Analysis

One venue's bad week made Ethereum the top DEX chain of the summer

Ethereum's $348B of DEX volume over 13 weeks ranks it first among chains, but $214B of it lands in a single week in which Balancer reports $206B, about 3,900 times its median. Replace that week with Ethereum's median and it falls to third, behind BNB Chain and Solana.

Data through 4 min read

Question

Which chain had the most DEX volume this summer? The obvious answer, summing Dune's curated dex.trades table by chain for the 13 complete weeks from June 22 to September 14, is Ethereum, and by a wide margin. The question this analysis asks is whether that answer survives looking at the weeks one at a time, because a ranking built on a sum is only as good as its worst week.

Approach

Both series are Dune's own maintained queries, read as stored results and re-aggregated here. The chain series covers every EVM chain in dex.trades plus Solana; the venue series covers EVM chains only.

Hypotheses considered

  1. Ethereum's lead is real. Falsified if its total depends on a small number of weeks that are far outside its own distribution.
  2. The spike is one venue, not the chain. Falsified if no single venue's week is large enough, and timed right, to account for the chain's excess.
  3. The spike is a genuine surge in trading. Cannot be settled from weekly aggregates alone. The query that would settle it, the distribution of that venue's trades by pool and transaction in that week, is written in the Data section but was not run here.

Verification independent of the thing being verified

The chain series and the venue series are separate aggregations of the same table, so they are not independent of the table. They are independent of each other's grouping, which is enough to test hypothesis 2: if the chain's excess and one venue's excess are the same size in the same week, the venue explains the chain.

The flag rule

A week is flagged when it exceeds 8 times the entity's median week over the window and $1B. The flagged week is not removed. For the ranking it is replaced with the median, and the result is checked again at a 5× threshold to see whether the conclusion depends on the choice.

Data

  • Chain series: Weekly DEX volume by chain, 13 complete weeks, 2026-06-22 to 2026-09-14, top six chains by median week.
  • Venue series: Weekly DEX volume by venue, same weeks, Balancer only.
  • Both read on 2026-09-24 and committed as JSON below. The current week is excluded because it is partial.
Source queries and the unrun checkSQL
-- Both series come from saved queries maintained by the Dune team; the SQL is
-- reproduced here so the committed JSON can be traced to its source.

-- 1. Weekly volume by chain (Dune query 5959322, https://dune.com/queries/5959322)
WITH weekly_volume AS (
  SELECT DATE_TRUNC('week', block_date) AS "Date", blockchain AS "Chain", SUM(amount_usd) AS "USD Volume"
  FROM dex.trades
  WHERE block_date > CURRENT_DATE - INTERVAL '180' DAY
  GROUP BY 1, 2
  UNION ALL
  SELECT DATE_TRUNC('week', block_date) AS "Date", 'Solana' AS "Chain", SUM(amount_usd) AS "USD Volume"
  FROM dex_solana.trades
  WHERE block_date > CURRENT_DATE - INTERVAL '180' DAY
  GROUP BY 1, 2
)
SELECT "Date", "Chain", "USD Volume" FROM weekly_volume ORDER BY "Date" DESC, "USD Volume" DESC;

-- 2. Weekly volume by venue, EVM chains only (Dune query 4323, https://dune.com/queries/4323)
SELECT project, DATE_TRUNC('week', block_time), SUM(CAST(amount_usd AS DOUBLE)) AS usd_volume
FROM dex."trades" AS t
WHERE block_time > NOW() - INTERVAL '365' day
GROUP BY 1, 2;

-- 3. The check that would settle the cause, not run here (needs query execution credits):
-- distribution of Balancer trades on Ethereum in the week of 2026-08-31 by pool and by
-- transaction, to see whether $206B is a few mispriced swaps or thousands of real ones.
SELECT project_contract_address AS pool, token_pair,
       COUNT(*) AS trades, SUM(amount_usd) AS usd_volume, MAX(amount_usd) AS largest_trade_usd
FROM dex.trades
WHERE blockchain = 'ethereum' AND project = 'balancer'
  AND block_date >= DATE '2026-08-31' AND block_date < DATE '2026-09-07'
GROUP BY 1, 2
ORDER BY usd_volume DESC
LIMIT 20;
Runs on Dune (Trino) · Download query.sql · Open the saved query

Findings

Ethereum's 13 weeks are unremarkable except for two. Eleven of them sit between $5.2B and $10.2B. The week of August 31 records $213.9B, 26 times the chain's median week of $8.2B, and the following week records $48.6B, 6 times the median. Those two weeks are 76% of the chain's summer total.

