By mid-2026, more than 95% of Solana's active stake runs the Jito-Solana validator client. Jito tips account for over 60% of all priority-fee volume on the network, up from roughly 10% in April 2023.
Jito is a private company. None of this is in the Solana protocol. There is no consensus rule that mentions it, no on-chain registry of participants, and no mechanism by which the network could remove it.
We recently took apart Ethereum's version of this problem, where roughly 91% of blocks arrive through off-protocol relays and three builders produce about 92% of them. Ethereum's answer, shipping in Glamsterdam, is to absorb that market into consensus.
Solana is going the other way, and the reason is that it never had the mechanism Ethereum is now trying to formalise.
Start with what Solana does not have
Ethereum has a public mempool. Transactions are broadcast to a peer-to-peer pool where anyone can read them before they execute. That transparency is what makes Ethereum MEV legible, and it is the surface the sandwich attack exploits.
Solana has no protocol-level mempool at all. Transactions are sent directly to the current leader, and the leader schedule is known in advance, so a client knows exactly which validator will produce the next several slots and forwards straight there.
That design removes the public preview window by construction. It also removes the coordination layer that a block market needs, which is why searchers on Solana must either run their own nodes or work with high-staked validators to see current state. Out-of-protocol infrastructure had to fill the gap, and whoever filled it would need adoption across a large fraction of stake to function.
Jito filled it.
What Jito actually is
Not one product. A stack, partly off-chain and partly on-chain.
The Jito-Solana client is an MEV-enabled fork of the standard validator software, adding a RelayerStage, a BlockEngineStage and a BundleStage.
The relayer is an outsourced transaction processing unit that sits in front of the validator over gRPC, filtering and verifying transactions on a separate machine. It holds transactions for 200 milliseconds before forwarding, which is the window in which bundles can be formed against them.
The block engine is the auction. Searchers submit bundles, the engine simulates and scores them, and the highest-paying combination that fits the block's compute budget goes to the validator for its leader slot.
Bundles are up to five transactions executed sequentially and atomically in a single slot. All succeed or none do, which is what makes cross-DEX arbitrage, liquidations and just-in-time liquidity work.
Tips are not priority fees, and conflating them is the most common technical error in this area. A priority fee goes to the leader validator through the protocol. A tip goes to one of eight hardcoded Jito accounts and is the actual auction signal. Competitive bundles run tips of 50 to 70 percent of expected profit. Pay less and you do not land; pay more and you are running a charity.
One precision point, because the headline number is easy to over-read
Running the Jito-Solana client is not the same as Jito building your block. The client lets a validator participate in Jito's bundle auction. It does not mean Jito assembles the block, and critically, a validator running Jito can participate in competing private systems at the same time.
So "95% of stake runs Jito" is a statement about which software is installed and which auction is available, not a claim that one company decides the contents of 95% of Solana blocks. That distinction matters, and it makes the real picture more troubling rather than less, for reasons the next section covers.
March 2024: Jito deleted its own product
Before March 2024, Jito's block engine exposed a public mempool. It gave searchers a 200-millisecond window to preview every transaction heading to the leader. That is the same visibility Ethereum's mempool provides, rebuilt off-protocol on a chain that deliberately lacks one.
It was used, systematically, for sandwich attacks.
Jito's engineers tried to filter the exploitative transactions out and could not keep pace with searchers adapting. So in March 2024, Jito shut the mempool down voluntarily and absorbed a significant revenue loss to do it.
By the standards usually applied to infrastructure operators, that is the correct decision, taken at real cost, without being forced.
It made the problem less visible rather than smaller
The demand did not go anywhere. Private mempools appeared to serve it, run by operators with no transparency obligations and no public accountability, benefiting whoever had exclusive access.
The scale is documented. DeezNode's sandwich bot alone accounts for nearly half of all sandwich attacks on Solana, executing 1.55 million attacks over a thirty-day window for $13.43 million in profit. Average take per attack is about $8.67, so the entire business is volume and automation rather than large individual hits. The disproportionate victims are memecoin traders, who set high slippage tolerances on illiquid assets because they are optimising for speed.
Private mempool operators then went after stake directly. Multiple validator operators have reported receiving detailed profit-sharing offers to join private mempools. A DeezNode validator accumulated $168.5 million in delegated stake within weeks, with nearly 20% of it sourced from Marinade's liquid staking pool.
Read that last sentence twice. Ordinary users deposited SOL into a liquid staking protocol, and a fifth of the resulting delegation to that validator financed a sandwich operation running against other ordinary users. MEV extraction did not just persist after the mempool shutdown. It bought validator stake with the proceeds, which is how extraction converts into consensus weight.
Enforcement is a private company's discretion
There is no protocol rule against sandwiching on Solana. What exists instead:
- Jito has banned validators from its ecosystem for sandwich activity, most recently a further 15 after evidence surfaced that validators were still front-running traders.
- The Solana Foundation removed more than 30 validators from its delegation programme in 2024 for the same reason.
Both are real and both had effect. Neither is a protocol mechanism. A private company and a foundation are policing a public network by revoking commercial and delegation relationships, which works exactly as far as their leverage extends and stops at validators who do not need either.
That is the actual governance model for ordering on Solana today.
BAM: the answer is trusted hardware
Jito's structural response is the Block Assembly Marketplace, announced July 2025 and live on mainnet September 25, 2025.
BAM introduces a network of off-chain block-builder nodes that run inside Trusted Execution Environments. Transactions submitted to a BAM node are encrypted and remain invisible even to the node operator until execution. That is an encrypted mempool enforced by hardware rather than by policy, and it removes the preview window that sandwiching depends on.
It also ships a plugin framework enabling application-controlled execution, which lets a protocol define its own ordering rules at the block-building layer: slippage protection running at sequencing time, programmable order types, explicit guards against toxic MEV. The application specifies how its own transactions may be ordered, rather than accepting whatever the auction produces.
Adoption is being bought. Jito ran an Early Adopter Subsidy for validators, scheduled to phase out over twelve weeks during 2026, after which BAM has to stand on its own economics.
Note what the trust assumption became. The 2024 answer to sandwiching was a company deleting a product. The 2026 answer is a company running block builders inside secure enclaves and asking you to trust the hardware manufacturer's attestation. That is a real improvement in user outcomes and it is not a reduction in the number of parties you are trusting.
Two chains, opposite directions
Set the two designs beside each other.
Ethereum built a block market outside the protocol, watched relays become unpaid trusted infrastructure and builders concentrate, and is now pulling that market into consensus. EIP-7732 makes the builder's payment obligation a staked, protocol-enforced commitment. The trusted intermediary is being deleted by making it unnecessary.
Solana never had a mempool to formalise. Its block market grew entirely outside the protocol, reached 95%+ of stake, discovered that transparency enabled predation, and responded by removing transparency and then re-adding it inside trusted hardware. The intermediary is not being deleted. It is being made more sophisticated and harder to inspect.
Neither approach has solved extraction. Ethereum's ePBS moves value from proposers to builders without reducing it. Solana's BAM reduces one specific attack by making order flow invisible, which also makes independent measurement of what happens inside a block considerably more difficult.
Securing the unseen
Are your transactions going through Jito's auction, a private mempool, or straight to the leader? If a private mempool, who operates it and what is their business model? Does the protocol you are trading on define its own ordering constraints, or does it accept whatever sequencing it is handed? And if your SOL sits in a liquid staking pool, do you know which validators your stake is delegated to and what they run?
That last one is not hypothetical. Liquid staking already routed retail deposits into a validator running the largest sandwich operation on the network.
Learn more at 0xhades.io/research