Here's Ethereum's embarrassing truth: after a decade of development and hundreds of billions in value, it can process about 15-30 transactions per second. During busy periods, a simple token swap can cost $50 in gas fees and take minutes to confirm.

Visa processes 65,000 transactions per second. A typical database can do millions. Ethereum can barely run a busy dApp.

The reason? Every node must process every transaction, store all data, and maintain the entire blockchain state. That's secure and decentralized, but it doesn't scale. Sharding is Ethereum's long-term solution: split the blockchain into parallel pieces that process transactions independently, multiplying throughput by orders of magnitude. It's elegant in theory, nightmarishly complex in practice. And after years of research, the approach has evolved into something quite different from what was originally envisioned.

One Blockchain Becomes Many

Sharding is a database concept blockchain borrowed: split a large dataset into smaller pieces distributed across multiple machines. Each shard handles a subset of data, allowing parallel processing.

In traditional blockchains, every validator processes every transaction. In sharded blockchains, the network splits into multiple shards. Each shard processes its own transactions. Validators are assigned to specific shards.

Result: 64 shards means theoretically 64x more transactions. Imagine a highway with 64 parallel lanes instead of one.

But these shards need to communicate, maintain security, and coordinate consensus—all without creating new attack vectors. That's the trillion-dollar problem.

The Plot Twist: Rollup-Centric Ethereum

Ethereum's original 2018 roadmap envisioned 64 execution shards—each effectively an independent Ethereum chain. Total throughput: potentially 1,000x higher.

But as research progressed, Vitalik Buterin and others realized this was unnecessarily complex and that Layer 2 rollups could achieve similar results more quickly. In 2020, Ethereum's roadmap pivoted dramatically.

The new vision: Ethereum Layer 1 becomes a settlement and data availability layer. Layer 2 rollups handle transaction execution. Sharding focuses purely on data availability for rollups, not execution.

Why? Execution shards were too complex. Rollups were already working, demonstrating Layer 2 could scale Ethereum 10-100x. Data availability was the bottleneck—rollups need to post data to Layer 1, but Layer 1 didn't have enough capacity.

Instead of 64 execution environments with complex cross-shard logic, have one secure Layer 1 and many Layer 2s that post data back. This led to Danksharding.

Danksharding: The Modern Approach

Danksharding, named after researcher Dankrad Feist, is Ethereum's current sharding design. Instead of sharding execution, shard data availability. Ethereum becomes a massive data availability layer for rollups.

Special transactions carry large data "blobs" up to 125 KB each. These blobs aren't part of the EVM state—they're temporary, prunable data that rollups use to post their transaction data. Validators don't download all blob data. They randomly sample small pieces to verify it's available. Cryptographic magic, specifically KZG polynomial commitments, ensures that if validators can sample the data, the full data must be available.

Start with a few blobs per block, eventually scale to thousands with full Danksharding. Rollups use blobs instead of expensive calldata—blob data can be pruned after about 18 days since rollups only need it temporarily.

Eventual scale: Full Danksharding could enable 100+ MB per block of data availability, enough to support hundreds of rollups processing millions of transactions per second combined.

Proto-Danksharding: The First Step

Full Danksharding is years away. So Ethereum implemented Proto-Danksharding via EIP-4844 in March 2024 as an intermediate step. It introduced blob-carrying transactions with 3-6 blobs per block, created a new transaction type for blob data, implemented the blob fee market separate from regular gas fees, and reduced Layer 2 costs by 90-95% overnight.

Before EIP-4844, rollups paid for expensive calldata. A transaction on Optimism or Arbitrum might cost $1-5. After EIP-4844, rollups use blob space, much cheaper. Same transaction now costs $0.05-0.50. Layer 2 fees dropped 90%+ within days. This was the most impactful Ethereum upgrade since The Merge.

Proto-Danksharding doesn't require data availability sampling—validators still download all blobs. But it proved the mechanism works and set the foundation.

Data Availability Sampling: The Secret Sauce

Here's the technical magic. If a block contains 100 MB of blob data, every validator downloading 100 MB doesn't scale. The solution: validators randomly sample small pieces. If enough random samples are available, you can prove statistically that the full data must be available.

