Aave V4 at Six Months: Following the Money
Every step in Aave V4’s growth lines up with a cap raise. We followed six months of deposits through its hubs and spokes to see where the money came from and whether it stayed.
1. Introduction
Aave V4 went live on Ethereum on 30 March 2026 with $3.5M supplied. Growth since then came in steps, and nearly every step lines up with a cap raise that LlamaRisk recommended on the forum and a Security Council multisig executed without an on-chain vote.
By 29 September, V4’s seven DAO-run hubs held $1.03B supplied and $366.1M borrowed. The four Ethereum hubs accounted for $741.4M supplied and $291.0M borrowed; Arc held $246.2M and $65.9M, Avalanche $37.7M and $8.7M, and the newly live Coinbase Stocks hub on Base $8.2M and $0.4M. EtherFi Cash is a separately managed OP Mainnet white-label, so it does not belong in those DAO totals.
Aave V4 · first six months
From $3.5M to $1.03B in six months: V4’s growth came in steps
Deposits by hub
Next: Cap round 18 (~$163M of Add Cap, posted 25 Sep) awaits execution · Risk Stewards on V4 (bounded, no-vote cap changes; AIP passed 29 Sep, payloads behind a one-day timelock) · a dedicated RWA hub
View data table
What was deposited
View data table
Weekly flows
View data table
Borrowed ÷ supplied
View data table
Sources: Dune (hub Add/Remove and Draw/Restore events on Ethereum, Avalanche, Arc and Base; principal only, excluding accrued interest; valued at fixed 29 Sep prices so the charts show flows rather than price moves: Dune prices on Ethereum, Chainlink via the Aave API elsewhere). KPI totals from the Aave API, which include accrued interest. The EtherFi Cash white-label on OP Mainnet, a separately managed V4 instance, is not included. Data through 29 Sep 2026.
About a third of the money deposited into V4’s largest spokes came from wallets that were already using V3, and most of them were not migrating so much as opening a second position.
2. Architecture in 60 seconds
2.1 Hubs and spokes
Where V3 put every asset in one pool and controlled risk with per-asset caps, V4 splits the two jobs. A hub holds the liquidity and does the accounting, and a spoke decides which assets you can post as collateral and what they are worth. The spoke draws on its hub through a credit line with two caps. The Add cap is the most that spoke may supply into the hub and the Draw cap is the most it may borrow out.
Spokes on the same hub share its liquidity, so they share its rate. A wstETH position in the Lido spoke and a weETH position in the EtherFi spoke face different risk rules but draw on the same Core liquidity. The spokes differ in what they accept and at what collateral factor, not in what the money earns. A spoke can also hold credit lines on more than one hub. The Maple spoke takes syrupUSDG as collateral on the Global Dollar hub but borrows part of its USDG from Core, and Bluechip, whose collateral sits on Prime, borrows USDC, USDT, frxUSD, USDG and EURC from Core. Ethena Ecosystem, whose sUSDe and USDe collateral sits on Plus, borrows frxUSD, USDC and USDT from Core.
We covered the full mechanism in more depth here.
Aave V4 · first six months
One pool, many doors: where the money actually sits
Asset, spoke, hub
Read the right-hand column first. Core holds 72% of everything on Ethereum, and one spoke, Main, accounts for three quarters of Core. The other eight spokes are specialist doors onto the same liquidity: EtherFi for weETH, Maple for syrupUSDG, Bluechip for blue-chip collateral, Gold for XAUt. Four of them, Maple, USDG Pendle, Bluechip and Ethena Ecosystem, also hold credit lines on Core that let them borrow from it without supplying into it, which is not drawn here.
Source: Dune query 8754686, supply and withdraw events on the nine material Ethereum spokes, valued at recent prices. Assets under $6M are grouped as Other; rsETH is shown separately because the spoke has been frozen since 18 April. PT-USDG matured on 24 September and now sits in Other. Data through 29 Sep 2026, 05:35 UTC.
2.2 Caps and governance
The Add and Draw caps decide how much TVL a spoke can take. Nothing enters a hub beyond the spoke’s Add cap, so every dollar in Figure 1 had to wait for its cap to rise. Over the first six months, LlamaRisk posted 17 cap recommendations that were executed. The same Security Council multisig executed every one without an on-chain AIP. Most went live within two to four days. Round 16, posted on 3 September, took eight, and Round 17 went live on 16 September, the day it was posted. The one change outside that sequence came on 15 July, when the multisig zeroed two Global Dollar credit lines, which were then deactivated on 6 August. An eighteenth recommendation was posted on 25 September but had not yet been executed.
V4 was activated by AIP 462 on 29 March, 433,908 AAVE for and 282,018 against, a 61/39 split. The main objection was that V4 replaces the risk model that kept V3 safe with one that has not been tested. Under the Aave Will Win framework, all V4 revenue goes to the DAO.
3. Growth and flows
3.1 What arrived
V4 on Ethereum added only three new assets after launch: PT-USDG on 24 June (AIP 496), syrupUSDG on 31 July (AIP 508) and PAXG on 9 September (AIP 516). Everything else in the table was there on day one, so most of the growth came from markets that were already open.
Aave V4 · first six months
What arrived, and when · at fixed prices
Supplied by asset, daily, valued at 29 September prices
Three of the twenty assets in the table below were not there on day one. PT-USDG arrived on 25 June, syrupUSDG on 1 August and PAXG on 10 September, all three into the Global Dollar hub. Everything else was listed when V4 opened and grew by filling a reserve that already existed. syrupUSDG is the steepest line on the chart: it went from nothing to $40M in its first month and passed $50M seven weeks after listing.
Day one against today, in native units
Source: Dune queries 8825952, 8825957 and 8826103 (table: 8724568), from hub Add and Remove events. Balances are principal and exclude accrued interest. Cutoff 29 Sep 2026; the last day is partial. Assets outside the twelve largest are grouped as Other in the chart and listed individually in the table.
