Databricks: Get ready for the "lakebase"

Welcome to Runtime! Today: Databricks floats a new portmanteau for the ever-evolving world of data infrastructure, a widespread Google Cloud outage takes down sites around the world, and the latest enterprise moves.

(Was this email forwarded to you? Sign up here to get Runtime each week.)

Different strokes for different folks

SAN FRANCISCO — First there was the database. Then came the data warehouse, which was followed by the data lake. People liked data warehouses and they liked data lakes, so data companies put data warehouses in data lakes and called them lakehouses. And now, after all these decades, Databricks feels like the time is right to bring the whole thing full circle with the "lakebase."

That was the term used by Databricks co-founder and CEO Ali Ghodsi Wednesday at the company's Data & AI Summit to describe his company's plans for Neon, the serverless PostgreSQL database vendor Databricks acquired for $1 billion last month. The lakebase is designed to challenge traditional OLTP (online transaction processing) databases still used extensively inside modern companies despite the notion that "the technology in transactional databases hasn't actually changed that much in the last 40 years," Ghodsi said in his keynote address.

• There are four major components to the "lakebase," which Ghodsi clarified in a press conference after the keynote is expected to describe a new, broader category of databases even though the company is calling the public preview of its managed Postgres product "Databricks Lakebase."
• Lakebases use open-source components such as the wildly popular Postgres database, and separate compute from storage using open formats.
• They are serverless, which in this context means they can scale up and down quickly and seamlessly, and they are designed to serve as the back end for AI agents.
• "Our prediction is every other transactional OLTP database will evolve towards this architecture in the coming years now," said Reynold Xin, co-founder and chief architect at Databricks, during the keynote.

Databricks thinks the lakebase will bring enterprise tech one step closer to an elusive vision data vendors have been promising for years: The ability to store all your corporate data in a cheap repository like AWS's S3 and funnel it to whatever computing or data-analysis engine you like. The data industry's embrace of open formats over the last year has helped, but gaps remain.

• Data vendors figured out long ago how to scale data storage independently of compute, which in the bad old days meant you were paying a lot more every time you added more data.
• However, there are still a lot of older transactional databases in use that have compute and storage tightly coupled, which made sense at the time given that transactional databases need lightning-fast access to data in order to process online orders.
• Database companies such as Cockroach Labs have found success handling OLTP workloads with a modern approach, but the company no longer offers its flagship product under an open-source license, which makes it ineligible for Databricks' just-invented lakebase category.

Ghodsi also predicted that as companies roll out AI agents, demand for databases will dramatically increase. Last year 30% of the databases created with Neon were created by agents, he said, and this year that number has soared to 80%.

• He acknowledged that the adoption of AI agents has "not been where we want it to be yet," but the tools are coming together that will help companies be more successful with agentic AI experiments.
• And just like pretty much every company in enterprise tech has done over the past year, Databricks rolled out an agent-building tool this week during the conference called Agent Bricks.
• Ghodsi could sell bags of sand to beachfront property owners, but if agentic AI really does lead to an explosion in software development, all those apps are going to need databases.
• "This is an over $100 billion [total addressable market] for us, so it is a big opportunity," he said. "I think it's going to be a journey, but I think it's going to be the most important marathon that Databricks runs over the next five to ten years."

Don't worry baby

Google Cloud suffered one of its worst outages in recent memory Thursday, taking down services such as Spotify, Cloudflare, and, perhaps most tragically, Major League Baseball's streaming service. The outage affected a huge number of global cloud infrastructure regions as well as key Google Cloud services like BigQuery.

Google first acknowledged the outage on its status page at 11:46am PT, but several sites reported outages or service disruptions earlier than that. At 2pm it reported that most services were starting to recover including instances running in its us-central1 region in Iowa, which appeared to be ground zero for this outage, but it did not share any details about what went wrong.

Cloud infrastructure services have had a pretty good run over the last few years when it comes to uptime, according to recent research from The Uptime Institute, but streaks are meant to be broken. Given the number of high-profile services that were taken down by this outage — which ironically is a sign of Google Cloud's growth over the last few years — expect to see a more-detailed explanation in the coming days.

Enterprise moves

Trevor Schulze is the new chief information officer at Genesys, joining the customer-experience company following similar roles at Alteryx, Ring Central, and Micron Technology.

The Runtime roundup

Oracle's stock rose 13% Thursday after it reported revenue and profit that beat Wall Street's expectations, citing strong growth in its cloud infrastructure business.

Google Cloud will subcontract some of its new computing deal with OpenAI to CoreWeave, according to Reuters, which is a little weird.

Salesforce no longer allows third-party AI software companies to access Slack data even if the owners of that Slack instance want them to, according to The Information, which is very lame.

Thanks for reading — see you Saturday!

This post was updated to correct the number of databases on Neon created by agents after Ghodsi provided an incorrect number during a press conference.