Level 3 · How it worksThe village#259

Sharding explained: split the copying, but someone still has to check every street

Sharding splits a blockchain's work so that no single computer has to handle all of it. Ethereum dropped its plan for separate shard chains and now splits only the data, while chains such as NEAR split the transactions themselves.

The village notebook has a problem. Every house copies every line, and the lines keep coming. By lunchtime, Grandma Rosa's old table is covered in paper.

So someone suggests the obvious fix: split the work. Mill Street copies the lines about flour. Church Street copies the lines about wool. Each house keeps a quarter of the notebook instead of all of it.

Rosa asks the question that decides everything: "If I only copy Church Street, who checks Mill Street?" That is sharding. The splitting is easy. Keeping every street honestly checked is the hard part.

What it is

Sharding splits a blockchain into parts, called shards, so that each node handles only some of the work. More shards can mean more transactions in total, because the parts run side by side.

There are two kinds to keep apart. Execution sharding splits the transactions and balances themselves: each shard runs its own share of the accounts. Data sharding splits only the job of storing and checking data, while the transactions are processed somewhere else, such as on a rollup.

The catch is the same for both. A shard watched by a few validators is easier to fool than a chain watched by all of them.

How it works: two plans, with numbers

Ethereum's first plan was execution sharding. In October 2019, Ethereum researcher Danny Ryan wrote that the starting shard count "must be reduced from 1024 to the new estimate of 64" (Ethereum Foundation). His reason is the security catch in one line: "there couldn't possibly be enough 32-ETH validators to safely allocate enough to each committee". With 1,024 streets, there weren't enough guards to post a safe number on each.

In September 2020, the roadmap had a "Phase 2" to "add full support for sharded smart contract execution" (EIP-2982).

Then the plan changed. On 2 October 2020, Vitalik Buterin proposed that "Ethereum base-layer scaling would primarily be focused on scaling how much data blocks can hold" (Ethereum Magicians). Rollups would run the transactions. The main chain would give them room for data. ethereum.org puts it flatly: "Shard chains are no longer part of the roadmap" (ethereum.org). What replaced them, danksharding, gets its own two articles next.

So Ethereum's main chain runs every transaction in one place. Block 26,449,058, written at 12:14:35 UTC on 19 November 2026, used 30,452,446 gas of its 60,000,000 limit (Etherscan). Every validator that checked it re-ran all of that work. Nothing was split.

NEAR took the other road. Its developers wrote in December 2025 that mainnet ran 9 shards, each processing "a disjoint portion of the state" (NEAR One). Producers assemble one "chunk" per shard, and messages between shards travel through the following blocks.

The same post reports a benchmark of 1 million transactions per second, using 70 shards and 140 validator nodes. Do the sum:

  1. Per shard: 1,000,000 divided by 70 is about 14,286 transactions per second.
  2. Guards per shard: 140 nodes divided by 70 shards is 2 nodes each.
  3. What was measured: native token transfers, the simplest kind of transaction, in a test network, not on mainnet.

The post says all of this openly. That is how a speed claim should be published: with the test conditions next to the number.

Why it matters

  • Who gains. Users, if more transactions fit through the network at once. Node operators, if they only need to keep and run one slice.
  • Who pays. Everyone, in complexity. A payment from one shard to another becomes a message between two parts of the system, with an extra step in between.
  • Where the risk sits. Each shard has fewer eyes on it. That was the reason Ethereum cut its plan from 1,024 shards to 64, and one reason it dropped shard chains altogether.
  • Why Ethereum chose data. Splitting data is simpler to keep safe than splitting execution. The rollups take on the running of transactions, and the main chain only has to make sure their data is published.

Where people get it wrong

  • "Sharding splits the coins." It splits the work. Your balance is one balance, kept in one place in the system.
  • "Ethereum has shards." It dropped shard chains. What it built instead is room for rollup data, explained in "Blobs and danksharding explained".
  • "Ten times the shards, ten times the speed." Messages between shards, the number of validators and the type of transaction all get in the way. Benchmarks with simple transfers are the best case.
  • "A shard is just a sidechain." A sidechain has its own validators and its own security. Shards are parts of one system that is supposed to guard all of them.

Ask who checks each street

When a network advertises sharding and a big transactions-per-second figure, ask two things. Was the figure measured on mainnet or in a benchmark? And how many validators check each shard?

A good project answers both in the same post as the number. A shard that nobody checks isn't a shard. It's a side street with the lights off.

Frequently asked questions

What is sharding in blockchain?

Sharding splits a blockchain into parts, called shards, so that each node handles only some of the work. Execution sharding splits the transactions and balances. Data sharding splits only the storage and checking of data, while transactions are processed elsewhere.

Does Ethereum use sharding?

Not in its original form. Ethereum planned 64 shard chains in 2019, but from October 2020 it moved to a rollup-centric plan, and ethereum.org states that shard chains are no longer part of the roadmap. Instead it is scaling the amount of rollup data blocks can hold, a plan called danksharding.

Why did Ethereum reduce its shard plan from 1,024 to 64?

In October 2019, the Ethereum Foundation explained that there could not be enough validators to put a safe number on each committee for 1,024 shards. Fewer shards meant more validators watching each one.

Which blockchains use sharding?

NEAR is one example, not a recommendation. Its developers wrote in December 2025 that mainnet ran 9 shards, each handling its own part of the network's state, and reported a benchmark of 1 million transactions per second using 70 shards.

Is a sharded blockchain less secure?

It can be, if each shard is checked by only a small group. An attacker then only needs to overpower that group rather than the whole network. Sharded designs try to prevent this by assigning validators carefully and letting others check the shards' work.

Sources

  1. Ethereum Foundation blog, Danny Ryan, eth2 quick update no. 2 (31 October 2019)
  2. EIP-2982, Serenity Phase 0 (created 15 September 2020)
  3. Vitalik Buterin, A rollup-centric ethereum roadmap (Ethereum Magicians, 2 October 2020)
  4. ethereum.org, Danksharding (page updated 22 October 2026)
  5. NEAR One, Scaling NEAR to 1M TPS (11 December 2025)
  6. Etherscan, Ethereum block 26,449,058 (read on 19 November 2026)

Read next.