The cap log shows when governance opened room, why, and whether anyone used it.
Aave V4 · first six months
Eighteen executions, one without a published round
Cap changes by spoke and asset
View data table
Source: Dune 8678614, 8702916, 8702923, 8754632 and 8754698, from UpdateSpokeConfig, Add and Remove events on the Core, Prime, Plus and Global Dollar hubs. Caps in token units. Each raise is observed only until the next cap change on the same spoke and asset, so a later balance is never scored against a stale cap; raises with a window under three days are excluded. Data through 29 Sep 2026, 05:35 UTC.
Every round took the same path. LlamaRisk posted the recommendation, Aave Labs wrote the payload, and the V4 Security Council multisig (0x187A…77e9) executed it through an executor contract.
3.1.1 Making room where capacity had already been consumed
This was the most direct reason to raise a cap. On 9 April (Round 1, executed 11 April), seven reserves were already above 80% of their limits. Main WETH was at 91.7%, while the EtherFi WETH draw cap was 97% used. The demand was already there, and the raise removed the limit on the next deposits and borrows.
The two tightest markets refilled fast. EtherFi weETH, whose Add cap went from 500 to 1,500, filled the new room within two days, and Main WETH (1,500 to 3,500) within four.
Other collateral markets refilled more slowly. LINK used 79% and WBTC 77% of their new room in the week after Round 1, which left borrowers space to add collateral without waiting for the next round.
The same pattern came back on 11 August (Round 13), when LlamaRisk recommended more capacity for a Maple market that was already at its limits. syrupUSDG had filled its original $10M cap, USDG was nearly exhausted, and several Avalanche reserves had crossed 80% utilisation.
The new $20M syrupUSDG cap filled immediately, and the $30M cap set on 24 August was filled before the next execution on 31 August. By the time the cap went to $50M on 31 August, syrupUSDG had hit three ceilings in a row. At the 29 September cut it held $49.6M, or 99.1% of the cap. The same late-August inflow pushed weETH, USDG, WETH, cbBTC and USDC close to their limits elsewhere in the book, and the raises let it through.
3.1.2 Responding to a rate subsidy
The 2% USDC borrow rebate began on 31 August and has remained unchanged since. It applies to USDC borrowing in the Core and Prime hubs. That was enough to move both hubs: by early September USDC utilisation was above 92% in each. Governance responded twice, raising the USDC rate slope on Core and Prime to 5% on Round 16 (11 September), and Round 17, on 16 September, raising Add cap from $18M to $40M and its Draw cap to $36M.
3.1.3 Creating room before a known catalyst
Sometimes governance moved before utilisation did. On 19 May (Round 4, executed 21 May), LlamaRisk raised Main USDG from $3.5M and frxUSD from $4.5M straight to $10M, ahead of incentives due to start that week. The idea was to give the incentive room to run without another round halfway through.
It is the cleanest case in the log of an incentive, a pre-emptive raise and a fill lining up. Main USDG went from a $500k cap at launch to $80M in Round 17, a 160× increase. At the 29 September cut it held $74.8M, or 93.5% of that cap.
Incentives did not produce the same outcome everywhere. Core USDG grew after rewards appeared, while frxUSD largely plateaued after its initial move. The three incentivised markets, Core USDG, Core frxUSD and Global Dollar USDG, were 11% of gross V4 supply on 26 September (Fuller). frxUSD stalled at about $28M against a $50M cap once its rewards settled; USDG kept filling.
3.1.4 Resetting the baseline
The 3 June recommendation (Round 6, executed 4 June) was different again. It raised caps across much of the book, including markets that were barely used, adding roughly $167M of capacity in one transaction.
Main GHO is the clearest example. On 3 June, its Add cap moved from $1.5M to $10M despite little existing use, and it remained largely empty through the following cap changes. The aim was to move V4 beyond launch-sized limits, so that one meaningful position would not consume an entire market’s usable capacity.
3.2 Where it came from
With Aave V3 still running, the natural suspicion is that V4 is cannibalising it, and that most of what arrived is just users moving positions from one version to the other.
In the seven days before their first V4 deposit, 1,089 wallets had withdrawn from Aave V3. Together they deposited $521.3M into V4, or 30.4% of all Ethereum V4 deposits in the dataset.
That looks like migration, but those wallets deposited $521.3M into V4 while their net V3 position fell by only $74.9M. Just 14.4% of what they brought can be matched to a net V3 exit. Most of the capital therefore came from outside the V3 position.
The window changes the headline: 20.7% of V4 deposits at one day, 30.4% at seven days and 43.8% at thirty. A wider window labels more wallets as migrants, but those additional dollars did not, in aggregate, leave V3.
Aave V4 · first six months
The launch money stayed. Later cohorts kept over half.
What V3 lost, and what V4 took in
V4 did not grow by emptying V3. For every dollar that genuinely left V3, between six and seven went into V4. Stretch the window to thirty days and those same wallets are net buyers of V3: $892M out, $959M back in. What looks like migration in a withdrawal count is mostly V3 users doing what they were already doing.
Supplier retention by cohort
The launch cohort is the outlier. Everyone who arrived in the first two days still holds 96.0% of the capital they put in at their peak. Every cohort after that keeps between 52% and 62%, and between 55% and 58% of the wallets. The two grids move together, which is what you would expect and what an earlier read of this data using gross deposits did not show.
View data table
Source: Dune queries 8741606 and 8841134 (wallet base 8702728), supply and withdraw events on the nine material Ethereum spokes. Positions are valued at fixed recent prices, so this measures capital rather than revaluation. A wallet counts as still in if it holds more than 5% of its peak position. Data through 29 Sep 2026 (retention to 05:35 UTC, migration to 08:40 UTC).
