Guide · Markets
Ethereum Network Upgrades Explained: Testnets, Hard Forks and Why Glamsterdam's Hoodi Date Was Pulled Four Days Out
Ethereum changes its rules through scheduled hard forks that every node must adopt at the same moment, after rehearsals on test networks. Here is how an upgrade moves from proposals to Sepolia, Hoodi and mainnet, why testnets exist, and why developers pulled Glamsterdam's October 26, 2026 Hoodi activation four days before it was due.
Ethereum has no company that can push a software update to its users. Its rules live in the software run by thousands of independent nodes, written by several separate client teams, and a change to those rules only works if every node switches at the same moment. That coordinated switch is a hard fork: from a given block, or a given time, nodes running the new software follow the new rules, and nodes that did not upgrade are left on a chain nobody else uses. Each upgrade bundles a set of Ethereum Improvement Proposals, or EIPs, numbered technical documents that describe one change each, and carries a city-and-star nickname that joins its execution-layer and consensus-layer halves: Pectra in May 2025, for example, and Glamsterdam, the upgrade scheduled for late 2026.
Because an error on the main network could freeze funds or split the chain, every upgrade is rehearsed. Developers first run short-lived development networks, then activate the fork on public test networks, or testnets, where the tokens have no value but the software, the validators and the applications behave like the real thing. In 2026 the two main public testnets are Sepolia, the long-running network used mostly by application developers, and Hoodi, launched in March 2025 as the closest copy of mainnet's validator set. Only when the fork has run cleanly on both do the client teams, on their fortnightly All Core Developers calls, pick a mainnet date. A testnet that misbehaves is the system working: in February 2025 Pectra's activation on the Holesky testnet failed to finalise because three execution clients had the wrong deposit-contract address configured, the network took about two weeks to finalise again, and the episode led to the creation of Hoodi. Mainnet was unaffected and Pectra went live in May.
Here is a real example. Glamsterdam completed its activation on Sepolia on October 6, 2026, and was scheduled to activate on Hoodi on October 26 at 17:42:48 UTC. On October 13 the security firm ChainSecurity published an analysis of two of its proposals, EIP-8037 and EIP-8038, which change how Ethereum charges gas for creating and accessing stored data. Under the new pricing, creating a storage slot could cost up to 110,020 gas, against 22,100 under the old rules, and ChainSecurity showed that some existing contracts could behave differently: a caller could supply just enough gas for an operation to fail while leaving enough for the contract to record the failure instead of reverting. On the All Core Developers call of Thursday, October 22, four days before the date, developers removed the Hoodi activation, together with a planned rise in Hoodi's gas limit to 200 million, and set no new date. ChainSecurity said it had not shown that a major live protocol was exploitable, and no existing contract was reported to be under attack.
For holders the practical consequences were small and specific. Nobody had to move ETH, because mainnet had no Glamsterdam date yet. Node and validator operators had to install client releases that dropped the October 26 activation, because a node still running the old configuration would have forked itself onto an abandoned Hoodi chain. And the timetable moved: a revised Hoodi date was expected only after Devcon in early November, which pushed any mainnet activation later. Ether, which had fallen from about $2,726 on October 5 to about $2,350 by October 25, did not need the delay to explain its weakness, but it removed one of the catalysts some traders had been pointing to.
Three habits help when an upgrade is in the news. Read the order of events, not the headline: a testnet postponement before mainnet is scheduled is a delay, while a bug discovered after mainnet activation is an incident. Check which layer is affected, since execution-layer changes such as gas pricing can break applications, while consensus-layer changes mostly concern validators. And treat upgrade dates as estimates until mainnet is fixed in client software, because markets that trade the date in advance, as they did with the Merge in 2022 and with Pectra, are trading a schedule that the developers themselves have said can move.