The Merge and the upgrades since
Ethereum changed its consensus mechanism while running, without stopping and without altering a single balance. Then it kept changing.
Ethereum launched as a proof-of-work network. The intention to move to proof of stake was stated early and took years longer than planned — a delay that became a running joke before it became a delivery.
The Merge
The existing chain, carrying every balance and contract, was joined to a separately-running proof-of-stake consensus layer that had been operating in parallel since December 2020. Mining stopped at a predetermined point and validators took over block production. No balance changed, no contract was redeployed, and the chain did not pause.
Energy consumption fell by an amount generally stated as over 99%, because the mining step — the entire purpose of which was to consume electricity — ceased to exist. See the energy question.
Why it was hard
The difficulty was not the idea but the constraint: replace the mechanism by which a live network agrees on reality, on a system holding a very large amount of other people's money, with no scheduled downtime and no ability to roll back. The approach — build the new consensus layer separately, run it for nearly two years, then switch — is the reason it worked.
What followed
- Withdrawals enabled (April 2023). Staked ether could be removed for the first time. Deposits had been one-way until then, which meant early validators locked funds with no exit.
- Proto-danksharding (March 2024). Introduced blobs — a cheap, temporary data space for rollups, which cut layer-2 costs sharply.
- Pectra (2025). A combined execution and consensus upgrade, raising blob capacity and improving account and validator mechanics.
- Fusaka (3 December 2025). Introduced PeerDAS, which lets each node store roughly an eighth of blob data rather than all of it, raising data capacity substantially without raising the hardware required to participate.
Each of these upgrades assumes most activity happens on layer 2 and that the main chain's job is to guarantee that layer-2 data was published and is available for verification. That is a deliberate architectural choice, made years ago, and it is why Ethereum's own throughput has not increased dramatically while the cost of transacting on rollups has fallen a great deal.
How to read upgrade announcements
Upgrades are named, scheduled by block height, specified in public improvement proposals, and implemented independently by several client teams. An upgrade that has activated has a block number and a date; one that has not is a plan, however confidently it is described. The distinction is worth holding onto when reading coverage.