The blob data is encoded using erasure coding—the same technique used in RAID storage. The blob is committed using KZG polynomial commitments—a cryptographic scheme that lets you verify individual pieces without having the full dataset.

Each validator randomly samples a few pieces. If a sufficient number successfully sample, you're statistically certain the full data is available.

Result: A validator only downloads a few KB to verify a multi-MB blob. This breaks the link between data size and node requirements. More data doesn't mean heavier nodes, it means more sampling across more validators.

Security Challenges

Sharding introduces new attack vectors. A malicious validator could post a blob commitment but not publish the data. Mitigation: erasure coding with redundancy. If an attacker withholds 49%, the other 51% reconstructs everything.

In execution sharding, an attacker could compromise a single shard—easier than the whole chain—and attack others. This is why Ethereum moved away from execution sharding. Rollups provide isolation.

These challenges are why sharding took so long—every attack vector must be closed before deploying with hundreds of billions at stake.

Why Sharding Is So Hard

Database sharding is mature. Blockchain sharding is cutting-edge research. In traditional databases, you trust the operator. In blockchains, you trust nothing. Every component must be verifiable and attack-resistant.

Adding shards shouldn't require more powerful hardware. Cross-shard transactions must agree atomically—really hard in a decentralized system.

This is why Ethereum's pivot to rollup-centric sharding is smart—it avoids most problems by not doing execution sharding.

The Sharding Timeline

The Beacon Chain launched December 2020. The Merge happened September 2022, transitioning Ethereum to proof-of-stake. Proto-Danksharding via EIP-4844 launched March 2024, introducing blobs.

Estimated 2025-2026: PeerDAS testing, where validators begin using data availability sampling instead of downloading full blobs. Estimated 2027+: Full Danksharding with thousands of blobs per block, massive data availability for rollups.

This timeline has slipped repeatedly—original Phase 1 was supposed to be 2021-2022. That's not failure, it's the reality of building infrastructure this complex while maintaining security.

The Endgame

If full Danksharding succeeds, Ethereum Layer 1 processes 15-30 TPS unchanged, provides 100+ MB per block of data availability via blobs, and acts as ultra-secure settlement and data availability layer. Dozens of Layer 2 rollups each process 1,000-10,000 TPS, posting data to Ethereum via blobs, with combined throughput of 100,000+ TPS across the ecosystem.

The user experience: you might never touch Layer 1 directly. All activity happens on rollups with instant, near-free transactions. Security is inherited from Layer 1. Cross-rollup transfers become seamless.

This isn't one sharded blockchain. It's a layered architecture where Layer 1 handles security and data availability, while Layer 2s handle scale. It's not what was originally designed. But it might be better.

The Philosophical Shift

Sharding represents a philosophical evolution. From 2015-2020, the thinking was "scale the base layer." From 2020-2025, it became "Layer 1 should be ultra-secure and simple, Layer 2s handle scale."

This is similar to how the internet evolved. Early internet had everything at the network layer. Modern internet uses layered protocols. Ethereum is becoming the TCP/IP of finance—the simple, secure base layer everything else builds on.

The Blockchain That Learned to Multiply

Sharding is blockchain's attempt to have its cake and eat it too: massive scale without sacrificing decentralization or security. The journey has been long and full of pivots. Execution sharding became data sharding. The 2021 timeline became 2026+.

But progress is real. EIP-4844 cut Layer 2 costs by 90%. Full Danksharding could cut them another 99%. Ethereum could secure millions of transactions per second across dozens of rollups.

That's the promise. Not one blockchain trying to do everything, but one blockchain enabling everything else to scale. The next few years will tell us if it works.

References

1. Ethereum Roadmap: Danksharding
2. EIP-4844: Proto-Danksharding
3. Dankrad Feist: KZG Polynomial Commitments
4. Vitalik Buterin: Sharding FAQ
5. The Block: EIP-4844 Impact on L2 Fees
6. Ethereum Foundation: Data Availability Sampling
7. NEAR Protocol: Nightshade Sharding
8. Polkadot: Parachain Architecture
9. Ethereum: The Merge Documentation
10. Flashbots: Ethereum Sharding Research