Gross V3 withdrawals rise from $98.2M in the one-day window to $892.0M in the thirty-day window, a 9.1× increase. Net exits remain small by comparison: $59.2M at one day, $74.9M at seven days, then a $67.3M net addition back into V3 over thirty days.
Most of these wallets kept their V3 position and opened a V4 one with money from outside Aave.
The flow analysis asks whether money left V3 immediately before arriving in V4. A snapshot of current balances asks a different question. At the 29 September cut, wallets that still hold at least $1,000 on V3, on any chain, account for 28% of what sits in V4’s Ethereum spokes. A further 46% belongs to wallets that have used V3 but have nothing meaningful left there, and 26% to wallets with no V3 history at all. That is meaningful overlap, but not a wholesale evacuation: only two of the fifty largest current Ethereum V3 suppliers have a position in V4.
3.3 Whether it stays
Everything that follows is Ethereum only. Arc, Avalanche, Base and the EtherFi Cash white-label are separate deployments and are covered in section 4. That leaves $741.4M across the four Ethereum hubs.
Grouping suppliers by the month of their first deposit and following each cohort for ninety days gives two different answers depending on what you count.
Capital retention is measured against the highest position a wallet ever held, because a looping wallet re-deposits the same collateral, so its gross deposits can be ten times the capital it ever had at risk. Positions are valued at one fixed price per asset throughout, so a wallet that has not moved a token shows a flat line whatever ETH did.
The wallets that arrived on 30 and 31 March still held 96.0% of their peak capital ninety days later, and 62.7% of them were still in. Every later cohort kept 52% to 62% of capital and 55% to 58% of wallets. The first two days brought money that behaves differently from anything V4 has attracted since.
In June, two addresses account for 41% of everything the cohort withdrew, and in May the three largest account for 34%. April is the only month where the money left broadly, with its three largest at 20% and the rest spread over hundreds of wallets.
Aave V4 · first six months
Who left, and who is holding
Who withdrew, by cohort
April is the only cohort where the money left broadly. Its three largest wallets account for a fifth of what the cohort withdrew, and the rest is spread across hundreds. In June two addresses account for 41% on their own, and in May three account for 34%. The dip in dollar retention after the launch cohort is a small number of large positions closing, not a base of depositors drifting away.
Who is holding it now
For a six-month-old lending market this is unremarkable. The largest wallet holds 6.9% and the top ten hold less than a third. Concentration lives at the spoke level instead: EtherFi has 44 depositors and its largest holds 40% of that spoke, while Main has 2,015 and its largest holds 5.6%.
Sources: Dune queries 8741508, 8702733 and 8702562, from spoke supply and withdraw events on the nine material Ethereum spokes. Positions valued at one fixed price per asset, so price movement does not enter. Wallet-level data cut 29 Sep 2026, 05:35 UTC.
The dip in later cohorts comes from a handful of large exits, which makes concentration the next question.
Across 2,743 wallets holding $726.3M, the largest holds 6.9%, the top five hold 19.6% and the top ten hold 27.5%.
At spoke level the spread is wide. EtherFi has 44 depositors, and its largest holds 39.9% of the $107.4M spoke, or about $42.9M. Main, with $390.9M, has 2,015 depositors, and its largest holds 5.6%. Maple syrupUSDG holds $86.4M across only 69 wallets, with 32.4% in its largest address.
4. The manual era
Every cap change in V4’s first six months took the same path. LlamaRisk posted a recommendation in the activation thread, then a Security Council signer submitted a transaction to the Aave V4 Gov Executor Safe. The Safe called an executor contract, which ran a payload written by Aave Labs, and the payload pushed the new caps through V4's config engine. Only the first round and a few one-off admin changes were batched through MultiSend. The Safe batched the configuration calls through MultiSend and the executor applied them directly to the hubs. Seventeen public rounds went through that path between 11 April and 16 September. An eighteenth recommendation was posted on 25 September but remained unexecuted at the 29 September cut. None required an on-chain AIP vote.
4.1 Risk Stewards
That process is about to change with the Risk Stewards ARFC. The proposal splits each instance’s two configurator domain-admin roles into five granular roles, then gives the Steward bounded and revocable authority over the risk-management and emergency actions it needs. Caps could move by up to 100% every 36 hours, rate curves by up to three percentage points, and collateral factors and liquidation bonuses by up to 0.5 points and reserves could also be frozen or paused. AIP 523 passed on 29 September with 371,400 AAVE for and none against. Its payloads for Ethereum, Avalanche and Base were queued the same day behind a one-day timelock, so at our data cut the stewards held no roles and had no operating history. Round 18, posted on 25 September, had still not been executed by the multisig.
A cap could double every 36 hours without a forum post, so the steps in Figure 1 should flatten out. Delegates including Abel189 asked for public reporting on steward actions, which the proposal does not require. At our data cut the stewards were not yet active on Ethereum: Round 17 still went through the multisig.
5. Partners
V4’s largest flows arrived through markets with a specific reason to exist: a stablecoin incentive, a collateral loop, a white-label lending book, or a rate subsidy. Because each spoke is a separate venue, we can trace each flow to the product that brought it and see what it did once it arrived.
5.1 Paxos
Global Dollar is now the second-largest hub on Ethereum, at $94.6M. It launched as the Paxos Hub on 24 June (AIP 496) with PT-USDG as its only collateral, drawing USDG from Core through the first cross-hub credit line in V4. It added Maple’s syrupUSDG on 31 July (AIP 508) and a PAXG Gold spoke on 9 September (AIP 516).
Paxos funded a Merkl campaign that paid suppliers of USDG on Main an additional 4% on top of the lending rate. That incentive gives suppliers a specific reason to bring USDG to V4; the repeated cap fills described above show that the market used the capacity governance made available.
