Session

NBCUniversal Media Group's Journey to Databricks

Overview

ExperienceIn Person
TrackData Engineering & Streaming
IndustryCommunications, Media & Entertainment
TechnologiesDatabricks SQL, Delta Sharing, Unity Catalog
Skill LevelBeginner
DOWNLOAD SESSION SLIDES

NBCUniversal launched Peacock in 2020 with a BigQuery based reporting and analytics data stack. This provided a solid platform for the business to grow on. As data volumes and computational needs increased, in 2025 it became necessary to evaluate other technology. With an eye on data sharing, scalability and of course, cost, Databricks rose to the top of the bunch. In this talk, we will discuss the factors that led to this decision, the process of migration and the road ahead.

Session Speakers

Louie Kuznia

/VP, Data Engineering
NBCUniversal

Full Summary

Inside Peacock's migration from BigQuery to Databricks: lessons from a high-stakes streaming platform

Peacock scaled fast after its 2020 launch, reaching petabytes of data, billions of daily events, and heavy concurrent usage by 2024. Keeping executive dashboards responsive, batch SLAs on time, and analysts productive forced a rethink of the core warehouse and execution layer. The result was a carefully sequenced move of silver and gold layers from BigQuery to the Databricks Platform that offers a practical playbook for any large-scale enterprise analytics platform evaluation.

FAQ


Apache Iceberg let the team store data once and serve it from multiple systems. Iceberg tables on Databricks were exposed back to BigQuery as external tables and were usable from Snowflake, cutting copies and reducing lock-in.

Silver and gold were rebuilt on Databricks while BigQuery stayed live. Iceberg tables were exposed as external tables in BigQuery, so most users only changed a few characters in table references, and the team validated results before turning off legacy paths.

Two stand out. They underestimated the hands-on support analysts needed to migrate code, and they started users on small warehouses, which felt slow before tuning. Proactive migration help plus a range of compute options worked better.

Yes. Shifting storage into cloud buckets provided immediate savings and compute tuning closed the gap. The team reports roughly a 30 percent reduction in monthly total ownership costs despite larger data volumes and more concurrent users.

They are running proofs of concept with Genie and investing in knowledge discovery. The near-term goal is helping users find the right datasets among thousands of tables, not replacing dashboards outright.