Don't Forget about Iceberg
Originally posted on Linkedin.
This year’s Summit included several headline-grabbing releases: Document AI, Native App Framework and Snowpark Container Services, which collectively address the more obvious risks the company faces around the ability to run AI/ML workloads.
But the Summit reveal that made the biggest impression on me was unified Iceberg Tables, because it begins to address two genuine risks to the Snowflake model that we’ve heard from operators and founders:
The introduction/advent of a new data control plane that enables flexible compute without the need for data centralization.
A lack of developer centricity despite the continued shift in decision-making power to Engineering.
(Full disclosure: I was at Snowflake before joining IVP).
It’s incredibly difficult to target a different user base a decade into a company’s existence, especially when that company has been one of the most successful in history. But that’s where the importance of Snowflake’s continued investment in interoperability through Iceberg comes in.
Unifying external and internal Iceberg Tables into a single type lays the groundwork for mitigating the above risks by:
Combatting the perception of Snowflake as a closed-source product that will aggressively lock you in.
Strengthening its positioning as an extensible platform by tapping into an independently growing ecosystem.
Demonstrating (over time – because they’re certainly not there yet) that Snowflake is also building for developers, rather than only for the IT, data, analytics and business teams.
Positioning Snowflake as the unified data control plane. If you’ve already built out your access control in Snowflake, wouldn’t you welcome the ability to apply those same controls anywhere that data is stored?
Though we’re in the early innings of these features, I’m curious to see how their customers respond. Large enterprises are trying to architect their data platform for the next decade, and for many the key decision is which horse to hitch their wagon to – Snowflake or Databricks. How will they evaluate these developments from Snowflake next to Databricks’ recent announcements? Between its acquisition of MosaicML and launch of UniForm, Vector Search, Fine-tuning and a Marketplace for models, Databricks is effectively positioning itself as the most complete platform for AI/ML development.
Which company is better positioned will remain a topic of debate, but we can all agree on one thing – it would be a lot simpler if they held their conferences on different days.