EIP-4844 introduces shard blob txs that allow storing non-persistent (14 day retention) data blobs on CL nodes. Most rollups expect to use blobs for data avaiability as it is expected to cost less than posting via calldata (data is not retained forever, resource is priced separately from execution gas). The current spec target/max blobs per block is 3/6 blobs/block which corresponds to .375/.750 MB/block so 31.25/62.5 kb/s. Full danksharding targets 16MB/block which is roughly 1.33MB/s assuming we stay at a 12s blocktime (unlikely probably goes up).

Celestia recently announced they are ready to launch. Their intial blocks will be 2MB but can go up to 8MB pending governance. Blocktime is set to 15 seconds. At genesis, Celestia can process 133KB/s which is > 4x what 4844 will support. Celestia can presumably push much further than this, but haven’t done much research into that.

EigenDA also had a recent announced. They claim to be targeting 10MB/s with a credible roadmap to 1GB/s. If they really do launch with 10MB/s that’s 75x more than celestia genesis config and 320x more than initial protodanksharding. At 10MB/s that’s 7.5x what full danksharding is targeting.

There’s good reasons to use Ethereum DA as it makes the rollup construction safer (other constructions look very similar to plasma??) and it will be the most battle tested at least for a while. I’d expect Celestia and EigenDA to be much less costly in addition to having more throughput, which will make them attractive for high-throughput applications. To be fair, 99.99999% of users probably do not care about how data availability is done and you can just get away with a DAC, though lose the rollup label. DA is commoditizing.

Ethereum DA is for security-minded ants.