Ethereum's Blob Transactions and the New Economics of Layer 2

The Night Ethereum Fees Collapsed

On March 13, 2024, Ethereum activated the Dencun upgrade at epoch 269568. Within hours, transaction fees on Arbitrum dropped from roughly $0.50 to under $0.01. On Base, fees that had averaged $0.20 fell below $0.001. Optimism users saw costs plunge 95%. zkSync fees hit fractions of a cent. This was not a temporary liquidity anomaly — it was the permanent structural consequence of EIP-4844, also known as proto-danksharding, and it rewrote the economics of Ethereum's Layer 2 ecosystem overnight.

What Are Blob-Carrying Transactions?

Before EIP-4844, Layer 2 rollups posted their compressed transaction batches to Ethereum as calldata — the same data field used for smart contract inputs. Calldata is permanent, stored forever in Ethereum's history, and priced accordingly. Every byte of calldata cost 16 gas (or 4 gas for zero bytes), and during periods of high demand, that cost ballooned to economically prohibitive levels for L2 users.

EIP-4844 introduced a new transaction type (type 3) that carries blobs — Binary Large OBjects of 128 KB each. Blobs attach to transactions but are stored separately from the EVM execution layer. Crucially, blobs are not permanently stored: consensus clients prune them after approximately 18 days (4096 epochs). The EVM itself cannot read blob data directly — only a cryptographic commitment (a KZG commitment) is accessible on-chain. This design means blob data is available long enough for fraud proofs and validity proofs to be submitted, then discarded, eliminating permanent storage overhead.

The Separate Fee Market: Why Blobs Are Cheaper

EIP-4844 established a completely independent fee market for blob data, separate from the standard EVM gas market. Blobs have their own base fee that adjusts via EIP-1559-style mechanics targeting 3 blobs per block (with a maximum of 6). When blob demand is low — which it has been for most of 2024 and 2025 — blob fees stay near their theoretical floor of 1 wei. This is structurally different from calldata, which competed directly with DeFi swaps, NFT mints, and every other Ethereum transaction for the same gas.

The immediate practical effect: L2s that previously spent $0.10–$0.50 per batch in calldata costs now spend a tiny fraction of that in blob fees. Those savings are passed directly to users as lower transaction fees.

Before and After: Real Fee Data

The numbers from the week of the Dencun activation tell the story clearly:

• Arbitrum One: Average swap fee dropped from ~$0.48 to ~$0.004 — a 99% reduction
• Optimism: Average transfer fee fell from ~$0.20 to ~$0.003 — a 98.5% reduction
• Base: Average transaction fee collapsed from ~$0.19 to ~$0.0008 — a 99.6% reduction
• zkSync Era: Fees dropped from ~$0.15 to ~$0.005 — a 97% reduction
• Starknet: Fees fell from ~$0.10 to ~$0.002 — a 98% reduction

These reductions were not temporary. By mid-2025, L2 fees have stabilized at 10–50x cheaper than pre-Dencun levels, with periodic spikes during congestion events that still resolve far below historical pre-EIP-4844 baselines.

Impact on Each Major Layer 2

Arbitrum (Nitro stack): Arbitrum was already the most cost-efficient optimistic rollup before Dencun. Post-EIP-4844, Arbitrum migrated to blob posting in its Nitro stack update. The result: Arbitrum became economically viable for micro-transactions and gaming applications that were previously impossible at $0.50 fees. Arbitrum Orbit chains — L3s built on Arbitrum — can now achieve sub-$0.001 fees by posting to Arbitrum rather than directly to Ethereum.

Optimism and Base (OP Stack): The OP Stack ecosystem, including Base (Coinbase's L2), adopted blobs rapidly. Base in particular saw explosive growth post-Dencun, with daily active addresses surpassing Ethereum mainnet on several occasions in 2024. The Superchain vision — multiple OP Stack chains sharing security and interoperability — became economically coherent once blob fees made cross-chain data sharing affordable. Optimism's Fault Proof system, which requires data availability during the 7-day challenge window, fits naturally with blobs' 18-day retention period.

zkSync Era: ZK rollups benefit from EIP-4844 differently than optimistic rollups. Their validity proofs are submitted on-chain (not blob data), but the underlying transaction data that proves correct state transitions is now posted as blobs. zkSync's hyperchain architecture — allowing application-specific ZK chains to settle to Era — gained economic viability post-Dencun. ZK proof generation costs, not data posting, are now the dominant cost driver for zkSync.

Starknet: Starknet adopted recursive STARK proofs that compress large batches efficiently. Post-EIP-4844, Starknet's data posting costs dropped dramatically. The team has focused on throughput increases, with transaction finality times decreasing as blob capacity allowed larger batch submissions. Starknet's Cairo programming language and native account abstraction make it distinct technically, but economically it now competes on a much more level playing field.

Polygon (CDK and zkEVM): Polygon's zkEVM and the Polygon CDK (Chain Development Kit) both integrated blob support. Polygon AggLayer — the aggregation layer connecting multiple Polygon CDK chains — uses blob data for cross-chain state proofs. The economic benefit here extends beyond just L2 fees: the entire Polygon ecosystem of application-specific chains became viable at production scale.

