Priority fee demand is no longer one market
Ethereum’s priority fee economy is not a single demand curve. In a report by Greenfield Capital, authors Christoph Rosenmayr and Jascha Samadi describe it as a set of separate order-flow submarkets, each driven by different incentives and competitive structures.
The paper looks at how Ethereum block building has changed since 2024. It points to shifts in order-flow distribution, more monetization at the wallet layer, the rise of private transaction routing, and changing builder competition as forces that have altered how value is created and captured across the block production stack.
This study is framed as the first part of a broader research series on Ethereum block building. Part one focuses on the demand side: who creates valuable order flow, who pays priority fees, and how those flows shape competition in block construction. A second part, according to the report, will turn to the supply side and examine how builders compete for those flows and convert them into builder surplus.
Greenfield says the analysis is based entirely on observable onchain data and that all datasets and charts are available through a Dune dashboard that will continue to be maintained. The authors also note that some factors that matter to builder competitiveness, such as latency differences and relay strategy, are outside the scope of this piece.
Why priority fees matter in Ethereum block building
Since the Merge in September 2022, MEV-Boost — described in the report as the off-protocol implementation of proposer-builder separation, or PBS — has become the main mechanism through which Ethereum validators monetize the execution layer. Validators can auction off ordering rights inside a block to specialized block builders, creating a market for transaction ordering.
The paper stresses that consensus-layer rewards are funded by newly issued ETH, while the EIP-1559 base fee is burned. That leaves execution-layer rewards from builder auctions as the only non-dilutive validator revenue stream directly linked to demand for blockspace.
That structure matters well beyond builder rankings. Greenfield argues that the market structure of block building affects the long-run sustainability of validator revenues, the efficiency with which blockspace is allocated among users, and the execution quality users experience when their transactions are routed and ordered.
How the report defines order flow and priority fees
The report defines order flow as all transactions submitted for inclusion onchain. Not every transaction is equally valuable. Many consume gas and pay normal fees but generate little additional economic value.
Priority fees, in this framework, are payments from order-flow originators to block builders in exchange for access, ordering, and inclusion. These payments are separate from the EIP-1559 base fee. They can appear as native priority fees or as direct builder payments through coinbase.transfer. The study combines both and treats them as one category throughout the analysis.
Greenfield argues that these payments exist because many profitable opportunities on Ethereum are competitive, time-sensitive, and highly dependent on exact ordering. In a CEX-DEX arbitrage trade, for example, the first transaction to hit an AMM pool captures the profit. In a memecoin launch, a retail trader may pay up to improve the odds of being included before others.
The broad picture: Ethereum moved from Telegram bot dominance to MEV dominance
The study first groups the market by order-flow originator addresses and finds that demand for priority access is highly variable. It can change with market conditions, periods of volatility, and isolated extreme events.
One of the clearest examples came in the week of March 9, which the report describes as the highest week for aggregate priority fees and the highest single-transaction priority fee in the past several years. That spike followed a high-slippage trade on the AAVE frontend in which a user lost $50 million, leading to a $34 million backrun priority fee surge. The paper says this single event shows up repeatedly across several metrics in the dataset.
Over a longer period, the composition of fee spend changed sharply. What was once a market dominated by two entities — jaredfromsubway and the Banana Gun Telegram bot — is now less concentrated. Greenfield also says total priority fee spend has fallen significantly since early 2024, in both dollar and ETH terms, as slower market conditions reduced the number of profitable opportunities, including a decline in retail speculation on Ethereum mainnet.
Using Flashbots labels available on Dune, the report sorts spenders into broad categories such as MEV, Telegram trading bots and terminals, DEX frontends, wallet trades, and smart-contract interactions. The authors caution that the labels are incomplete, especially because MEV teams frequently change strategy contracts, so the measured contribution of MEV is likely a slight underestimate.
Even with that caveat, the trend is clear. Retail-facing Telegram bots and terminals dominated priority fee spending in early 2024. That is no longer the case. MEV-related flow now leads, accounting for roughly 66% to 80% of weekly priority fee spend. The shift was driven mainly by lower spending from Telegram bots as token-sniping opportunities on Ethereum dried up.
At the same time, the report notes that spending from Telegram trading bots and terminals has picked up strongly again in recent weeks.
Retail flow shifted from DEX frontends toward wallet trades
Within retail-oriented flow, Greenfield finds a separate structural move: spending has shifted away from DEX frontends and toward wallet trades, mainly MetaMask Swaps. That share has grown steadily over multiple years.
The report also singles out a “smart contracts” category. This covers priority fees paid when users interact with major token contracts such as USDC and USDT, including simple transfers and approvals. Nominal spending in that category stayed relatively stable through the year at about $100,000 to $200,000 a week. As other categories fell in nominal terms, its share rose.
