What is EIP-4844?

EIP-4844, or Proto-Danksharding, introduced a new transaction type to Ethereum: the blob-carrying transaction (type 0x03). Instead of paying for permanent calldata storage, rollups can attach large data chunks that consensus clients verify once and discard after roughly 18 days.

The result is a separate, much cheaper market for data availability sitting alongside the regular gas market. Rollups inherit Ethereum’s security without paying execution-layer prices for raw data they only need to keep around for fraud proofs and reconstruction.

At launch, each block could carry between 0 and 6 blobs with a target of 3. Each blob holds up to 128 KB of data, secured by KZG cryptographic commitments that let the network verify availability without every node downloading the data permanently.

EIP-4844 was always designed as a stepping stone toward full Danksharding, the long-term plan to give Ethereum massive data bandwidth through data availability sampling.

How Blob Transactions Work

Blob transactions sit alongside regular transactions but carry a "sidecar" of data that lives on the consensus layer rather than the execution layer. The mechanics are deliberately simple:

• Temporary storage: Each blob is retained for about 4,096 epochs (~18 days) before being pruned. Long enough for honest rollup participants to download the data, short enough to keep node disk requirements manageable.
• Fixed size: Every blob holds up to 128 KB of data, regardless of whether it is fully utilised. Senders pay for the entire blob.
• EVM invisibility: Smart contracts only see a versioned hash of the blob, which keeps execution-layer costs low.
• Separate fee market: Blobs use their own EIP-1559-style pricing, with a base fee that adjusts based on whether the previous block was above or below target. All blob base fees are burned.

This design gave rollups vastly cheaper data availability without bloating the execution layer or pricing out everyday users.

Retrospective: Two Years of Blobs

When Proto-Danksharding shipped in March 2024, the impact was immediate. Median fees on Arbitrum, Optimism, Base, and zkSync collapsed from tens of cents to fractions of a cent overnight. Daily activity on Base alone tripled in the months that followed.

By late 2024, demand was running close to the 3-blob target almost every block. A single airdrop or token launch could push fees from fractions of a cent to several dollars per blob, eroding the cost advantages that made blobs attractive in the first place.

Two structural problems became obvious as the year wore on:

1. Blob supply was too tight. With only 3 target blobs per block and a hard ceiling of 6, the system could not absorb growing rollup demand without fee spikes.
2. Blob fees were sometimes too low to be meaningful. During quiet periods, the base fee would collapse to 1 wei. Great for rollups, but it broke price discovery and meant nodes were doing real cryptographic work for almost no compensation.

Both issues were addressed in the upgrades that followed.

Pectra: Doubling Blob Capacity

The Pectra upgrade, activated on May 7, 2025, was Ethereum’s first major hard fork of 2025. The headline features were validator-facing changes like EIP-7251 and EIP-7702, but the most important change for rollups was buried inside EIP-7691.

EIP-7691 doubled blob throughput overnight. The per-block target moved from 3 to 6 and the maximum from 6 to 9. The base fee adjustment fraction was also tuned, with full blocks now raising the base fee by ~8.2% per block instead of 12.5%.

Pectra also shipped EIP-7623, which raised the gas cost of calldata from 16 to 42 gas per byte. The intent was simple: nudge any remaining rollups still using calldata to migrate to blobs.

For most of 2025, this combination held up well. L2 fees stayed low and blob utilisation hovered around the new 6-blob target. But the next bottleneck was already visible, and developers were preparing something more ambitious.

Fusaka and PeerDAS: The Real Scaling Leap

Fusaka activated on December 3, 2025 and represents the biggest change to blob infrastructure since EIP-4844 itself. Where Pectra was a tuning exercise, Fusaka introduced an entirely new way for the network to handle blob data, confirmed live by the Ethereum Foundation.

The headline feature is PeerDAS (Peer Data Availability Sampling), defined in EIP-7594. Before Fusaka, every node had to download every blob in full, which placed a hard ceiling on how much data the network could process without compromising decentralisation.

PeerDAS removes that constraint by splitting blob data into 128 columns distributed across nodes, with each node only storing and serving a subset. The network as a whole still has access to everything, but no individual node carries the full load.

Regular full nodes only need to subscribe to 4 of the 128 column subnets, cutting blob bandwidth requirements by roughly 8x. This is the same data availability sampling concept that underpins full Danksharding, deployed in a form production validators can run today.

BPO Forks: Scaling in Real Time

Alongside PeerDAS, Fusaka introduced EIP-7892, which defines a new type of mini-upgrade called a Blob Parameter Only (BPO) fork. BPO forks adjust the blob target, maximum, and base fee update fraction without bundling any other protocol changes.

