The technical reality of Ethereum’s Verkle tree migration

Running a full Ethereum node requires hundreds of gigabytes of storage because the network currently uses the Patricia Merkle Trie (MPT) data structure. This structure forces nodes to store the entire state, which limits decentralization by raising hardware requirements. Verkle trees change this by using vector commitments instead of hash-based branching. These trees enable stateless clients to verify and produce blocks without storing the full state. These clients receive witnesses, which act as small cryptographic proofs, alongside each block to prove the correctness of the state data. You probably know that state bloat threatens network decentralization, but Verkle trees provide a way to bypass the need for massive local storage. This shift belongs to the Verge phase of Vitalik Buterin’s staged roadmap, which focuses on making Ethereum nodes dramatically lighter and more accessible.

Merkle-Patricia Tries versus vector commitments

The Ethereum state consists of a large key-value map where keys include account addresses or contract storage slots and values include balances or stored data. Currently, Ethereum encodes these in an MPT where the root hash appears in every block header. The root is a master fingerprint for the entire state. If anything inside the state changes, the root changes. To prove a value in this large map, a program must send the trie root and adjacent nodes along the path. This process requires checking all encoded nodes, which leads to large proofs. A single proof can be several kilobytes, and a whole block’s worth of proofs can reach tens of megabytes.

Verkle trees upgrade the math by using vector commitments. Each internal node uses a vector commitment, pointing to other nodes, to commit to many children at once. This allows the tree to be shallower and keeps proofs compact. The trade-off involves fewer bytes on the network but more complex cryptography and heavier CPU work. The locality principle for statelessness suggests that keeping recently touched parts of the state nearby on the state tree leads to smaller witness sizes. A proposal suggests including the account address as a value alongside the balance, nonce, code, and storage. This allows the tree location to depend on the time the account last changed. Whenever an account is touched in a block, the subtree is deleted from its location and reinserted at a new location based on the current block height and index within the block.

Feature Merkle-Patricia Trie (MPT) Verkle Trees
Proof Type Hash-based intermediate nodes Vector commitments
Proof Size Large (megabytes per block) Small (compact witnesses)
CPU Load Lower (hashing) Higher (commitment checks)
Node Storage Full state required Stateless capability possible

Testing bottlenecks and implementation hurdles

Developers currently test Verkle Trees on two testnets: Beverly Hills and Kaustinen. The Beverly Hills testnet is a minimal implementation that focuses solely on Execution Layer clients and the Verkle Tree specification. The Kaustinen testnet implements changes to the Ethereum Consensus Layer along with Verkle Trees. The current implementation of Verkle Trees remains about 40% slower than the existing Merkle Patricia Trees, which creates significant performance concerns for developers who want to transition the network without sacrificing any transaction throughput capacity or stability. I find the 18-hour conversion time on the Erigon implementation to be completely unacceptable. Because the conversion from Merkle Patricia Trees to Verkle Trees takes eighteen hours on the fastest Erigon implementation, this technical bottleneck makes the adoption of Verkle Trees significantly more difficult for all Ethereum clients.

Testing has also left several gaps. Developers tested Verkle Trees against chain reorganizations, but they have not assessed JSON RPC performance or tested with shadow forks. No multi-client devnet forks have executed, which raises concerns about compatibility across various clients. Another challenge involves the need for a new sync mode. Developers must also address the difficulty of the subtree delete and move operation, which is a complicated primitive to implement.

The 2026 roadmap and the Hegota outlook

Ethereum developers follow a schedule of two hard forks per year. Following the Pectra and Fusaka upgrades, the network moves toward Glamsterdam in the first half of 2026 and Hegota in the second half of 2026. Hegota might include Verkle Trees, but the exact content remains undefined. Guillaume Ballet reported that the Verkle Trees specifications appear stable and make good progress in early testing. Other developers, such as Lukasz Rozmej of Nethermind, suggest focusing on a feature-rich fork to avoid delays rather than prioritizing Verkle Trees. This debate defines the current development atmosphere. If all clients do not adopt a similar database architecture to Erigon, they must rely on specific nodes to gossip the Verkle Trees, which poses a risk. How will the network ensure compatibility across various clients if multi-client devnet forks do not execute?

Newsletter