5.2 Maple
Maple gave USDG a second use. A user could supply syrupUSDG as collateral in the Maple spoke and borrow USDG against it through the Global Dollar hub. The result was a loop between a yield-bearing stablecoin position and USDG borrowing, rather than a one-way supplier incentive.
PT-USDG matured on 24 September, unwinding the first, maturity-specific market in the Global Dollar hub. The most likely next step is not a replacement PT market but a migration of USDG borrowing capacity towards Maple’s syrupUSDG loop. Round 17 already moved in that direction, cutting the USDG Pendle draw cap from $20M to $15M while raising Maple’s USDG line from $5M to $10M. If the Pendle market is removed, a further increase in Maple’s line would let the USDG trade continue against syrupUSDG collateral.
5.3 EtherFi
EtherFi is two separate stories.
On Ethereum, the EtherFi spoke is one of V4’s largest collateral markets. weETH deposits grew alongside WETH borrowing capacity, allowing users to keep an EtherFi yield-bearing position while borrowing against it.
The two books look nothing alike. EtherFi Cash holds about 73,000 of V4's 79,000 addresses with open balances but only 9% of its debt: its 8,889 borrowers carry a mean of $3,773 and a median of $4.63. Strip Cash out and 6,023 depositors account for 91% of V4's debt.
weETH kept growing but stopped refilling every raise the way USDG and syrupUSDG did. Our read is that collateral markets need spare room: a borrower whose health factor drops has to be able to add collateral without waiting for a cap round or unwinding a loop.
On OP Mainnet, EtherFi Cash is a dedicated, isolated V4 white-label instance for the Cash product. It went live around 13 August and held $284M supplied and $34.1M borrowed on 29 September. The ARFC gives EtherFi 80% of instance revenue and the Aave DAO 20%, with EtherFi bringing the assets and running the market while Aave provides the V4 code and licence.
The ARFC committed up to $175M from EtherFi at launch, plus $20M supplied by the Optimism Foundation. The instance had already exceeded that initial capitalization by 10 September. Its $1.0 to 1.2M annual DAO-revenue estimate assumes roughly $500M of instance assets by the end of 2026 and $5 to 6M of annual reserve-factor revenue, of which the DAO receives 20%.
A large partner spoke can fill quickly, remain economically useful and still depend materially on one position.
5.4 Ethena
Ethena incentives rewarded a paired USDe and sUSDe lending position, with stablecoin borrowing required to qualify. That made the Plus hub useful for a leveraged trade. The subsequent cap utilisation shows that the programme attracted both collateral and borrowing demand.
5.5 Avalanche
V4 went live on Avalanche on 14 July through AIP 504, its first deployment beyond Ethereum. It launched with one Core Hub and Main, AVAX Correlated and Forex spokes sharing liquidity across WAVAX, sAVAX, BTC.b, USDC, USDT, WETH.e and EURC. It now holds roughly $37.7M supplied, against $8.7M borrowed: enough to show that the deployment has found initial collateral demand, but still a small and lightly levered book beside Ethereum.
5.6 Base
Base went live on 25 September and its first V4 market was a dedicated Equities Hub with USDC as the sole borrowable asset, while Coinbase’s seven B20 equities are collateral only. That isolated tokenized-equity credit risk from Aave’s other Base liquidity while giving an existing V3 user base and a new cohort of users a new reason to use V4.
LlamaRisk tracked the first weekend. From 8 p.m. ET Friday to 8 p.m. ET Sunday, while US equity markets were closed and Chainlink's 24/5 feeds were paused, 34 addresses borrowed $22k of USDC across 56 transactions. That was 76% of all borrowing by addresses other than the seeder. The median borrow was $100. When the feeds resumed on Sunday night every price landed within 0.4% of Friday's close, and borrowers with more than $50 of debt sat at health factors between 1.27 and 5.34. The market is small, but it is the first V4 book where most of the borrowing happens while the collateral price cannot move.
6. What’s next
Ethereum and Avalanche followed the same pattern: launch with small caps, watch which ones fill, then give those markets more room. Arc started differently.
6.1 Arc
Arc went live on 16 September with USDC, EURC, cirBTC and WETH, and its USDC cap filled within hours. At the 29 September cut it held $246.2M supplied and $65.9M borrowed. USDC accounts for $143.2M of supply and $65.8M of debt, while cirBTC accounts for $85.4M of supplied collateral.
Most of that debt is one position. The $83M of cirBTC that arrived on 24 September belongs largely to a single holder, who borrowed about $49M of USDC against it, roughly three quarters of Arc's debt. Before that, Arc's $143M of USDC sat almost unborrowed.
Cross-chain governance is still a follow-up task. The deployment comes with a $2M-a-year revenue floor for five years, with any shortfall covered by Arc ecosystem participants. For scale, all of V4 earned the DAO about $100k in the last 30 days.
cirBTC is part of that same bet. It launched as one of Arc’s initial reserves, with a 78% collateral factor after the final configuration update. It gives V4 an early test of whether the hub-and-spoke structure can support custody-linked BTC collateral without putting it into the same risk bucket as every other asset.
6.2 An Ethereum RWA hub
An Ethereum RWA hub has already been proposed. The Custodied Collateral Lending ARFC would create an isolated V4 hub and spoke for custody-linked collateral, separating that exposure from the existing Ethereum hubs.
7. Conclusion
Six months in, V4 has shown that governance can meter capacity into markets that want it. It has not yet shown that the capacity pays. DefiLlama puts V4 revenue at about $100k over the last 30 days, roughly $1.5M a year at the current pace, which is less than Arc’s guaranteed floor alone. The next test is whether the new operating model becomes a business. Base is already live with an isolated equities market, while Risk Stewards passed their AIP on 29 September, with the payloads behind a one-day timelock, but have not yet produced an operating history on V4. If those mechanisms turn cap capacity into recurring borrowing and revenue, V4 can become a meaningful business for AAVE.