The paper says these transactions generally are not competing for priority state access, so there is little reason to pay extra. But because those contracts are used so heavily, millions of transactions paying slightly more than necessary add up. As a result, the category accounts for 24% of total priority fee spend, even though average priority fees there are 99% lower than in other categories.
The top 100 spenders over the past 45 days
Looking at May 2026 and the trailing 45-day period, Greenfield sees a steep power-law distribution. Two MEV-linked addresses — jaredfromsubway and 0xbdb3 — account for most of the spending, with more than $2.4 million in combined priority fees over the previous 45 days. After the top 15 spenders, the chart drops off sharply.
In that window, MEV strategies accounted for 69% of total priority fee spend, down from around 80% at the start of the year.
The highest-ranked non-MEV spenders were Telegram bots, led by GMGN Terminal, Maestro, and Banana Gun. GMGN stood out in particular, ranking third and posting nominal fee spend similar to jaredfromsubway and Wintermute. Major token contracts including USDC and USDT followed, along with MetaMask wallet trades.
Frontends, routers, and meta-aggregators
The report then breaks fee share down by frontend, meta-aggregator, router, and MEV labels. The categories overlap. MetaMask, for example, appears both as a frontend and as a trading router.
Among frontends, the paper shows a decline in the large Telegram bots Banana Gun and Maestro and a more recent rise in MetaMask trades as one of the biggest priority-fee-paying frontends. More recently, all Telegram bots have seen a rebound, but GMGN Terminal stands out because of its expansion to Ethereum mainnet. Greenfield says GMGN now leads both the frontend and router label categories, accounting for 41% and 37% of total priority fee spend in those buckets.
By nominal fee outlay, the report says GMGN gained share mainly by bringing new users from its existing product base onto Ethereum mainnet rather than by taking spend directly from others. One example offered in the paper: MetaMask’s nominal priority fee spend stayed flat while its market share dropped sharply.
In the meta-aggregator category, solver-based systems including 1inch Fusion, CoW Swap, and Uniswap X appear effective at keeping priority fee spending low. The report attributes that to the role solvers play in handling execution complexity for users, along with batch execution in systems like CoW Swap.
Among routers, MetaMask Swap Router had led since mid-2025 but was recently overtaken by Maestro Router. Uniswap, 0x, Aave, 1inch, and Banana Gun Router follow with relatively similar shares.
For MEV-labeled contracts, the authors again warn that the labels are incomplete. Even so, they say Wintermute has re-emerged as the dominant spender among labeled MEV contracts, while jaredfromsubway has also posted an eye-catching comeback in recent weeks.
MetaMask and GMGN users are paying much more per trade
One of the clearest retail-side findings in the report is the change in average fee paid per trade. MetaMask wallet trades have not grown dramatically since early 2024, but the average priority fee paid by those users rose from about $0.15 per transaction in early 2024 to $1.2, based on what the report describes as manually selected MetaMask flow.
GMGN users are paying even more. The paper says they spend more than $2.1 on average per transaction, a level close to Wintermute’s roughly $3 average per transaction. The difference, of course, is that Wintermute’s trades are generally tied to profit capture, while retail users are not operating under the same economics.
That raises the question the report asks directly: why did MetaMask users become willing to spend so much more on priority fees after 2024?
Order flow auctions and fee rebates may distort what onchain data shows
Greenfield says one important source of distortion is difficult to capture directly from onchain data: the growing use of priority fee rebates in order flow auctions, or OFAs.
In addition to selling front-running opportunities to the highest-bidding searcher and returning that bid to the order-flow provider, OFAs have increasingly offered a fixed rebate on all priority fees. Under that setup, transactions that enter through an OFA RPC are shared with eligible builders, and the winning builder is often the one returning a fixed share of the fee — usually 90%, according to the report.
The refund logic and business model vary across OFAs. The report says those rebates are usually not sent back in the same block but paid out periodically, making it impossible to attribute refunds to specific transactions using onchain data alone.
That matters because observed priority fees may overstate what some users or applications are truly willing to pay for priority access. A meaningful share may ultimately be returned upstream. Even so, the authors say the overall conclusions are unlikely to change in a material way. In MEV, where bids are calibrated closely to beat rivals in builder-level bundle auctions, receiving a 90% rebate is economically similar to simply bidding 90% less for the same opportunity.
The report argues that consumer applications are especially well positioned to monetize OFA rebates because they control RPC selection and handle relatively uninformed order flow, where fees are often set loosely and overpayment is common. Greenfield says this may explain the elevated average fee levels seen in wallet-originated flow such as MetaMask.
The paper goes a step further and says those rebates are often not passed through to MetaMask users. That could create a conflict between minimizing user costs and maximizing wallet revenue. Wallets encourage usage, often recommend default priority fee settings, and many users accept those defaults without adjustment.
