.