Aave V4 at Six Months: Following the Money
Every step in Aave V4’s growth lines up with a cap raise. We followed six months of deposits through its hubs and spokes to see where the money came from and whether it stayed.
1. Introduction
Aave V4 went live on Ethereum on 30 March 2026 with $3.5M supplied. Growth since then came in steps, and nearly every step lines up with a cap raise that LlamaRisk recommended on the forum and a Security Council multisig executed without an on-chain vote.
By 29 September, V4’s seven DAO-run hubs held $1.03B supplied and $366.1M borrowed. The four Ethereum hubs accounted for $741.4M supplied and $291.0M borrowed; Arc held $246.2M and $65.9M, Avalanche $37.7M and $8.7M, and the newly live Coinbase Stocks hub on Base $8.2M and $0.4M. EtherFi Cash is a separately managed OP Mainnet white-label, so it does not belong in those DAO totals.
Aave V4 · first six months
From $3.5M to $1.03B in six months: V4’s growth came in steps
Deposits by hub
Next: Cap round 18 (~$163M of Add Cap, posted 25 Sep) awaits execution · Risk Stewards on V4 (bounded, no-vote cap changes; AIP passed 29 Sep, payloads behind a one-day timelock) · a dedicated RWA hub
View data table
What was deposited
View data table
Weekly flows
View data table
Borrowed ÷ supplied
View data table
Sources: Dune (hub Add/Remove and Draw/Restore events on Ethereum, Avalanche, Arc and Base; principal only, excluding accrued interest; valued at fixed 29 Sep prices so the charts show flows rather than price moves: Dune prices on Ethereum, Chainlink via the Aave API elsewhere). KPI totals from the Aave API, which include accrued interest. The EtherFi Cash white-label on OP Mainnet, a separately managed V4 instance, is not included. Data through 29 Sep 2026.
About a third of the money deposited into V4’s largest spokes came from wallets that were already using V3, and most of them were not migrating so much as opening a second position.
2. Architecture in 60 seconds
2.1 Hubs and spokes
Where V3 put every asset in one pool and controlled risk with per-asset caps, V4 splits the two jobs. A hub holds the liquidity and does the accounting, and a spoke decides which assets you can post as collateral and what they are worth. The spoke draws on its hub through a credit line with two caps. The Add cap is the most that spoke may supply into the hub and the Draw cap is the most it may borrow out.
Spokes on the same hub share its liquidity, so they share its rate. A wstETH position in the Lido spoke and a weETH position in the EtherFi spoke face different risk rules but draw on the same Core liquidity. The spokes differ in what they accept and at what collateral factor, not in what the money earns. A spoke can also hold credit lines on more than one hub. The Maple spoke takes syrupUSDG as collateral on the Global Dollar hub but borrows part of its USDG from Core, and Bluechip, whose collateral sits on Prime, borrows USDC, USDT, frxUSD, USDG and EURC from Core. Ethena Ecosystem, whose sUSDe and USDe collateral sits on Plus, borrows frxUSD, USDC and USDT from Core.
We covered the full mechanism in more depth here.
Aave V4 · first six months
One pool, many doors: where the money actually sits
Asset, spoke, hub
Read the right-hand column first. Core holds 72% of everything on Ethereum, and one spoke, Main, accounts for three quarters of Core. The other eight spokes are specialist doors onto the same liquidity: EtherFi for weETH, Maple for syrupUSDG, Bluechip for blue-chip collateral, Gold for XAUt. Four of them, Maple, USDG Pendle, Bluechip and Ethena Ecosystem, also hold credit lines on Core that let them borrow from it without supplying into it, which is not drawn here.
Source: Dune query 8754686, supply and withdraw events on the nine material Ethereum spokes, valued at recent prices. Assets under $6M are grouped as Other; rsETH is shown separately because the spoke has been frozen since 18 April. PT-USDG matured on 24 September and now sits in Other. Data through 29 Sep 2026, 05:35 UTC.
2.2 Caps and governance
The Add and Draw caps decide how much TVL a spoke can take. Nothing enters a hub beyond the spoke’s Add cap, so every dollar in Figure 1 had to wait for its cap to rise. Over the first six months, LlamaRisk posted 17 cap recommendations that were executed. The same Security Council multisig executed every one without an on-chain AIP. Most went live within two to four days. Round 16, posted on 3 September, took eight, and Round 17 went live on 16 September, the day it was posted. The one change outside that sequence came on 15 July, when the multisig zeroed two Global Dollar credit lines, which were then deactivated on 6 August. An eighteenth recommendation was posted on 25 September but had not yet been executed.
V4 was activated by AIP 462 on 29 March, 433,908 AAVE for and 282,018 against, a 61/39 split. The main objection was that V4 replaces the risk model that kept V3 safe with one that has not been tested. Under the Aave Will Win framework, all V4 revenue goes to the DAO.
3. Growth and flows
3.1 What arrived
V4 on Ethereum added only three new assets after launch: PT-USDG on 24 June (AIP 496), syrupUSDG on 31 July (AIP 508) and PAXG on 9 September (AIP 516). Everything else in the table was there on day one, so most of the growth came from markets that were already open.
Aave V4 · first six months
What arrived, and when · at fixed prices
Supplied by asset, daily, valued at 29 September prices
Three of the twenty assets in the table below were not there on day one. PT-USDG arrived on 25 June, syrupUSDG on 1 August and PAXG on 10 September, all three into the Global Dollar hub. Everything else was listed when V4 opened and grew by filling a reserve that already existed. syrupUSDG is the steepest line on the chart: it went from nothing to $40M in its first month and passed $50M seven weeks after listing.
Day one against today, in native units
Source: Dune queries 8825952, 8825957 and 8826103 (table: 8724568), from hub Add and Remove events. Balances are principal and exclude accrued interest. Cutoff 29 Sep 2026; the last day is partial. Assets outside the twelve largest are grouped as Other in the chart and listed individually in the table.
