The Ethereum ecosystem — the most important commercial network in crypto, handling billions of dollars worth of tokens — is getting increasingly centralized.

On Sept. 27, blockchain infrastructure provider Blocknative said it will stop operating as a so-called relay, a key participant in adding blocks of transactions to the Ethereum blockchain. Its exit leaves only four other major relay players to handle most Ethereum blocks and raises concern of potential problems, ranging from the censorship of transactions to stealing of other key operators’ profits.

The relay issue comes as Ethereum already faces concentration concerns in other parts of its operations. The network is run jointly by relays as well as parties called builders, which compile most transactions into blocks, and validators, which order blocks into a blockchain. Both builder and validator functions are dominated by a handful of participants. The relay decline is especially worrisome.

“This doesn’t matter until it matters, everything is fine until it isn’t fine,” said Uri Klarman, chief executive officer of bloXroute Labs, which provides services such as a relay for the Ethereum ecosystem. “And then it’s too late to fix it.”

Relays may reject transactions from specific digital wallets, for example, to comply with government sanctions — or on their own accord. They could also potentially steal other key network operating parties’ profits, Klarman said.

A relay malfunction can also hugely impact the network, and could “result in late or missed blocks on the chain,” said Jim McDonald, chief technology officer at Attestant, a large Ethereum staking provider.

The situation goes against the decentralized ethos of crypto, which was born in the fallout of the Global Financial Crisis as a reaction against powers that be: corporate and government control.

Currently, relays have little financial incentives to act, and many are simply there because they want to support the network or have side businesses that can benefit from offering the function. It costs $500,000 to $1 million a year to operate a relay, Klarman said, and relays don’t get paid. While several ways to compensate relays have been considered in the past year — since Ethereum moved to its new proof-of-stake system of ordering transactions that uses relays, builders and validators — none has taken off.

“It’s been a year, this is unsustainable,” Klarman said. To stop relays from dropping out, builders may need to start sharing revenue with them, he said.

The builder category is highly centralized, also. While a year ago one builder — Flashbots — controlled the market, nowadays four builders, including one run by crypto market maker Wintermute, account for the majority of blocks built, according to Relayscan.io.

“Concentration of builders is the the No. 1 problem Ethereum faces, beyond scaling of course,” said Leland Lee, a vice president at Galaxy Digital and an authority on crypto’s concentration risks. Ethereum programmers are working on potential solutions, whose development could be accelerated if needed, said Ben Edgington, a product owner at Ethereum infrastructure provider Consensys. They are looking at things like automatically detecting censored transactions and punishing the censoring builders, he said.

To add to the problems, a pool called Lido currently controls 32.3% of total validator power. A validator controlling 34% could potentially falsify transactions.

“While it is highly unlikely that Lido will attack Ethereum in any capacity, the potential damage from any attack increases after Lido breaches the 33% threshold,” according to a recent report from researcher Messari. “It is prudent for Ethereum to remove the possibility of such an attack.”

A change to Ethereum’s software that is expected to take place in a few months and limit the entrance of new validators could further help Lido — with its tokens including stEth — grow even bigger, McDonald said. Lido is working to decentralize its operations, so as not to be perceived or act as a single entity, but it’s not there yet.

“There is nothing in-protocol that we are doing or can realistically do to limit Lido’s staking share,” Edgington said. “Beyond social consensus not to invest more in stEth, the only lever we have is the threat of potentially evicting or censoring Lido-controlled validators in future if they are believed to be acting against the protocol’s best interests. However, such a decision would not be solely up to core devs, but would involve all stakeholders running nodes. This can be considered a last-resort option, but ultimately viable.”