The table format war ended without a press release

Two years ago we reviewed the open table format standoff and issued the most useful verdict in this journal's toolbox: wait. Committing an estate to Iceberg or Delta while both camps were sprinting felt like picking a videotape format in 1979.

The waiting worked. Iceberg v3 folded in the features Delta partisans used to win arguments with: deletion vectors, row lineage, variant types. Snowflake took v3 to general availability in May, and Databricks, Delta's own parent, shipped it in public preview while proposing that Iceberg v4 and a future Delta 5.0 share one metadata layout outright. When both armies adopt the other's weapons, the war is over regardless of what the flags say.

Note what actually decided it. Not benchmarks, which stayed inconclusive to the end. Catalogs. Once every major engine could read and write Iceberg through a catalog interface, the format stopped being a loyalty oath and became plumbing. The remaining differences are real but operational: compaction behaviour, streaming ergonomics, the pace of engine support for v3 features.

Revised verdict. The format question is settled enough that relitigating it in your architecture review is a waste of a meeting. Ask instead who runs your catalog, because that is where the lock-in moved, and it did so while everyone was watching the format war. The vendors know this, which is why catalog pricing pages have grown so cheerful.

We were right to tell you to wait. We are telling you to stop waiting. It is plumbing now; buy the boring one and go argue about something that still matters.