Fusaka & PeerDAS — blobs get cheaper, again

Dencun gave rollups blobs — a cheap, throwaway data lane. Fusaka takes the next step: it lets the network carry *more* blobs without asking every node to download *all* of them. The trick is PeerDAS — Peer Data Availability Sampling.

The problem with just adding more blobs

After Dencun, the limit on blob throughput wasn't math — it was bandwidth. Every full node downloaded every blob to guarantee the data was actually published. Double the blobs, double everyone's bandwidth. That doesn't scale.

What PeerDAS changes

Instead of downloading the whole blob, a node downloads a few random samples and gossips with peers. If the samples are available, the math says (with overwhelming probability) the *whole* blob is available. Data is erasure-coded, so even a fraction of the pieces can reconstruct the rest.

BEFORE (Dencun)            AFTER (PeerDAS)
node ── downloads ALL      node ── samples a few columns
 blob data every slot       + checks with peers
                            │
 bandwidth ∝ total blobs    bandwidth ∝ samples (≈ constant)
 → caps blob count          → blob count can grow

The payoff lands downstream: more blob capacity → lower competition for blob space → cheaper L2 transactions, inherited for free by every rollup that posts data to Ethereum.

Why sampling is safe

The worry with "don't download everything" is obvious: what if the missing piece is the one being hidden? Erasure coding is the answer. The blob is expanded so any sufficiently large *fraction* of the pieces can rebuild the whole. To hide data, an adversary would have to withhold so much that random sampling would almost certainly hit a missing piece — and any node that samples raises the alarm. A handful of random checks per node becomes a network-wide guarantee.

blob ──erasure code──→ 2x pieces (any ~50% rebuilds all)
 each node samples a few random pieces:
   all present  → available (overwhelming probability)
   one missing  → raise the alarm, reject the block

This is the unlock for full danksharding later: once nodes verify availability by sampling, blob *count* can scale with the size of the network instead of with any single node's bandwidth. Fusaka is the first production step onto that path.

Ethereum keeps the same security promise — "the data was published" — while asking each node to do less work to verify it.