The cap log shows when governance opened room, why, and whether anyone used it.
Aave V4 · first six months
Eighteen executions, one without a published round
Cap changes by spoke and asset
View data table
Source: Dune 8678614, 8702916, 8702923, 8754632 and 8754698, from UpdateSpokeConfig, Add and Remove events on the Core, Prime, Plus and Global Dollar hubs. Caps in token units. Each raise is observed only until the next cap change on the same spoke and asset, so a later balance is never scored against a stale cap; raises with a window under three days are excluded. Data through 29 Sep 2026, 05:35 UTC.
Every round took the same path. LlamaRisk posted the recommendation, Aave Labs wrote the payload, and the V4 Security Council multisig (0x187A…77e9) executed it through an executor contract.
3.1.1 Making room where capacity had already been consumed
This was the most direct reason to raise a cap. On 9 April (Round 1, executed 11 April), seven reserves were already above 80% of their limits. Main WETH was at 91.7%, while the EtherFi WETH draw cap was 97% used. The demand was already there, and the raise removed the limit on the next deposits and borrows.
The two tightest markets refilled fast. EtherFi weETH, whose Add cap went from 500 to 1,500, filled the new room within two days, and Main WETH (1,500 to 3,500) within four.
Other collateral markets refilled more slowly. LINK used 79% and WBTC 77% of their new room in the week after Round 1, which left borrowers space to add collateral without waiting for the next round.
The same pattern came back on 11 August (Round 13), when LlamaRisk recommended more capacity for a Maple market that was already at its limits. syrupUSDG had filled its original $10M cap, USDG was nearly exhausted, and several Avalanche reserves had crossed 80% utilisation.
The new $20M syrupUSDG cap filled immediately, and the $30M cap set on 24 August was filled before the next execution on 31 August. By the time the cap went to $50M on 31 August, syrupUSDG had hit three ceilings in a row. At the 29 September cut it held $49.6M, or 99.1% of the cap. The same late-August inflow pushed weETH, USDG, WETH, cbBTC and USDC close to their limits elsewhere in the book, and the raises let it through.
3.1.2 Responding to a rate subsidy
The 2% USDC borrow rebate began on 31 August and has remained unchanged since. It applies to USDC borrowing in the Core and Prime hubs. That was enough to move both hubs: by early September USDC utilisation was above 92% in each. Governance responded twice, raising the USDC rate slope on Core and Prime to 5% on Round 16 (11 September), and Round 17, on 16 September, raising Add cap from $18M to $40M and its Draw cap to $36M.
3.1.3 Creating room before a known catalyst
Sometimes governance moved before utilisation did. On 19 May (Round 4, executed 21 May), LlamaRisk raised Main USDG from $3.5M and frxUSD from $4.5M straight to $10M, ahead of incentives due to start that week. The idea was to give the incentive room to run without another round halfway through.
It is the cleanest case in the log of an incentive, a pre-emptive raise and a fill lining up. Main USDG went from a $500k cap at launch to $80M in Round 17, a 160× increase. At the 29 September cut it held $74.8M, or 93.5% of that cap.
Incentives did not produce the same outcome everywhere. Core USDG grew after rewards appeared, while frxUSD largely plateaued after its initial move. The three incentivised markets, Core USDG, Core frxUSD and Global Dollar USDG, were 11% of gross V4 supply on 26 September (Fuller). frxUSD stalled at about $28M against a $50M cap once its rewards settled; USDG kept filling.
3.1.4 Resetting the baseline
The 3 June recommendation (Round 6, executed 4 June) was different again. It raised caps across much of the book, including markets that were barely used, adding roughly $167M of capacity in one transaction.
Main GHO is the clearest example. On 3 June, its Add cap moved from $1.5M to $10M despite little existing use, and it remained largely empty through the following cap changes. The aim was to move V4 beyond launch-sized limits, so that one meaningful position would not consume an entire market’s usable capacity.
3.2 Where it came from
With Aave V3 still running, the natural suspicion is that V4 is cannibalising it, and that most of what arrived is just users moving positions from one version to the other.
In the seven days before their first V4 deposit, 1,089 wallets had withdrawn from Aave V3. Together they deposited $521.3M into V4, or 30.4% of all Ethereum V4 deposits in the dataset.
That looks like migration, but those wallets deposited $521.3M into V4 while their net V3 position fell by only $74.9M. Just 14.4% of what they brought can be matched to a net V3 exit. Most of the capital therefore came from outside the V3 position.
The window changes the headline: 20.7% of V4 deposits at one day, 30.4% at seven days and 43.8% at thirty. A wider window labels more wallets as migrants, but those additional dollars did not, in aggregate, leave V3.
Aave V4 · first six months
The launch money stayed. Later cohorts kept over half.
What V3 lost, and what V4 took in
V4 did not grow by emptying V3. For every dollar that genuinely left V3, between six and seven went into V4. Stretch the window to thirty days and those same wallets are net buyers of V3: $892M out, $959M back in. What looks like migration in a withdrawal count is mostly V3 users doing what they were already doing.
Supplier retention by cohort
The launch cohort is the outlier. Everyone who arrived in the first two days still holds 96.0% of the capital they put in at their peak. Every cohort after that keeps between 52% and 62%, and between 55% and 58% of the wallets. The two grids move together, which is what you would expect and what an earlier read of this data using gross deposits did not show.
View data table
Source: Dune queries 8741606 and 8841134 (wallet base 8702728), supply and withdraw events on the nine material Ethereum spokes. Positions are valued at fixed recent prices, so this measures capital rather than revaluation. A wallet counts as still in if it holds more than 5% of its peak position. Data through 29 Sep 2026 (retention to 05:35 UTC, migration to 08:40 UTC).
