13M Downloads a Week. PGlite Now Belongs to Databricks.

TL;DR

On August 11, Databricks announced it is acquiring Electric, the small team behind PGlite, the WebAssembly build of Postgres that went from 1 million to 13 million weekly downloads in twelve months. Two days later Databricks closed a $5 billion round at a $190 billion valuation and named Lakebase, its Postgres for agents, as one of three products the money is earmarked for. Everything Electric previously open sourced stays open source. Electric Cloud, the hosted service, is winding down, and if you were running on it you now own a migration.

What Databricks actually bought

PGlite is Postgres compiled to WebAssembly: a real query planner, real SQL, real extension loading for things like pgvector and PostGIS, in a bundle under 3MB gzipped. It runs in a browser tab, in Node, in a container, in whatever sandbox your agent happens to occupy. No server, no connection pool, no network hop. It is dual licensed under Apache 2.0 and the PostgreSQL License.

The other half of Electric is a sync engine that streams changes between those embedded copies and a central Postgres. Databricks says PGlite grew from 1M to 13M weekly downloads in twelve months, which is the number that turns a side project into an acquisition target.

There is a neat piece of circularity here. PGlite started as a proof of concept by Stas Kelvich, a co-founder of Neon, the serverless Postgres company Databricks bought for about $1 billion in May 2025 and turned into Lakebase. Electric took the experiment and made it a product. Databricks has now bought both ends of the same idea, and Electric is joining the Neon team inside Databricks.

Why an agent wants its own database

The pitch is that agents are a bad fit for the shared-database model. An agent does a lot of small reads and writes at machine speed, needs to scribble intermediate state somewhere, and periodically needs to throw all of that away because the step it just took was wrong. Doing that against a central production Postgres means every scratch write is a network round trip, and every mistake is a cleanup job.

Think of it as the difference between thirty contractors marking up one shared blueprint on one table, and each getting a photocopy to scribble on while a foreman reconciles the marks at close of day. The local copy is where the mess happens; the central copy stays coherent.

Mike Leone of Moor Insights told TechTarget that Electric "shrank Postgres down small enough to run beside the agent instead of across a network," and that the local copy lets an agent undo a step cleanly instead of leaving a half-finished mess. That second part is the underrated one. Rollback is a much bigger deal for autonomous agents than latency is.

Lakebase already leans on this vocabulary. Its product page describes serverless Postgres that "branches like code," with instant branches for development and testing and compute that scales to zero when idle. PGlite pushes the same branch metaphor one step further out, into the sandbox itself.

The competitive read

Devin Pratt of IDC gave TechTarget the sharpest framing of the deal: "Snowflake bought its Neon, [but] nobody's bought their Electric yet. That's the gap Databricks just closed on itself." Read that as a land grab on the agent-side data layer while the big clouds are still selling Postgres for ordinary workloads.

The part that costs you something

Electric co-founders James Arthur and Valter Balegas wrote the announcement themselves, and it is admirably direct. Everything Electric previously open sourced stays open source: Postgres Sync, PGlite, TanStack DB, and Durable Streams. That is the sentence the 13 million weekly downloads needed to hear.

Then the other sentence: Electric Cloud is winding down. Cloud users have to self-host or move to another provider. Electric says it contacted every existing cloud user directly and will support the transition, but the post gives no sunset date. If you are on it, your database just became strategic infrastructure for somebody else and your hosting became a migration ticket.

Terms of the acquisition were not disclosed.

The $5 billion behind it

Two days after the Electric post, Databricks announced a $5 billion round at a $190 billion valuation, led by Coatue with Blackstone, MGX, accounts advised by T. Rowe Price, new investor Sixth Street Growth, and returning backers including BOND, Clearlake, Point72, Premji Invest and TPG.

The numbers in the press release, all annualized run-rate as of Q2 2026:

• Over $7 billion total run-rate, growing more than 80% year over year.
• Over $1.5 billion for the lakehouse data warehousing business, growing more than 100% year over year.
• Over $100 million for Lakebase, which launched in June 2025.
• More than 1,000 customers above $1M run-rate, more than 100 above $10M.
• Positive adjusted free cash flow over the trailing twelve months.

The money is earmarked for three things: Lakebase, Genie (querying business data in natural language), and Unity AI Gateway (a model catalog with usage and cost controls). Two of the three are agent infrastructure.

CEO Ali Ghodsi told TechCrunch the company set out to raise $1 billion, a report leaked mid-conference, and the phone did not stop: "The interest level was just insane. Just from this select group of investors that we looked at, there was $15 billion of interest." They took $5 billion. It is a nice problem to have.

What this means if you build agents

If you already use PGlite, nothing breaks today. The license is unchanged, the repos stay open, and the maintainers are now paid by a company with $5 billion of fresh cash and a direct product reason to keep the project healthy. That is usually the good version of an acquisition.

If you are on Electric Cloud, start your migration planning now rather than when a date appears. Self-hosting the sync engine is the documented path, and it is open source, so this is a scheduling problem rather than a rewrite.

The wider signal is architectural. Two Postgres companies in fifteen months, both aimed at agent state, is a bet that the interesting database problem in 2026 is not scale but isolation: giving each of a hundred concurrent agents a place to be wrong in private before anything lands in the system of record.

What is not confirmed

• No purchase price. No Electric Cloud shutdown date.
• No shipping timeline for PGlite-in-Lakebase as a product, and no published latency or throughput numbers for the combined stack.
• Governance beyond licensing is unstated. "Stays open source" is a license commitment, not a promise about who merges pull requests in 2028.

Key Takeaways

• Databricks is acquiring Electric, maker of PGlite, the sub-3MB WebAssembly Postgres that hit 13M weekly downloads, up from 1M a year ago. Terms undisclosed.
• PGlite, Postgres Sync, TanStack DB and Durable Streams stay open source. Electric Cloud is winding down with no announced sunset date, so hosted users must self-host or migrate.
• The plan is a local Postgres inside each agent sandbox, synced by Electric back to Lakebase, Databricks' serverless Postgres built on its ~$1 billion Neon acquisition from May 2025.
• Two days later Databricks closed $5 billion at a $190 billion valuation, led by Coatue, on a reported $7B+ run-rate growing more than 80% year over year with positive adjusted free cash flow.
• Lakebase is past $100M run-rate, about 1.4% of revenue, and is one of three products the raise explicitly funds alongside Genie and Unity AI Gateway.
• The bet worth watching is isolation over scale: per-agent database copies with clean rollback, not a bigger shared cluster.

Sources: Databricks blog: Electric joins Databricks, Electric: Electric is joining Databricks, Databricks press release: over 80% YoY growth, $7B run-rate, TechCrunch, TechTarget, PGlite, Databricks Lakebase, CNBC on the Neon acquisition