Two weeks account for $262B of Ethereum's $348B summer DEX volume. Bars are clipped at $40B; the true totals are printed above them.Bar chart of Ethereum weekly DEX volume across 13 points.$0$10B$20B$30B$40BJun 22Jul 6Jul 27Aug 10Aug 31Sep 14
Two weeks account for $262B of Ethereum's $348B summer DEX volume. Bars are clipped at $40B; the true totals are printed above them.Source: Dune query 5959322

The venue series points at one name. Balancer's median week across the same window is $52M. In the week of August 31 it reports $205.9B, about 3,900 times that, and $41.4B the week after, about 790 times. Ethereum's excess over its median in those weeks is $205.7B and $40.5B. The venue's excess and the chain's excess match to within 1% in the first week and about 2% in the second.

Balancer reports under $120M in every week but two. Bars are clipped at $200M; the two clipped weeks are $205.9B and $41.4B.Bar chart of Balancer weekly DEX volume, all EVM chains across 13 points.$0$50M$100M$150M$200MJun 22Jul 6Jul 27Aug 10Aug 31Sep 14
Balancer reports under $120M in every week but two. Bars are clipped at $200M; the two clipped weeks are $205.9B and $41.4B.Source: Dune query 4323

Treat the flagged week as its median and the ranking changes. Ethereum's total falls from $347.6B to $141.8B and its rank from first to third. Base and Robinhood Chain each have a flagged week of their own and fall too, but keep their positions. BNB Chain and Solana have no flagged weeks and move up to first and second.

Ethereum drops from $348B to $142B once its flagged week is replaced by its median; BNB Chain and Solana are unchanged at $267B and $195B.Bar chart of Summed as reported, Flagged weeks replaced by median across 6 points.$0$100B$200B$300B$400BBNB ChainSolanaEthereumBaseRobinhood ChainArbitrum
Ethereum drops from $348B to $142B once its flagged week is replaced by its median; BNB Chain and Solana are unchanged at $267B and $195B.Source: Dune query 5959322
Thirteen-week totals, June 22 to September 14, 2026, summed as reported and with weeks above 8× the chain's median replaced by that median.
ChainSummedRankRobustRankFlagged weeks
BNB Chain$266.7B2$266.7B1none
Solana$195.1B3$195.1B2none
Ethereum$347.6B1$141.8B32026-08-31
Base$171.5B4$111.3B42026-08-10
Robinhood Chain$123.5B5$77.7B52026-08-31
Arbitrum$20.3B6$20.3B6none
Thirteen-week totals, June 22 to September 14, 2026, summed as reported and with weeks above 8× the chain's median replaced by that median.

The conclusion does not depend on the threshold. At 5× instead of 8×, Ethereum's second spike week and one more Base week are also flagged; Ethereum's robust total falls further, to about $101B, and it stays third.

Source: chains-naive-vs-robust.json ·

Every figure above is recomputed from the committed rows by the site at build time.

Limitations

The venue is inferred, not attributed

The venue series is not split by chain, so this analysis cannot show from these two queries that Balancer's $205.9B occurred on Ethereum. The inference rests on size and timing: the venue's excess matches the chain's excess to within a few percent in two consecutive weeks. That is strong, and it is not the same as a join.

Flagged does not mean wrong

A week 3,900 times a venue's median is far more likely to be a pricing or decoding artifact in one pool than a real surge, but weekly aggregates cannot distinguish the two. The transaction-level query that would is in the Data section. It was not run here because reading stored results costs datapoints while executing costs compute credits this site does not spend.

Thirteen weeks is a short window for a median

Medians over 13 points are stable against one or two outliers, which is why they are used, but a chain whose activity genuinely stepped up mid-window would have a median that lags it. None of the six chains shows that pattern in these weeks; a longer window would make the point stronger.

Coverage is Dune's coverage

Both queries count only venues Dune has decoded into its curated tables. A chain with more undecoded volume is undercounted throughout, before and after any correction, and no wash-trading or round-trip filter is applied, so all figures are reported volume rather than monetizable volume.

Reproduce