Gross V3 withdrawals rise from $98.2M in the one-day window to $892.0M in the thirty-day window, a 9.1× increase. Net exits remain small by comparison: $59.2M at one day, $74.9M at seven days, then a $67.3M net addition back into V3 over thirty days.
Most of these wallets kept their V3 position and opened a V4 one with money from outside Aave.
The flow analysis asks whether money left V3 immediately before arriving in V4. A snapshot of current balances asks a different question. At the 29 September cut, wallets that still hold at least $1,000 on V3, on any chain, account for 28% of what sits in V4’s Ethereum spokes. A further 46% belongs to wallets that have used V3 but have nothing meaningful left there, and 26% to wallets with no V3 history at all. That is meaningful overlap, but not a wholesale evacuation: only two of the fifty largest current Ethereum V3 suppliers have a position in V4.
3.3 Whether it stays
Everything that follows is Ethereum only. Arc, Avalanche, Base and the EtherFi Cash white-label are separate deployments and are covered in section 4. That leaves $741.4M across the four Ethereum hubs.
Grouping suppliers by the month of their first deposit and following each cohort for ninety days gives two different answers depending on what you count.
Capital retention is measured against the highest position a wallet ever held, because a looping wallet re-deposits the same collateral, so its gross deposits can be ten times the capital it ever had at risk. Positions are valued at one fixed price per asset throughout, so a wallet that has not moved a token shows a flat line whatever ETH did.
The wallets that arrived on 30 and 31 March still held 96.0% of their peak capital ninety days later, and 62.7% of them were still in. Every later cohort kept 52% to 62% of capital and 55% to 58% of wallets. The first two days brought money that behaves differently from anything V4 has attracted since.
In June, two addresses account for 41% of everything the cohort withdrew, and in May the three largest account for 34%. April is the only month where the money left broadly, with its three largest at 20% and the rest spread over hundreds of wallets.
Aave V4 · first six months
Who left, and who is holding
Who withdrew, by cohort
April is the only cohort where the money left broadly. Its three largest wallets account for a fifth of what the cohort withdrew, and the rest is spread across hundreds. In June two addresses account for 41% on their own, and in May three account for 34%. The dip in dollar retention after the launch cohort is a small number of large positions closing, not a base of depositors drifting away.
Who is holding it now
For a six-month-old lending market this is unremarkable. The largest wallet holds 6.9% and the top ten hold less than a third. Concentration lives at the spoke level instead: EtherFi has 44 depositors and its largest holds 40% of that spoke, while Main has 2,015 and its largest holds 5.6%.
Sources: Dune queries 8741508, 8702733 and 8702562, from spoke supply and withdraw events on the nine material Ethereum spokes. Positions valued at one fixed price per asset, so price movement does not enter. Wallet-level data cut 29 Sep 2026, 05:35 UTC.
The dip in later cohorts comes from a handful of large exits, which makes concentration the next question.
Across 2,743 wallets holding $726.3M, the largest holds 6.9%, the top five hold 19.6% and the top ten hold 27.5%.
At spoke level the spread is wide. EtherFi has 44 depositors, and its largest holds 39.9% of the $107.4M spoke, or about $42.9M. Main, with $390.9M, has 2,015 depositors, and its largest holds 5.6%. Maple syrupUSDG holds $86.4M across only 69 wallets, with 32.4% in its largest address.
4. The manual era
Every cap change in V4’s first six months took the same path. LlamaRisk posted a recommendation in the activation thread, then a Security Council signer submitted a transaction to the Aave V4 Gov Executor Safe. The Safe called an executor contract, which ran a payload written by Aave Labs, and the payload pushed the new caps through V4's config engine. Only the first round and a few one-off admin changes were batched through MultiSend. The Safe batched the configuration calls through MultiSend and the executor applied them directly to the hubs. Seventeen public rounds went through that path between 11 April and 16 September. An eighteenth recommendation was posted on 25 September but remained unexecuted at the 29 September cut. None required an on-chain AIP vote.
4.1 Risk Stewards
That process is about to change with the Risk Stewards ARFC. The proposal splits each instance’s two configurator domain-admin roles into five granular roles, then gives the Steward bounded and revocable authority over the risk-management and emergency actions it needs. Caps could move by up to 100% every 36 hours, rate curves by up to three percentage points, and collateral factors and liquidation bonuses by up to 0.5 points and reserves could also be frozen or paused. AIP 523 passed on 29 September with 371,400 AAVE for and none against. Its payloads for Ethereum, Avalanche and Base were queued the same day behind a one-day timelock, so at our data cut the stewards held no roles and had no operating history. Round 18, posted on 25 September, had still not been executed by the multisig.
A cap could double every 36 hours without a forum post, so the steps in Figure 1 should flatten out. Delegates including Abel189 asked for public reporting on steward actions, which the proposal does not require. At our data cut the stewards were not yet active on Ethereum: Round 17 still went through the multisig.
5. Partners
V4’s largest flows arrived through markets with a specific reason to exist: a stablecoin incentive, a collateral loop, a white-label lending book, or a rate subsidy. Because each spoke is a separate venue, we can trace each flow to the product that brought it and see what it did once it arrived.
5.1 Paxos
Global Dollar is now the second-largest hub on Ethereum, at $94.6M. It launched as the Paxos Hub on 24 June (AIP 496) with PT-USDG as its only collateral, drawing USDG from Core through the first cross-hub credit line in V4. It added Maple’s syrupUSDG on 31 July (AIP 508) and a PAXG Gold spoke on 9 September (AIP 516).
Paxos funded a Merkl campaign that paid suppliers of USDG on Main an additional 4% on top of the lending rate. That incentive gives suppliers a specific reason to bring USDG to V4; the repeated cap fills described above show that the market used the capacity governance made available.
