The technology worked. The business never did. I will give you the verdict first and defend it after, because here the conclusion is not a matter of taste — it is on DefiLlama’s dashboard. Restaking secured $10.02 billion as of Sept 8, 2026 and generated $99,977 in fees over the prior week. Ordinary liquid staking secured $51.87 billion and generated $27.35 million over the same week. Per dollar secured, plain liquid staking earns roughly 53 times more than restaking. That is not a bad quarter. That is a product whose unit economics were never there.
The flagship is walking away. ether.fi, which launched in 2024 as the purest expression of the idea, will by the end of Q3 2026 have cut the last structural link between its staking tokens and EigenLayer. The money did not vanish, though; it moved. The shape of it — a token you hold that is a receipt for somebody else’s risk, sold to you as free yield — has already reappeared in curated lending vaults, where it is early enough that the losses are still being counted. It is not the mechanism that failed. It is the wrapper, and the wrapper is being rebuilt right now.
One Pot of Money, Two Yields
Trace the lineage before you judge it, because the flaw was in the design, not the execution. Staking is simple: you lock ETH to secure Ethereum and get paid a yield for doing it. Restaking took the same locked ETH and rented its security out to other services — oracles, data availability layers — that would pay for the protection. On paper you earn twice on one pot of money. That is the whole pitch, and it is a good pitch, which is exactly why it aimed at where the ball was going rather than where it was.
Then came the wrapper. A liquid restaking token, an LRT, is a tradable receipt on top of the restaking position. Instead of locking your ETH away, you hold a token you can sell or post as collateral, and weETH was the largest of them. ether.fi’s edge was convenience: deposits were automatically restaked on EigenLayer. The market loved the ease. EigenLayer held $19.7 billion at its peak, and LRTs grew more than 1,000% in the first six weeks of 2024. Read that again — a wrapper on a wrapper, one thousand percent in six weeks — and note that none of that growth was driven by the services buying the security.
Here is the arithmetic that killed it. The services buying security never paid enough to cover both the base staking yield and a premium on top of it. That is the entire second yield, gone. The whole reason to restake was the premium, and the premium was never funded by a customer. It was funded by points, by airdrops, by token emissions, by someone else’s balance sheet. When your core revenue line is a subsidy, you have not built a product. You have built a promotion.
The subsidy stopped, and the downside got a price
Two things happened in 2025, and together they removed what was left of the incentive. First, the subsidizing of deposits wound down through the year. Second, slashing went live in April 2025. Slashing confiscates part of an operator’s staked ETH when it misbehaves — going offline or signing conflicting messages. So restaking suddenly carried a real, priced downside where the risk had been an abstraction, and there was still no extra yield to compensate. The subsidy stopped and the downside got a price. From that moment the structure had to pay for itself, and the ledger says it could not.
The ledger is not subtle. The five largest remaining LRTs — Renzo, Kelp, Swell, Puffer Finance, Bedrock — booked a combined gross profit of $953,350 in Q2 2026, down from $2.18 million three quarters earlier, according to CoinDesk. Puffer raised $23 million and recorded $21,590 for the quarter. Swell recorded $22,370. And on Kelp’s books, EIGEN token rewards appear as $460,600 of gross revenue and $460,600 of cost of revenue — the money arrives and passes straight to depositors, leaving nothing with the protocol. Puffer and Swell book staking rewards the same way. Whatever profit these companies made came from the ordinary staking fees charged underneath the restaking layer. The restaking layer supplied the story; the staking layer supplied the money.
I have seen too many teams build a company on a yield that was never revenue. It always looks like this one: a dashboard that goes up and to the right, a token that appreciates, a treasury that pays salaries for a while, and no customer who would keep paying once the incentive stopped. The team believes the yield will turn into revenue once usage is big enough. Usage was never the constraint. The customer was. ether.fi’s CEO, Mike Silagadze, put the exit plainly to CoinDesk:
“There were no meaningful yield opportunities in restaking and there was some perceived risk from stakers, so we decided it made sense to exit.”
