EIP-4844 is the Ethereum upgrade that introduced “blobs” — cheap, temporary data packets that rollups use to post transaction data to the main chain instead of paying for expensive calldata. It shipped as part of the Dencun upgrade on March 13, 2024, and it’s the reason Layer 2 transaction fees dropped from roughly a dollar to a fraction of a cent almost overnight.
The proposal is also known as proto-danksharding, a preview of Ethereum’s longer-term plan to shard blockchain data across the network. Here’s what it actually does, why Ethereum needed it, who relies on it day to day, and how the mechanism has evolved through 2026.
What Is EIP-4844?
EIP-4844 added a new transaction type — the blob-carrying transaction — that lets a rollup attach up to 128 KB of data to a block without that data becoming part of Ethereum’s permanent execution history. Validators store each blob for about 18 days, long enough for anyone to verify it, then discard it.
That temporary storage is the whole point. Before EIP-4844, rollups had to publish their transaction batches as calldata, a field that gets bundled permanently into Ethereum’s state and priced accordingly. Blobs get their own separate fee market, so they’re priced independently of regular gas and don’t compete with normal transactions for block space.
Why Ethereum Needed EIP-4844
The Calldata Problem Before Blobs
Rollups such as Arbitrum, Optimism, and Base process transactions off-chain, then post compressed batches back to Ethereum so anyone can verify what happened. Before Dencun, that posting step used calldata, and calldata is expensive because Ethereum treats it as permanent history that every future node has to store forever.
By early 2024, calldata costs from rollups were consuming a meaningful share of Ethereum block space, and that cost got passed straight to users. OP Labs estimated before launch that a dedicated, temporary data channel could cut rollup L1 costs by at least 20x — the core justification for building blobs instead of just tweaking calldata pricing.
The Fix: Data That Doesn’t Need to Last Forever
Rollup data only needs to be available long enough for the network to verify it and for anyone to challenge a fraudulent batch. It doesn’t need to sit in Ethereum’s state permanently the way a token balance does. EIP-4844 exploits that distinction: blobs are cryptographically committed to the chain but pruned after roughly two weeks, which is why they’re so much cheaper than calldata.
Who Uses EIP-4844?
Blobs aren’t something an end user interacts with directly — they’re infrastructure that specific players depend on:
- Rollup operators (Arbitrum, Optimism, Base, zkSync, Starknet, and dozens more) post transaction batches as blobs instead of calldata, which is now their primary cost driver.
- Wallets and dapp frontends benefit indirectly: lower L1 settlement costs are a big part of why swapping or minting on an L2 can cost a few cents instead of several dollars.
- Block explorers and analytics platforms track blob usage, blob base fees, and per-rollup data consumption as a proxy for network demand.
- Node operators and infrastructure providers need clients that can store, gossip, and serve blob data alongside regular blocks, which is a meaningfully different job than serving standard JSON-RPC calls.
How Blob Transactions Work
Blob Size and Structure
A single blob holds 4,096 field elements of 32 bytes each, for a fixed size of 131,072 bytes — about 128 KB. A transaction can carry multiple blobs, but each one is priced and verified independently. This fixed sizing is deliberate: it makes blob data predictable to gossip across the peer-to-peer network and compatible with future upgrades.
KZG Commitments
Instead of publishing the raw content of every blob to every node, Ethereum uses a KZG commitment — a compact cryptographic proof that lets a node verify a blob is correct without downloading the whole thing. That’s a deliberate design choice: the same commitment scheme is what full danksharding will eventually need for data-availability sampling, so proto-danksharding is forward-compatible rather than a throwaway shortcut.
A Separate Fee Market
Blob gas has its own base fee, adjusted block-by-block using the same exponential logic as EIP-1559, but tracked against a blob-specific target and maximum rather than the regular 30 million gas limit. When blob usage runs above target, the blob base fee rises; when it runs below target, it falls. This keeps blob demand from ever competing with ordinary transactions for the same fee market.
| Calldata (pre-blobs) | Blob transactions | |
|---|---|---|
| Storage | Permanent, part of chain history | Pruned after ~18 days |
| Fee market | Shares regular gas market | Separate blob base fee |
| Data per unit | Priced per byte, no fixed cap | Fixed 128 KB per blob |
| Verified via | Full data included in block | KZG commitment |
From Dencun to Fusaka: How Blob Capacity Has Grown
Blob capacity was never meant to stay static. Ethereum launched conservatively and has raised the ceiling through a series of scheduled upgrades as the network proved it could handle more.
| Upgrade | Date | Blob target | Blob max |
|---|---|---|---|
| Dencun (EIP-4844 launch) | March 13, 2024 | 3 | 6 |
| Fusaka (PeerDAS) | December 3, 2025 | 6 | 9 |
| BPO1 | December 9, 2025 | 10 | 15 |
| BPO2 | January 7, 2026 | 14 | 21 |
That progression matters because blob demand caught up with supply faster than expected. In September 2024, with blob usage already running near 75% of target, Vitalik Buterin warned that “the ecosystem is sleeping on the fact that it’s uncomfortably close to a ceiling.” That pressure is a direct reason the Fusaka upgrade shipped PeerDAS — Peer Data Availability Sampling — which lets nodes verify blob availability by checking small samples instead of downloading every blob in full. That’s what made it safe to raise the ceiling repeatedly without forcing every node to handle proportionally more bandwidth.