Ethereum can now raise blob capacity in measured steps as the network proves it can handle higher throughput, instead of waiting 12 to 18 months for the next hard fork. The first two BPOs activated on schedule:

• BPO1 (December 9, 2025): Raised the blob target from 6 to 10 and the maximum from 9 to 15.
• BPO2 (January 7, 2026): Raised the blob target from 10 to 14 and the maximum from 15 to 21.

These two forks alone tripled blob capacity in a month. Core developers are already planning further BPOs aiming for 48 blobs per block by mid-2026, with the long-term target of 128 blobs per slot under full Danksharding.

Blob Fee Revenue Since Launch

Every blob base fee paid on Ethereum is burned, which makes blob revenue a direct contributor to ETH’s deflationary mechanism. The track record so far is more nuanced than the headline numbers suggest.

In the first year after Dencun, blob transactions paid roughly 1,020 ETH in EIP-4844 base fees plus another ~2,000 ETH in EIP-1559 fees, for a total of around $8 million. Modest by L1 standards, and well below the $34 million per month L2s were spending on calldata before the upgrade.

The reason is simple: outside of brief spikes from events like Blobscriptions and the LayerZero airdrop, the blob base fee spent most of 2024 and 2025 sitting at or near 1 wei. Rollups were getting near-free data availability, and Ethereum was capturing almost none of the value.

Fidelity Digital Assets research modelled what blob revenue would have looked like with a proper fee floor in place. Their estimate: an additional ~24,641 ETH (roughly $78 million) of cumulative revenue would have been generated between Dencun and October 2025 if EIP-7918 had been live.

That is exactly what Fusaka fixed.

EIP-7918: Making Blob Fees Meaningful

EIP-7918, shipped in Fusaka, ties the minimum blob base fee to the L1 execution base fee. Blob pricing now always reflects the actual computational cost of verifying KZG proofs, with the floor set at 1/15.258 of L1 gas.

The mechanism does not affect rollups during periods of genuine demand, since the equilibrium fee in those moments sits well above the floor. What it prevents is the price collapsing to nothing when L1 gas is cheap.

The immediate effect was dramatic. The blob base fee surged by roughly 15 million times relative to its pre-Fusaka floor. Industry estimates suggest blob fees could contribute 30-50% of total ETH burn by 2026, depending on how L2 activity scales.

Blob revenue is becoming a real product line for Ethereum rather than a subsidised service for rollups, which changes the long-term economic relationship between L1 and L2.

Impact on Layer 2 Fees and Activity

The combined effect of EIP-4844, Pectra, and Fusaka has been transformative for the Layer 2 ecosystem:

• Fee compression: A typical L2 transaction that cost ~$0.50 in late 2025 dropped to between $0.20 and $0.30 in the weeks following Fusaka, with further reductions as BPO capacity is utilised.
• Throughput: The combined L2 ecosystem currently processes ~5,600 TPS, with developers projecting 24,000+ TPS as the BPO schedule progresses through 2026.
• Rollup base case: Ethereum Foundation modelling suggests rollup fees could fall an additional 40-60% during the first months of Fusaka, with potential drops of 90%+ as capacity ramps further.
• L1 throughput: Fusaka also raised the default block gas limit target to 60 million via EIP-7935, letting the base layer process ~20-30% more transactions per block.

The net effect is an Ethereum L1 that is significantly more capable, paired with an L2 ecosystem that finally has the headroom to onboard the next wave of users.

What Comes Next

The path from here is clearer than at any point in Ethereum’s scaling roadmap. PeerDAS is the technical foundation for full Danksharding, and BPO forks give core developers a way to scale capacity without full network upgrades. Each BPO is a stress test that validates whether the network can handle the next jump.

Beyond the existing schedule, two further milestones are on the Ethereum roadmap:

• Glamsterdam (planned for 2026): The next major hard fork after Fusaka, expected to bring further execution layer improvements alongside continued blob scaling.
• Full Danksharding: The endgame for Ethereum data availability, with a target of around 128 blobs per slot. No longer the speculative concept it was when EIP-4844 launched.

Bottom Line

EIP-4844 was Ethereum’s most consequential scaling change since The Merge, and the two years since launch have validated the rollup-centric roadmap in a way few would have predicted in early 2024. L2 fees have collapsed, rollup activity has grown into the dominant share of Ethereum usage, and the network has the infrastructure to keep scaling for years.

Pectra doubled blob capacity, Fusaka introduced PeerDAS and BPO forks, and EIP-7918 made sure blob fees would actually mean something as throughput grows. Proto-Danksharding was always the prototype. What is running on mainnet today is starting to look a lot more like the real thing.