The study also notes recent consolidation at the OFA layer, including Blink’s acquisition of Merkle and the acquisition of MEV-Blocker from CoW DAO by Consensys/SMG, which operates SERVO MEV Protection. These relationships, the report says, increasingly shape flow routing and block outcomes, even if large parts of their economics remain hard to observe onchain.
Because of that attribution problem, Greenfield says measured priority fees for consumer applications should often be treated as an upper bound on what users are truly willing to pay for inclusion and priority state access. The firm says it plans to examine this further in a future study focused on consumer order flow.
MEV is now the main source of priority fee spending
Across Ethereum, MEV-related activity drives more than 66% to 80% of weekly priority fee spend, according to the report. To unpack that market, Greenfield splits MEV into three common categories on Ethereum: statistical arbitrage, sandwich MEV, and atomic arbitrage.
The classification for statistical arbitrage reuses Greenfield’s earlier work, while the sandwich and atomic arbitrage categories rely on existing classifications from Dune user @hildobby.
The authors note that competition is central to MEV. Searchers often have to pay a large share of extractable profit as priority fees in order to win. Estimating net profit precisely is difficult and error-prone. In previous research, Greenfield used Binance marks to estimate margins and assumed an average margin of about 10% for statistical arbitrage, implying roughly 90% of gross extractable value goes to priority fees, while the largest stat-arb opportunities can carry margins of up to 30%.
Even if priority fees do not map one-to-one to profit, Greenfield says they remain a useful and relatively robust proxy for understanding market structure.
The three MEV markets behave differently
Statistical arbitrage is the largest of the three by both DEX volume and fee outlay. It accounts for 70% to 85% of MEV-related DEX volume and typically around 50% of MEV priority fees.
Atomic arbitrage, by contrast, shows sharper spikes and has trended higher since early 2025. In recent weeks it has often contributed more than 25% of total MEV priority fees. Greenfield suggests that Ethereum’s gas limit increase may have improved the competitiveness of atomic arbitrage, which often relies on multi-hop trades, and says the increase in November 2025 coincides with a higher DEX volume share linked to this strategy.
Sandwich-related fees have been in a long decline from early 2024, when sandwiching represented more than 60% of MEV-related priority fee spend. The report ties that decline to a larger share of retail transactions bypassing the public mempool as private RPCs and dApp-level protections matured. It points to one data point in particular: 92% of DEX volume in recent months was routed through private mempools.
Still, sandwich activity has picked up again. In recent weeks, 35% to 51% of MEV priority fee spend has once again come from that strategy.
Fee intensity differs across the three categories. Statistical arbitrage has the lowest fee intensity, both per trade and as a share of DEX volume, because opportunities are frequent but each one tends to carry lower extractable value. Atomic arbitrage is much denser. The report says an average of 0.4% of atomic-arbitrage-related DEX volume is spent on priority fees.
How much of Uniswap volume is tied to MEV
Greenfield says about 38% of Uniswap volume currently comes from trades related to statistical arbitrage. That figure was as low as around 5% in early 2026 and above 45% in early 2025.
The report sees a strong correlation between the share of Uniswap volume linked to statistical arbitrage and Deribit’s BTC DVOL index. It says the relationship is even stronger when looking at ETH realized volatility over 12-second block intervals. The logic is simple: stat-arb opportunities widen when the same asset moves more sharply between centralized exchanges and AMMs.
Across the three identified MEV strategies, the paper estimates that roughly 49% of total Uniswap volume currently comes from MEV-related trades, down from as high as 56% in early 2024.
Statistical arbitrage: Wintermute remains powerful, but a challenger emerged
In statistical arbitrage, Greenfield compares participants by DEX volume, total priority fees paid, and relative fee share. Historically, the report says, the market was led by Wintermute and SCP/beaverbuild. From early 2024 through September 2024, the pair controlled about half the market between them, with the rest spread across a long tail.
After September 2024, a searcher previously labeled “Kayle” in Greenfield’s earlier work quickly gained share. But by April 2025, Wintermute and SCP had regained and even expanded their position, pushing other players out and controlling more than 75% of the market.
Later, as beaverbuild gradually transitioned to BuilderNet — a point the report says will matter again later — SCP’s searcher began losing share from December 2025 onward. In the first three months of 2026, Wintermute briefly held around 70%, with SCP at roughly 5% and 0x7cDa58 behind them. Greenfield describes that period as the most concentrated phase in the stat-arb market.
In recent months, 0xbdb3ba has emerged as a serious challenger. By priority fee spend, its market share is now broadly similar to Wintermute’s, and the two together account for 83% of total fee spend in the statistical arbitrage category. The report highlights one unusual part of the 0xbdb3ba strategy: bundling all stat-arb swaps into a single transaction, in some cases with more than 50 single-leg swaps.