By January 2026, Ethereum’s maximum data throughput per block had grown to roughly 2,688 KB, up from 768 KB at Dencun’s launch — a 3.5x increase in under two years, with more BPO forks planned.
EIP-4844 vs. Full Danksharding
Proto-danksharding is explicitly a stepping stone, not the finished design. The distinction matters if you’re trying to understand where Ethereum’s broader scaling roadmap is actually headed, and how blobs fit into the wider scalability, security, and decentralization trade-offs every blockchain has to balance.
| Proto-danksharding (EIP-4844) | Full danksharding | |
|---|---|---|
| Blob source | Proposer includes full blob data | Data spread across the validator set |
| Node requirement | Nodes still process every blob | Nodes sample a fraction via data-availability sampling |
| Target capacity | Tens of blobs per block | 64+ blobs per block |
| Status | Live since March 2024, expanding via BPO forks | Still in active research and design |
Full danksharding will eventually let the network scale blob capacity without every node needing to store or verify all of it directly — the sampling approach PeerDAS already previews at a smaller scale.
Real-World Impact: What Changed for Layer 2 Fees
The numbers back up why this upgrade got so much attention. Layer 2 transaction fees fell by a factor of 100 to 200 in the weeks after Dencun, with average costs dropping from around $1 to well under a cent on several rollups. By April 2026, after BPO2 pushed the blob ceiling to 21, simple swaps on rollups like Base were routinely settling for a few cents even during busier periods.
Blob utilization has stayed comfortably below the new ceiling — averaging somewhere in the 20-30% range of capacity through the first quarter of 2026 — which suggests Ethereum built in real headroom rather than just chasing the previous quarter’s demand. That headroom is also what makes further blob-fee spikes less likely in the near term, though rollup adoption keeps climbing as more chains launch on Ethereum’s data-availability layer.
Where Infrastructure Providers Fit In
Serving blob data isn’t the same job as answering a standard JSON-RPC call. A node has to store, gossip, and prune blobs on their own schedule, and a growing blob ceiling means growing storage and bandwidth requirements for anyone running that infrastructure.
NOWNodes is a blockchain infrastructure provider that gives developers API-based access to 120+ networks, including Ethereum, without each team having to run and maintain that infrastructure themselves. For a rollup team or a dapp built on top of one, that means reading blockchain state, checking balances, or broadcasting transactions through a maintained endpoint instead of operating the underlying client software in-house. Coverage and available methods vary by network, so it’s worth checking NOWNodes’ Ethereum access page for the specific interfaces a given project needs.
Risks and Limitations
EIP-4844 solved a real cost problem, but it isn’t a complete fix for everything people mean by “Ethereum scaling.”
- It doesn’t reduce L1 execution costs. Regular Ethereum transactions and smart-contract calls are priced the same as before; only rollup data posting got cheaper.
- Blob data is temporary by design. Rollups and any archival service that needs blob history beyond the pruning window have to copy and store it themselves.
- Capacity increases require careful pacing. Each BPO fork raises bandwidth and storage demands on every node, so Ethereum’s core developers deliberately stage increases rather than jumping straight to danksharding-level targets.
- It benefits rollups, not every use case equally. Applications that settle directly on Ethereum L1 rather than through a rollup see no direct fee benefit from blobs.
Conclusion
EIP-4844 didn’t change what Ethereum can execute — it changed how rollups pay to prove what they executed, replacing permanent, expensive calldata with temporary, purpose-priced blobs. That single change, refined through Dencun, Fusaka, and two BPO forks since, is the biggest reason Layer 2 fees now sit in the cents rather than dollars. The roadmap toward full danksharding is still unfolding, but proto-danksharding already did the hard part: proving the blob model works at scale.
FAQ
How long do blobs stay on Ethereum?
Blobs are pruned after roughly 18 days (4,096 epochs). Anyone who needs the data longer — a rollup’s own archive, an indexer, or a block explorer — has to copy it before it’s deleted from consensus-layer storage.
Do blob transactions affect regular Ethereum gas fees?
No. Blob gas is priced on a separate fee market with its own base fee, target, and maximum, so blob demand doesn’t compete with or inflate the cost of ordinary transactions.
How many blobs can fit in one Ethereum block?
The limit has changed several times since launch. It started at a maximum of 6 blobs per block at Dencun in March 2024 and reached a maximum of 21 following the BPO2 fork in January 2026, with further increases planned.
Is EIP-4844 the same as danksharding?
No. EIP-4844 is proto-danksharding, an intermediate step that uses the same cryptographic commitments full danksharding will eventually need, but without the data-availability sampling across the validator set that full danksharding requires.
Which rollups use EIP-4844 blobs?
Most major Ethereum rollups, including Arbitrum, Optimism, Base, and zkSync, post their transaction data as blobs rather than calldata. Blob usage across these networks is the main driver of blob demand on Ethereum today.