Mike Silagadze, CEO of ether.fi
The exit is cleaner than the entry. In August 2026, ether.fi stripped restaking out of weETH — the version that circulates and is accepted as collateral across DeFi — leaving it a plain liquid staking token. Anyone who still wants restaking must opt into a separate token built on Symbiotic, a rival platform. Protocol documentation put under 1% of assets still restaked as of August, and EigenPod withdrawal credentials are due to be removed by the end of the year. That is not a pivot. That is the deletion of a layer, and it is the right call.
The wrapper, not the mechanism
The temptation is to read restaking’s failure as a failure of restaking. It is not X — it is Y. The mechanism held up fine. EigenLayer itself did not fail: nothing was slashed, no restaking mechanism broke. The money was lost in the wrapper — the tradable receipt on top of the restaking, one more piece of software between the depositor and the asset. That distinction is not academic. It is where the entire loss lived.
On April 18 an attacker exploited Kelp’s cross-chain bridge — the system that moves its token between blockchains — and created 116,500 rsETH in 46 minutes, worth about $293 million, with no ETH backing it. The attacker deposited the tokens into Aave as collateral and borrowed real ether against them. Around $6 billion left Aave in the days that followed, with potential bad debt of $123 million to $230 million. In May, Aave rewrote its collateral listing standards to assess cybersecurity and technical architecture alongside price volatility. Silagadze objects to reading the incident as a failure of leverage: “The cause of the Kelp hack was poor security practices with respect to cross-chain, not related to leverage.” He is right, and his being right is the point — the bridge was the weak joint, the wrapper was the barn. By April, anyone holding one of these tokens was accepting an extra piece of software that could be attacked and getting no extra yield for the exposure. The wrapper had stopped paying for itself and was still open for business.
So where did the money go?
Capital leaving restaking did not leave crypto lending. It changed asset and it changed venue. In 2024 the loop was: stake ETH, restake it, wrap it in an LRT, borrow against that, buy more — stacking exposure to a single asset. By 2026 the same behavior runs through curated vaults, and the ETH is now a dollar. That is the honest read of the migration: the leverage did not go away, the collateral just got a new name, as CoinDesk’s reporting on the category shows.
A curated vault is a lending pool where an outside firm — a curator, not the lending protocol itself — decides which assets the pool accepts and on what terms, in exchange for a share of the fees. Morpho, the largest venue for this, holds around $5.8 billion. Notice the shape before the details: as in restaking, the token a depositor holds is a receipt whose risks are set by someone else and accepted as collateral by a third party. Same structure, new label, and the label is doing the work the math should be doing.
Curated vaults have already produced their own version of the Kelp collapse. On Nov 4, 2025, Stream Finance disclosed roughly $93 million of losses and froze withdrawals. Its xUSD token, a yield-bearing dollar token meant to hold $1, fell 77% in a day. Curators had built vaults on Morpho where depositors supplied real stablecoins against xUSD, and the borrowed stablecoins were used to buy more xUSD, inflating the token far beyond what backed it. Those markets valued xUSD at a fixed $1 rather than its market price, so when the real price fell, the automatic liquidations that should have closed the loans never triggered. Researchers later mapped roughly $285 million of debt exposure across lending platforms. A second dollar token, 65% backed by loans to Stream, fell about 98% and was wound down. Every piece of that is the restaking wrapper again: a receipt, rented risk, a price that is assumed rather than measured.
Even the wrapper’s own operators say the wrapper doesn’t pay
You do not have to take my reading of curated-vault economics, because the people running the wrapper said it themselves and then deleted the evidence. On Sept 26, 2026, a post from Morpho’s official X account said most curator businesses are not self-sustaining from vault fees and that the business lives in private distribution agreements. It was deleted the same day, and Morpho blamed a third-party AI marketing tool. Who pressed send does not matter. The content does. It is a confession that the fees collected at the wrapper layer do not cover the wrapper — precisely the condition on which we just watched restaking fail. The Defiant reported it, and Aave founder Stani Kulechov called it the most bearish take for MORPHO holders.
Two days earlier, on Sept 24, 2026, Morpho CEO Paul Frambot had proposed splitting onchain vaults into “noncustodial” and “discretionary” categories for regulators. Kulechov called the split “self-serving”, and vault builders pushed back that a timelock does not remove the manager. The Defiant carried that exchange too. Look at what the fight is actually about: how to label the wrapper, conducted by the people who collect the wrapper’s fees, while the question of whether the wrapper can pay for itself stays conveniently unaddressed. When the operators start arguing over the taxonomy of the risk instead of the size of the revenue, the numbers have already told you the answer.
