The reality of Ethereum blob capacity and PeerDAS scaling

Ethereum reached a significant threshold in May 2025 when Pectra launched. This upgrade increased the blob target from 3 to 6. It also introduced EIP-7702, which allows externally owned accounts to execute smart contract code for single transactions. This enables transaction batching and social recovery. Pectra also raised the maximum effective validator balance to 2,048 ETH via EIP-7251. The community also raised the mainnet gas limit from 30M to 60M. Following this, the Fusaka upgrade arrived in December 2025. It introduced PeerDAS (EIP-7594) to solve the bandwidth bottleneck. This mechanism allows for a 48 blob target per block. This expansion increases throughput from ~220 to ~3,500 UOPS. Rollups depend on this abundant, low cost blobspace to sustain user activity.

PeerDAS and node economics

PeerDAS modifies how blobs exist on the network. It uses Reed-Solomon erasure coding to expand the size of each blob. The block proposer organizes these encoded blobs into a two-dimensional matrix. This matrix consists of 128 columns that the proposer splits and distributes. Each node then subscribes to specific subnets to receive its assigned data slices. A regular full node without validators handles at least 4 data columns in custody and samples another 4. These nodes do not need to download entire blobs.

Validators with more stake subscribe to more subnets. For every 32 ETH staked, nodes store an extra data column. Nodes staking 1,824 ETH or more handle at least 64 columns, which allows them to reconstruct the whole set. Nodes staking 3,872 ETH or more must handle all 128 columns and act as supernodes. Each cell in the 128 columns holds 2 kB of data. Nodes use unique, randomly generated IDs to determine which subnets they join. By subscribing to a subset of subnets, a node only receives 1/8 of the original data. This reduces the bandwidth and storage requirements for the majority of the network.

Instability at high blob counts

The network capacity does not always follow protocol targets. MigaLabs found that the network failed to reach the target blob count of 14. The median blob count actually fell from 6 to 4 after the BPO1 update. I view the reliability issues at high blob counts as a major problem for the ecosystem. The MigaLabs study shows that while the median blob count fell from 6 to 4 after BPO1, the miss rate for slots with 21 blobs reaches 1.79% which is significantly higher than the baseline. You probably think more blobs always equals better scaling, but the network struggles when it hits high counts.

At the Hoodi testnet, the p99 latency for 15 blobs reached 12.4s. This exceeds the 4s deadline for validator attestation. High blob counts also increase the failed attestation rate. For 15 blobs, the p95 failed attestation rate reaches 7%. At 16 or more blobs, the miss rate for slots ranges from 0.77% to 1.79%. This is more than double the 0.5% baseline observed for 0 to 15 blobs. Fusaka also introduced EIP-7934, which caps the RLP execution block size at 10 MiB to prevent DoS attacks. This limit leaves 2 MiB for consensus framing. Will the network reach the 48 blob target without hitting a wall of missed slots?

L2 scaling and pricing behavior

L2 economics rely on the ability to post data predictably. Arbitrum introduced the ArbOS Dia upgrade in January 2026 to stabilize fees. It replaced single-window pricing with multiple gas targets and adjustment windows. Short windows handle bursts, while long windows keep long-run system constraints safe. This change makes the system behave like shock absorbers to handle demand bursts. The minimum L2 base fee also rose from 0.01 gwei to 0.02 gwei. This change makes spam-style usage less cheap.

L2 operators like Base face challenges with replacement fees. The Ethereum Foundation recommends a matched node-cost benchmark and a replacement-fee audit for Base. These investigations aim to resolve the 40 higher-tip transactions seen in August 2025. Better L2 posting reliability helps keep fees low for end users. I see the connection between L1 blob capacity and L2 sequencer budgets as the primary driver for the next phase of scaling. EIP-7918 also helps by pinning blob prices to execution costs to prevent blob prices from crashing to 1 wei. This ensures rollups pay a meaningful fraction of the compute and space they cause.