EIP-4844 (proto-danksharding) compressed blob fees to near zero in 2024. Two years in, the L2 cost curve has plateaued and the next leg down requires full danksharding — still 18+ months out. Here is where Base, Arbitrum, Optimism, and zkSync sit on cost, throughput, and revenue today, and where the curve goes from here.
Key takeaways
- Median L2 tx fee: $0.012 (Base), $0.018 (Arbitrum), $0.016 (Optimism), $0.024 (zkSync).
- Total L2 TVL: $42B across the top-10 L2s; Base + Arbitrum are 64% of it.
- L2 fee revenue to Ethereum L1 (via blob posting): ~$11M/month, down 70% from 2024 peak.
- Full danksharding (the next compression catalyst): targeted Dencun+1 hardfork, ~Q3 2027 estimate.
What proto-danksharding actually changed
Before EIP-4844, L2s posted calldata to Ethereum L1 — paying the same gas any other transaction would. After 4844, they post “blobs” that are pruned after 18 days, and the gas market for blobs is separate from regular execution gas. Blob fees collapsed; L2 tx costs followed.
For users, this was a 10–50x cost reduction. For Ethereum L1, it was a 70% revenue reduction from L2 traffic. Both are true.
Where each L2 sits today
| L2 | Median tx fee | TPS (peak) | TVL | Posture |
|---|---|---|---|---|
| Base (Optimism stack) | $0.012 | ~340 | $15.6B | Coinbase user funnel, social/Farcaster heavy |
| Arbitrum One (Nitro) | $0.018 | ~280 | $11.4B | DeFi-heavy, oldest ecosystem |
| Optimism (Bedrock) | $0.016 | ~210 | $5.2B | Superchain (Base, Mode, World) |
| zkSync Era | $0.024 | ~180 | $2.8B | ZK rollup, account-abstraction native |
| Starknet | $0.041 | ~110 | $0.9B | Cairo VM, slower adoption |
The plateau
Fees have stopped declining. Base hit ~$0.005 median in Q1 then drifted back to $0.012 as activity rose. Arbitrum sits in the $0.015–0.020 band consistently. The bottleneck is no longer blob cost — it is sequencer overhead and the fact that blob throughput is currently capped at 3 per slot (will rise to 6 with the next blob-target increase).
Until full danksharding lands (PeerDAS, blob count to 64+ per slot), the L2 cost curve plateaus here.
What this means for L2 token holders
L2 native tokens (ARB, OP, MNT, others) have struggled relative to ETH through 2025. Two reasons: lower fee revenue means lower sequencer revenue means thinner future-cash-flow case; and the proliferation of L2s means each is fighting for a smaller share of users.
That said, the Superchain thesis (Optimism Collective: Base, Optimism, Mode, World shared sequencing and unified liquidity) has measurable traction. If Superchain TVL aggregates above $25B by year-end, the OP-token-as-claim-on-sequencer-share thesis re-tightens.
“L2 token holders are betting the sequencer becomes a real cash-flow business. So far, the cash flow is small and the competition is large.”
For users vs holders
If you use L2s, you have already won — fees are 10–50x lower than L1. If you hold L2 tokens, the cost compression that helped users came at the cost of the value accrual story for the token.
Model output for ARB: 30d base near current price; 1y bear-case wider than usual. For ETH itself: the L2 revenue compression is already priced in but is a structural headwind to “ETH is cash-flow asset” narratives.
Why this matters
L2 fees define crypto’s UX at the wallet level. The 2024 fee compression was the single biggest UX upgrade Ethereum has shipped in years. The 2025-2026 plateau is real but temporary — and the next compression wave is 18+ months out. Build product accordingly.