The team behind 0xbdb3ba remains unknown, according to the study. Manual inspection suggests it shares Wintermute’s preference for routing to Titan, but the authors say that only indicates the contract does not appear tied to another known builder or vertically integrated setup.
Sandwiching: jaredfromsubway still dominates
The sandwich market shows a different structure. Even though sandwich-related fees have fallen from earlier highs, jaredfromsubway remains the dominant force. The report says its relative share of sandwich priority fee spending has remained in a 70% to 85% range for years.
Greenfield adds an important note: the data shown here was collected before the theft from jaredfromsubway’s main contract, an incident that cost more than $15 million. The report says it remains unclear how that event will affect sandwich competition in the months ahead.
If jaredfromsubway is removed from the chart, the remaining market looks much less stable. Participants rise and fall quickly, far faster than the long tail in statistical arbitrage once Wintermute and SCP are filtered out. In other words, the top of sandwiching is steady, but the layer beneath it is not.
The report identifies 0x01FD as the new second-place participant, paying about 20% of total sandwich MEV priority fees and holding that position for several weeks. Greenfield calls it the most credible challenger jaredfromsubway has faced in a long time.
As for the sharp rise in sandwich-related fees over the past three weeks — 45% to 60% of MEV fee spend — the authors say the exact cause is hard to know. Their working explanation is that stronger competition between jaredfromsubway and 0x01FD may have triggered a fee war, reducing margins and driving up fee outlays. The paper says these two are the only real drivers behind the increase.
That view is supported by higher fee intensity. On a priority-fee-per-dollar-of-DEX-sandwich-volume basis, jaredfromsubway is spending roughly three times more than it was four weeks earlier, which the report suggests may be a reaction to pressure from 0x01FD.
The authors add that 0x01FD appears likely to be run by the same team behind BobTheBuilder.
Atomic arbitrage: smaller, spikier, and more fragmented
Atomic arbitrage looks different again. The report says the market is smaller overall, but opportunities are far more variable, which produces bigger spikes.
That fits the mechanics. Large atomic arbitrage opportunities often start with an individual user making a high-price-impact swap in an AMM pool, intentionally or by mistake. Atomic arbitrage then realigns prices across AMMs because the distortion is onchain rather than between centralized exchanges and AMMs.
The report returns here to the AAVE frontend incident mentioned earlier, where a $50 million swap turned into just $35,000. Greenfield says that event generated the highest atomic arbitrage priority fees observed since early 2024.
These episodes usually come from user input errors and are infrequent. That makes atomic arbitrage fundamentally different from statistical arbitrage, which appears whenever CEX and DEX prices diverge over a 12-second block interval, and from sandwiching, where some users continue to send trades through the public mempool without protection. The result is a more volatile fee profile.
Atomic arbitrage is also the only one of the three MEV markets that remains highly fragmented, with no single participant consistently dominating.
Greenfield suggests two reasons. First, the category is smaller in aggregate. Second, pure atomic arbitrage offers less structural scale advantage because tools like flash loans lower capital barriers. That stands in contrast to statistical arbitrage, where latency, inventory, and CEX fee tiers matter and favor firms like Wintermute, and to sandwiching, where inventory-heavy strategies helped jaredfromsubway maintain long-term dominance.
Part one conclusion
Greenfield’s conclusion is that Ethereum priority fee demand is made up of multiple order-flow businesses rather than one market. Since 2024, the composition of valuable order flow has changed materially at both the category and participant level, with each segment evolving under its own competitive dynamics.
The distinction is especially clear inside MEV. Statistical arbitrage, sandwich attacks, and atomic arbitrage differ meaningfully in concentration, opportunity frequency, and fee intensity.
The first part of the report stays on the demand side, asking who creates valuable order flow and why they pay. The second part, the authors say, will examine how builders compete to capture that flow, how builder surplus is generated, and how increasingly specialized order-flow providers are reshaping block production.
For feedback, the report lists a Telegram contact at https://telegram.me/@thirdeye33 and an email address at [email protected]. It also says all charts and underlying queries are available on the Dune dashboard and will continue to be maintained.
Disclosure
Greenfield says it has investments in Gattaca, the parent company of Titan, as well as in CoW Protocol and NuConstruct.
The firm says the analysis is based entirely on public onchain data obtained through Dune Analytics and reflects independent research views rather than portfolio marketing or investment advice.
It also states that the analysis may mention projects, protocols, builders, searchers, and market participants in which Greenfield may have direct or indirect interests. Any such mention, the report says, is for research and informational purposes only and should not be interpreted as an endorsement, recommendation, or statement about future performance.