The second act
None of this means the people who built restaking are wrong, or that they cannot run a business. ether.fi’s second act is real, and it is worth reading as an income statement rather than a press release. The company now runs a card that lets users spend against their crypto without selling it, a borrowing market on the Ethereum layer-2 network Optimism, and a set of vaults; it describes itself as a crypto neobank. In August it added tokenized stocks, metals and fiat rails. Silagadze puts the neobanking market at roughly $300 billion in annual revenue, about 300 times DeFi’s — which is the most useful sentence in the whole story, because it explains the exit better than any pivot deck. He is not leaving a market because it broke. He is leaving because it was small.
The card’s share of monthly revenue went from 17% in January to 46% in July. Silagadze told CoinDesk that “neobank revenue has fully replaced the revenue lost from restaking and lower ETH price”, and that “we are on track to increase revenue overall run rate this year by about 38%, while staking and restaking revenue has declined by 70%. Diversification of our revenue has been a huge success.” Read the two clauses together and they are perfectly consistent: a fast-growing new line can fully replace a fast-dying old one at the headline while the total still falls. That is not a contradiction. That is a company in transition, and I would rather see the transition than the story.
But here is the part that keeps an engineer honest. DefiLlama’s figures show ether.fi’s gross profit falling 47%, from $18.71 million in Q3 2025 to $9.99 million in Q2 2026. Both can be true at once, because gross revenue is not gross profit and a forward run rate is not a trailing quarter — but ether.fi has not published the basis for the 38%, and until it does, the run-rate is a claim, not a measurement. And the P&L line that should stop you is this: by source in Q2 2026, card fees produced $3.14 million and EigenLayer restaking produced $2.87 million — ahead of core ETH staking and ahead of vault fees, borrowing and management fees combined. On DefiLlama’s accounting, restaking was ether.fi’s second most profitable line at the exact point it decided to leave.
Silagadze disputes one input, and it is worth stating fairly. Cashback — the rewards paid to card users — appears in DefiLlama’s figures at $5.83 million in both revenue and cost of revenue, adding nothing to profit. He says third-party partners used to pay the cashback rewards but no longer do, so current revenue reporting does not include the cashback subsidy grants, while DefiLlama’s adapter still books it on both sides through the most recent quarter. He may be right. I do not have the internal ledger, and I will not pretend I do. What I will say is that when the difference between a good quarter and a bad one is how you book a subsidy, the subsidy is still the business.
What is still unresolved
Draw the boundary between what is settled and what is not, because that is the only honest place to stop. What is settled is that EigenLayer stopped selling restaking as the product. Rebranded as EigenCloud, it now markets verifiable computing — applications proving that work carried out off-chain was done correctly — with restaked collateral as the layer underneath rather than the thing on sale. Its holdings stand at $5.10 billion, down from $22.06 billion in August 2025, a fall of roughly 75%. EigenDA, its data availability service, runs on mainnet at 100 MB/s and remains the largest service by value secured, and Symbiotic, where ether.fi moved its own restaking, has integrated more than 50 networks. The primitives survived. CoinDesk framed the question the way I would have:
“The question was never whether restaking works. It was whether it generates enough revenue to justify building a business on it, and for the protocols that made restaking their entire product, the answer has been no.”
CoinDesk
What is not settled is the verdict the market is still arguing about. A sector that shrank roughly 75% while continuing to secure everything it secured before may have failed. Or it may simply have been four times larger than the work required — a market that ran ahead of its demand and then corrected to it, which is not a failure but a size. I genuinely do not know which, and neither does anyone who claims to know this early. Silagadze did not wait to find out; he moved ether.fi out while the argument was still running, and on the evidence above that was the revenue-first choice, even at the cost of giving up a profitable line.
What should worry you more is the analog. Curated lending vaults are the same wrapper — a receipt whose risks are set by someone else and posted as collateral by a third party, sold as yield — and they are early enough that we are counting the first losses, not the last. Restaking took about two years and a hack to teach this lesson. I would not assume the next wrapper is faster. The technology will probably work. Watch whether the business does.