The Long-Term Roadmap: Full Danksharding and PeerDAS

EIP-4844 is explicitly a stepping stone. The name "proto-danksharding" signals its relationship to the full danksharding vision proposed by researcher Dankrad Feist. Here is what comes next:

PeerDAS (Peer Data Availability Sampling): Currently in active development and testnet deployment as of 2025, PeerDAS will allow Ethereum nodes to verify blob availability without downloading entire blobs. Nodes sample small random chunks and use erasure coding to confirm the full blob is available across the network. This will enable a dramatic increase in blob capacity — the current 3-blob target will scale to potentially 32–64 blobs per block.

Full Danksharding: The eventual target is 256 blobs per block, each 128 KB, for approximately 32 MB of data availability per block. At this capacity, L2s could post data cheaply even at very high throughput, potentially supporting millions of transactions per second across the entire Ethereum ecosystem. The timeline for full danksharding remains multi-year, likely 2027–2028.

The blob fee market at scale: As blob capacity increases, blob fees will remain low unless L2 throughput grows proportionally. The current 3-blob target was deliberately conservative to allow the ecosystem to stress-test the new architecture. As PeerDAS lands, the target will increase incrementally, keeping the fee market balanced.

Revenue Impact on Ethereum: The Sound Money Question

EIP-4844 has a direct and measurable negative impact on Ethereum's fee burn mechanism. Under EIP-1559, base fees are burned, reducing ETH supply and supporting the "ultrasound money" narrative. Blob fees are also burned, but they are structurally lower than the calldata fees they replaced.

The data is stark: in the months before Dencun, L2s were paying millions of dollars per week in calldata fees to Ethereum, all of which was burned. Post-Dencun, blob fee revenue to Ethereum dropped by over 90%. Combined with a general reduction in mainnet activity as users migrated to cheaper L2s, ETH issuance began exceeding fee burn for extended periods in 2024 — a reversal of the deflationary dynamic that had prevailed since the Merge.

This has reopened debate about ETH's monetary policy. The counter-argument from Ethereum core developers: the long-term value of Ethereum comes from being the settlement layer for a high-throughput ecosystem, not from extracting maximum rent from L2s. A thriving L2 ecosystem with billions of users transacting cheaply creates more aggregate demand for ETH (for gas, staking, DeFi collateral) than a more expensive but less-used system.

Practical Guide: Choosing an L2 in 2025-2026

With fees now a fraction of pre-Dencun levels across all major L2s, fee cost alone is no longer the primary differentiator. Here is how to evaluate L2s for specific use cases:

• DeFi trading and high-frequency activity: Arbitrum One remains the deepest liquidity venue with the most mature DeFi ecosystem (GMX, Uniswap V3 pools, Aave, Pendle). Base is a strong second for Coinbase-integrated products and social applications.
• Gaming and micro-transactions: ImmutableX (ZK-based, focused on gaming) and Arbitrum Nova (using AnyTrust data availability for ultra-low fees) are optimized for high-volume, low-value transactions. Arbitrum Orbit L3s offer the best economics for dedicated gaming applications.
• Enterprise and compliance-sensitive applications: zkSync's native account abstraction and Starknet's Cairo VM offer programmable signing and session keys — critical for enterprise UX. Polygon CDK chains offer customizable data availability options.
• NFTs and social: Base has emerged as the dominant NFT and social chain post-Dencun, driven by Coinbase distribution and low fees enabling affordable minting.
• Cross-chain applications: The OP Superchain and Polygon AggLayer are building shared liquidity and messaging layers. Applications that need to operate across multiple chains should evaluate these ecosystems first.

How to Evaluate Fees in Practice

Do not rely on static fee comparisons — L2 fees are dynamic and depend on current blob market conditions, L2-specific sequencer pricing, and activity levels. Use L2Fees.info or GrowThePie.xyz for real-time cross-L2 fee data. When assessing an L2 for a production application, calculate the all-in cost: sequencer fee (what users pay), data posting cost (blob fees, which flow to Ethereum), and proof submission cost (for ZK rollups). The last two are invisible to users but determine L2 economics and long-term sustainability.

One underappreciated dynamic: as blob capacity increases with PeerDAS, the marginal cost of posting data to Ethereum will approach near-zero. At that point, the competitive differentiation between L2s will shift entirely to execution environment quality, developer tooling, liquidity depth, and security track record — not data availability costs. The L2s investing in these dimensions now are positioning for the post-scarcity data availability world.

Takeaways

• EIP-4844 permanently reduced L2 data costs by 90–99%, making micro-transactions viable across Arbitrum, Optimism, Base, zkSync, Starknet, and Polygon
• Blobs use a separate fee market with 18-day retention — purpose-built for rollup data availability without permanent storage overhead
• PeerDAS will scale blob capacity 10–20x, likely by 2026, pushing fees even lower
• Fee cost is no longer the primary L2 differentiator — evaluate liquidity, ecosystem, security maturity, and developer tooling instead
• Ethereum fee burn decreased significantly post-Dencun; the trade-off is a larger, more active L2 ecosystem as the long-term value driver
• For most users in 2025–2026: use Base or Arbitrum for DeFi, Arbitrum Nova or Orbit for gaming, zkSync or Starknet for enterprise-grade account abstraction