EIP-4844 and Proto-Danksharding
How Ethereum's March 2024 Dencun upgrade introduced blob space — a new transaction type that cut rollup data costs by ~10x and set the stage for full danksharding.
The basics
Why Rollups Needed Cheaper Data
Layer-2 rollups (Arbitrum, Optimism, Base, zkSync, etc.) inherit their security from Ethereum by posting all their transaction data back to Ethereum. Before March 2024, they did this by stuffing the data into expensive 'calldata' fields in regular Ethereum transactions. Calldata is permanent — every Ethereum node stores it forever — and pricing it like permanent storage made rollup fees high.
What EIP-4844 Did
EIP-4844 (Ethereum Improvement Proposal 4844), activated in the Dencun upgrade on March 13, 2024, added a new kind of data attachment called a 'blob.' Blobs are big chunks of data (128 KB each; up to 6 per block at launch, a limit since raised) that piggyback on transactions but are stored separately and deleted after about 18 days. Because they're temporary, they're much cheaper than calldata. Rollups can use blobs to post their batch data, and rollup fees dropped by roughly 10x almost immediately.
What Users Saw
Ahead of Dencun, Base's engineering team said it expected the upgrade to drive down fees for users by 10-100x. In practice, the same actions on rollups that adopted blobs became much cheaper than when rollups posted their data as calldata.
- Activated in the Dencun upgrade on March 13, 2024
- Adds 'blobs' — big temporary data attachments cheaper than calldata
- Cut L2 fees by ~10x almost overnight
- First step toward full 'danksharding' (a much bigger blob capacity in future upgrades)
Key Takeaways
- Rollups need to post data to Ethereum to inherit its security
- Pre-Dencun, that data went in expensive permanent calldata
- EIP-4844 added blobs: large, temporary, ~10x cheaper
- L2 fees dropped immediately on activation; the upgrade is widely seen as one of Ethereum's most successful
Going deeper
Blobs in Detail
A blob is a 4096-field-element data structure where each field is 32 bytes — so each blob is exactly 128 KB. At Dencun launch each block could carry up to 6 blobs (768 KB total), with a target of 3 (384 KB), so the fee market can equilibrate. Pectra (7 May 2025, EIP-7691) raised this to a target of 6 and a maximum of 9, and Blob Parameter Only forks after the Fusaka upgrade raised it to a target of 10 and a maximum of 15 (9 December 2025), then a target of 14 and a maximum of 21 (7 January 2026). As of November 2026 the mainnet limits are a target of 14 and a maximum of 21 blobs per block. Blobs are propagated through a separate sidecar network and only the data's KZG commitment (a 48-byte cryptographic summary) goes into the actual block — the full data lives in a side channel and gets deleted after ~18 days.
KZG Commitments — The Math That Makes It Work
KZG (Kate-Zaverucha-Goldberg) commitments let a blob's content be cryptographically summarised into a tiny commitment that fits in a block. Anyone can later prove that a specific value at a specific position is part of the committed blob, using a small proof. This is what makes blob data verifiable without making every node store every blob forever. The KZG setup required a trusted setup ceremony — the largest in cryptography history, with ~140,000 participants in 2023.
The Blob Fee Market
Blobs have their own fee market separate from regular gas. When the blob target (14 per block as of November 2026) is exceeded, the per-blob price increases exponentially via an EIP-1559-style mechanism. When demand is below target, it decreases, but since the Fusaka upgrade (EIP-7918) a reserve price tied to execution gas costs stops the blob fee from being stranded at 1 wei. In quiet periods, blobs are cheap; during congestion (heavy L2 activity), prices can spike. The two fee markets — regular gas for execution, blob gas for data — are independent.
What 'Proto-Danksharding' Means
EIP-4844 is called 'proto-danksharding' because it's a stepping stone to full danksharding (named after researcher Dankrad Feist). Full danksharding would increase blob capacity far beyond the current maximum of 21 per block (as of November 2026), requires full data availability sampling (its first step, PeerDAS, went live with the Fusaka upgrade on 3 December 2025), and remains a future Ethereum upgrade. The 'proto' version got the blob infrastructure into production so rollups could start using it immediately.
Key Takeaways
- Each blob is 128 KB; as of November 2026 up to 21 per block (target 14)
- Only a 48-byte KZG commitment goes into the block; the data lives in a separate sidecar network
- Blobs delete after ~18 days; storage cost stays bounded for nodes
- Blob fees use an independent EIP-1559-style market separate from regular gas
- This is 'proto' — full danksharding (many more blobs per block + full DAS) remains on the roadmap; PeerDAS, its first step, shipped in December 2025
Under the hood
Why Blobs Improve Censorship Resistance
Calldata-based rollup data is in the execution payload — censoring builders can exclude it like any other transaction. Blob data is in a separate sidecar that the consensus layer treats as a commitment. As blob inclusion becomes the standard, the censorship-resistance properties of L2s improve because the data path is cleaner and the consensus layer has stronger guarantees about what was posted vs what was withheld.
