Iceberg won. Now what?

The table format wars are over and Apache Iceberg won. Snowflake reads it, Databricks caved and embraced it alongside Delta, BigQuery supports it, every catalog vendor pitches “your data, open formats, no lock-in.” As someone who works in Snowflake daily, I think the peace is real. I also think people are over-reading what it means.

What Iceberg settles: the storage layer. Your tables live in object storage in an open format, and in principle any engine can read them. That’s a genuinely big deal for the ten-year horizon of not being hostage to one vendor’s pricing.

What it doesn’t settle: everything above the storage layer, which is most of the job.

Somebody still owns the catalog, and whoever owns the catalog owns you a little. Permissions, compaction, snapshot expiration, statistics, write coordination when two engines touch one table. None of that got easier because the file format is open. Some of it got harder, because now there are more moving parts with more owners.

I watched a team spend a quarter on an “open lakehouse” migration whose actual outcome was swapping one set of proprietary conveniences for three new operational duties nobody had budgeted for. Their query bills went down. Their on-call load went up. Whether that trade was good depends on numbers nobody wrote down beforehand, which is its own lesson.

My honest position, as someone early in her career who expects to live with these decisions for decades: default to boring. If you’re on Snowflake and it works, native tables plus an eye on Iceberg for the datasets where engine flexibility genuinely matters is a defensible place to stand in 2026. Adopting Iceberg because the discourse says lock-in is scary is just lock-in fear doing vendor marketing’s job for it.

Open formats are a real option worth holding. An option is not an obligation to exercise it everywhere at once.