5.2 Maple
Maple gave USDG a second use. A user could supply syrupUSDG as collateral in the Maple spoke and borrow USDG against it through the Global Dollar hub. The result was a loop between a yield-bearing stablecoin position and USDG borrowing, rather than a one-way supplier incentive.
PT-USDG matured on 24 September, unwinding the first, maturity-specific market in the Global Dollar hub. The most likely next step is not a replacement PT market but a migration of USDG borrowing capacity towards Maple’s syrupUSDG loop. Round 17 already moved in that direction, cutting the USDG Pendle draw cap from $20M to $15M while raising Maple’s USDG line from $5M to $10M. If the Pendle market is removed, a further increase in Maple’s line would let the USDG trade continue against syrupUSDG collateral.
5.3 EtherFi
EtherFi is two separate stories.
On Ethereum, the EtherFi spoke is one of V4’s largest collateral markets. weETH deposits grew alongside WETH borrowing capacity, allowing users to keep an EtherFi yield-bearing position while borrowing against it.
The two books look nothing alike. EtherFi Cash holds about 73,000 of V4's 79,000 addresses with open balances but only 9% of its debt: its 8,889 borrowers carry a mean of $3,773 and a median of $4.63. Strip Cash out and 6,023 depositors account for 91% of V4's debt.
weETH kept growing but stopped refilling every raise the way USDG and syrupUSDG did. Our read is that collateral markets need spare room: a borrower whose health factor drops has to be able to add collateral without waiting for a cap round or unwinding a loop.
On OP Mainnet, EtherFi Cash is a dedicated, isolated V4 white-label instance for the Cash product. It went live around 13 August and held $284M supplied and $34.1M borrowed on 29 September. The ARFC gives EtherFi 80% of instance revenue and the Aave DAO 20%, with EtherFi bringing the assets and running the market while Aave provides the V4 code and licence.
The ARFC committed up to $175M from EtherFi at launch, plus $20M supplied by the Optimism Foundation. The instance had already exceeded that initial capitalization by 10 September. Its $1.0 to 1.2M annual DAO-revenue estimate assumes roughly $500M of instance assets by the end of 2026 and $5 to 6M of annual reserve-factor revenue, of which the DAO receives 20%.
A large partner spoke can fill quickly, remain economically useful and still depend materially on one position.
5.4 Ethena
Ethena incentives rewarded a paired USDe and sUSDe lending position, with stablecoin borrowing required to qualify. That made the Plus hub useful for a leveraged trade. The subsequent cap utilisation shows that the programme attracted both collateral and borrowing demand.
5.5 Avalanche
V4 went live on Avalanche on 14 July through AIP 504, its first deployment beyond Ethereum. It launched with one Core Hub and Main, AVAX Correlated and Forex spokes sharing liquidity across WAVAX, sAVAX, BTC.b, USDC, USDT, WETH.e and EURC. It now holds roughly $37.7M supplied, against $8.7M borrowed: enough to show that the deployment has found initial collateral demand, but still a small and lightly levered book beside Ethereum.
5.6 Base
Base went live on 25 September and its first V4 market was a dedicated Equities Hub with USDC as the sole borrowable asset, while Coinbase’s seven B20 equities are collateral only. That isolated tokenized-equity credit risk from Aave’s other Base liquidity while giving an existing V3 user base and a new cohort of users a new reason to use V4.
LlamaRisk tracked the first weekend. From 8 p.m. ET Friday to 8 p.m. ET Sunday, while US equity markets were closed and Chainlink's 24/5 feeds were paused, 34 addresses borrowed $22k of USDC across 56 transactions. That was 76% of all borrowing by addresses other than the seeder. The median borrow was $100. When the feeds resumed on Sunday night every price landed within 0.4% of Friday's close, and borrowers with more than $50 of debt sat at health factors between 1.27 and 5.34. The market is small, but it is the first V4 book where most of the borrowing happens while the collateral price cannot move.
6. What’s next
Ethereum and Avalanche followed the same pattern: launch with small caps, watch which ones fill, then give those markets more room. Arc started differently.
6.1 Arc
Arc went live on 16 September with USDC, EURC, cirBTC and WETH, and its USDC cap filled within hours. At the 29 September cut it held $246.2M supplied and $65.9M borrowed. USDC accounts for $143.2M of supply and $65.8M of debt, while cirBTC accounts for $85.4M of supplied collateral.
Most of that debt is one position. The $83M of cirBTC that arrived on 24 September belongs largely to a single holder, who borrowed about $49M of USDC against it, roughly three quarters of Arc's debt. Before that, Arc's $143M of USDC sat almost unborrowed.
Cross-chain governance is still a follow-up task. The deployment comes with a $2M-a-year revenue floor for five years, with any shortfall covered by Arc ecosystem participants. For scale, all of V4 earned the DAO about $100k in the last 30 days.
cirBTC is part of that same bet. It launched as one of Arc’s initial reserves, with a 78% collateral factor after the final configuration update. It gives V4 an early test of whether the hub-and-spoke structure can support custody-linked BTC collateral without putting it into the same risk bucket as every other asset.
6.2 An Ethereum RWA hub
An Ethereum RWA hub has already been proposed. The Custodied Collateral Lending ARFC would create an isolated V4 hub and spoke for custody-linked collateral, separating that exposure from the existing Ethereum hubs.
7. Conclusion
Six months in, V4 has shown that governance can meter capacity into markets that want it. It has not yet shown that the capacity pays. DefiLlama puts V4 revenue at about $100k over the last 30 days, roughly $1.5M a year at the current pace, which is less than Arc’s guaranteed floor alone. The next test is whether the new operating model becomes a business. Base is already live with an isolated equities market, while Risk Stewards passed their AIP on 29 September, with the payloads behind a one-day timelock, but have not yet produced an operating history on V4. If those mechanisms turn cap capacity into recurring borrowing and revenue, V4 can become a meaningful business for AAVE.
.svg)