Databricks has made a significant move in the evolving landscape of database technology by acquiring Electric, a startup renowned for its WASM-based Postgres project, PGlite, and the innovative Electric sync engine. This acquisition comes at a time when agentic applications are reshaping how developers interact with databases.
The Electric team will integrate with Neon, a serverless Postgres company that Databricks acquired for approximately billion last year. Neon serves as the backbone of Databricks’ Lakebase database service. While the financial details of the acquisition remain undisclosed, the strategic implications are clear.
What Databricks bought
PGlite represents a full Postgres database implemented in WebAssembly (WASM), capable of running in various environments, including browsers and Node.js processes. It allows for dynamic extension loading, notably supporting pgvector, which is a favored extension for Postgres vector operations. The popularity of PGlite is evident, with its weekly downloads skyrocketing from 1 million to an impressive 13 million over the past year.
The sync engine at the core of Electric
At the heart of Databricks’ interest in Electric lies the Electric sync engine. This technology enables a central Postgres database to be synchronized in near real-time across various platforms, including browser tabs, mobile applications, and agents. This functionality mirrors the collaborative models seen in tools like Figma and Google Docs, but tailored for Postgres and its associated agents.
The Neon team has highlighted the complexities involved in real-time synchronization, such as conflict resolution and partial replication, which can make building such systems from scratch a daunting task. This context likely influenced Databricks’ decision to acquire Electric rather than developing a similar solution independently.
Looking ahead, Electric’s founders, James Arthur and Valter Balegas, have assured users that all previously open-sourced components, including the sync engine, PGlite, Durable Streams, and TanStack DB, will remain accessible. However, the hosted service, Electric Cloud, will be phased out, requiring existing users to either self-host or transition to alternative providers.
This acquisition is part of a broader trend of database-related purchases by Databricks, which includes the acquisition of Neon and, more recently, the transactional processing startup Mooncake.
A database that lives for 10 seconds
Databricks posits that traditional non-agentic applications typically share a single database among multiple clients, which tends to be a permanent fixture within the tech stack. However, the emergence of agent workloads is altering this paradigm.
In a recent analysis, Databricks’ Ippokratis Pandis, Nikita Shamgunov, and Reynold Xin noted that agents are now responsible for creating approximately four times more databases than human users on Lakebase. They also pointed out that the average project currently encompasses around 10 database branches, with some projects extending to over 500 iterations.
Interestingly, for certain applications on Lakebase, the average lifespan of a database compute instance is now less than 10 seconds. This rapid turnover reflects agents’ propensity to branch databases similarly to how they branch code, a concept that Neon has built into its architecture.
In practice, a coding agent can quickly spin up a sandbox, instantiate PGlite, conduct builds and tests against the database, and then either discard the entire setup or sync the outcomes with a corresponding Lakebase branch. The architecture of Lakebase, which separates storage from compute and maintains data in Postgres page formats on object storage, facilitates this process as a cost-effective copy-on-write operation.
As the Neon team aptly notes, as coding agents drive the cost of creation toward zero, the number of applications proliferates, with many being relatively small. A database server, even a serverless variant that can scale down to zero, imposes a minimum operational cost, which can hinder the proliferation of small applications. The argument is made that an era of abundance in application development cannot thrive if every app necessitates a fixed baseline of compute resources.
‘Two halves of the same idea’
It is noteworthy that PGlite’s origins trace back to an experiment by Stas Kelvich, co-founder of Neon, who initially compiled Postgres to WASM to explore its client-side capabilities. Electric later adopted this initiative, transforming it into a viable production project. As articulated in Databricks’ announcement, this acquisition effectively reunites two halves of the same innovative concept.