Data Availability Sampling (The Endgame)
Full danksharding plans to expand blob capacity far beyond today's limits. But that creates a problem — no single full node could download all of it. The solution is data availability sampling (DAS): nodes randomly sample tiny pieces of each blob and use erasure coding to be cryptographically confident that the whole blob is available without downloading it. DAS is hard to implement correctly and the focus of substantial research (see the related learn topic on DAS for detail). Proto-danksharding did not implement DAS. A first form of DAS, PeerDAS (EIP-7594), went live with the Fusaka upgrade on 3 December 2025; further DAS work is part of the 'full' danksharding upgrade.
What This Means for Rollup Economics
The 10x cost reduction shifts where rollup costs come from. Pre-Dencun, L1 data was the dominant cost of an L2 transaction. Post-Dencun, the L1 data share dropped sharply, and execution costs (proving, sequencer fees) became the larger share for many rollups. This changed the competitive landscape: rollups now compete more on prover efficiency and sequencer cost than on data-batching tricks. It also means future blob-capacity expansions will have diminishing returns for end users — once L1 data is a small fraction of total cost, halving it again saves less than halving the first time.
Validator Bandwidth and Hardware Implications
Adding blob data to every block raised bandwidth requirements for validators. At Dencun launch each block could carry a target of 3 blobs (~384 KB) and a maximum of 6 (~768 KB), and every full node downloaded every blob. Since then the limits have risen: as of November 2026 the target is 14 blobs (~1.8 MB) and the maximum 21 (~2.7 MB), but since the Fusaka upgrade nodes check blob availability by sampling rather than downloading entire blobs. This was a deliberate trade-off — validators absorb more bandwidth to enable cheaper rollups for everyone. Subsequent upgrades (Pectra, then the Blob Parameter Only forks that followed Fusaka) have increased blob capacity, putting pressure on validator hardware requirements and informing the long-running debate about how much to centralise (heavier validators) vs. decentralise (lighter validators with DAS) the consensus layer.
Watching Blob Usage Tells You About L2 Demand
Real-time blob-usage trackers (e.g., on Etherscan's blob explorer) show how full each block's blob capacity is. When blobs are consistently at the per-block maximum, L2 demand is exceeding current capacity and fees rise. When they're far below the target, L2 demand is light and blobs are cheap. Watching this is a clean read on L2 activity that's hard to manipulate.
Key Takeaways
- Blob-based data improves censorship resistance vs calldata-based posting
- Full danksharding builds on data availability sampling; PeerDAS, its first step, shipped with Fusaka on 3 December 2025, and full danksharding remains on the roadmap
- L1 data's share of L2 costs dropped sharply post-Dencun, shifting competition to prover/sequencer efficiency
- Validators pay the bandwidth cost; capacity expansion is an ongoing centralisation-vs-decentralisation trade-off
- Blob fullness is a clean real-time read on L2 demand
Related Content
Related Coins
Key Terms
2026 Trends
More Topics
Browse all 189 topicsCrypto Tax Rules by Country
How cryptocurrency is taxed in major jurisdictions — US (1099-DA, cost basis), UK (HMRC Section 104), EU (DAC8), and key Asian markets. Not tax advice — what the rules actually say.
AI Meets Crypto: What's Real, What's Hype
Where artificial intelligence and cryptocurrency actually intersect — AI agents with wallets, decentralized compute, verifiable inference, and the new scam vectors AI creates.
Bitcoin ETFs & Institutional Crypto
How spot Bitcoin ETFs work, what institutional adoption means for the market, and why owning ETF shares is fundamentally different from owning the coin.
UTXO vs Account Model
How Bitcoin's UTXO model differs from Ethereum's account model — what each gets right, what each makes harder, and why the choice shapes everything from parallelism to replay protection.
Account Abstraction (ERC-4337)
How ERC-4337 brings smart-contract wallets, social recovery, gasless transactions, and arbitrary signature schemes to Ethereum without changing the protocol — and why every wallet UX of the next decade will use it.
Data Availability Sampling
How light nodes can be cryptographically confident that a block's full data was published — without downloading it — and why this is the key unlock for Ethereum's full danksharding and the modular blockchain stack.
Finality and Fork Handling
What it means for a transaction to be 'final,' why different chains finalise differently, how reorgs happen, and the MEV incentives that complicate the picture.
State Pruning and Expiry
Why Ethereum's node storage needs have grown to terabytes, what 'archive,' 'full,' and 'light' nodes actually store, and how proposals like EIP-4444 and new state-tree designs (verkle or binary) aim to keep node hardware accessible for years to come.
References & further reading
- primary
- secondary
- primaryEthereum Foundation Blog - Fusaka Mainnet Announcement
Fusaka activation time and the BPO1/BPO2 schedule (10/15, then 14/21 blobs).
- primaryethereum.org - Fusaka
PeerDAS, blob parameter only forks and the EIP-7918 blob fee reserve price.