What is happening
Support for the Sepolia testnet ends on 30 September 2026. A replacement launched in March 2026, and rather than cutting Sepolia off on the day, the Ethereum client teams agreed to run both networks side by side through a grace period. That overlap is the point: it exists so you can migrate while both are up, instead of finding out the morning after.
The current status of every Ethereum and L2 testnet we track, with the dates and the sources, is on the front page: https://chaindrop.org
Why this one lands harder than Goerli did
Goerli died in 2023 and most teams shrugged, because Sepolia was already the recommended default and the migration path was obvious. This time Sepolia is the default. It is what the tutorials say, what the boilerplates ship with, what the course material teaches, and what a great deal of CI configuration has hardcoded.
Hoodi, launched in March 2025 to replace Holesky, is supported to 2028 and is the right target for validator and staking work. It is not a drop-in replacement for application testing, so one migration does not cover both cases.
The dependency most people have not checked
Base Sepolia, Arbitrum Sepolia and OP Sepolia settle to Sepolia. When we last reviewed this, none of the three had published a replacement plan for their own testnets. That does not mean there is no plan. It means there was nothing public to point at, and that the answer has to come from each L2 rather than from an Ethereum announcement.
If your project touches any of them, ask now rather than in late September. Those three are marked unconfirmed on our status map for this reason. We would rather say we do not know than publish a date nobody has committed to.
What to do first
Search your repositories for sepolia, and not only in the chain config. RPC URLs, chain IDs, explorer links, faucet links, CI environment variables, deployment scripts, hardhat and foundry profiles, subgraph manifests, and any documentation you have written for other people. Search for 11155111 as well, the Sepolia chain ID, because it surfaces places the word itself never appears.
Then split the results in two. Anything with a lifespan past September moves to the replacement network during the overlap. Anything short-lived can stay where it is and die with the testnet, which is a perfectly good decision as long as somebody makes it deliberately.
Pin your testnet targets explicitly instead of relying on whatever a library defaults to. Defaults are how a dead network survives three years inside a codebase.
The wider habit
Testnets expire on published schedules and almost nobody diaries them. The rest of the stack expires too, just less visibly. On 4 August 2026 a worm called ChainDrop reached 444 npm packages in under four hours by riding legitimate release automation, and what carried it to most victims was an ordinary patch-version bump nobody had a reason to look at. The reference, and a lockfile checker that runs entirely in your browser, is here: https://chaindrop.org/chaindrop-npm-worm
Sources for the dates above: https://ethereum.org/developers/docs/networks/ and https://blog.ethereum.org/2025/03/18/hoodi-holesky