PlanetScale is riding the Postgres wave (while still loving MySQL)

Sam Lambert has been in the world of scaling relational databases for 20 years. He ran databases at GitHub before PlanetScale. He's seen MySQL power essentially every top 100 internet site—Facebook runs the largest MySQL deployment in the world, handling billions of queries per second across hundreds of thousands of servers.

PlanetScale is built on Vitess, the open source MySQL sharding solution YouTube created when they needed to scale beyond a single server. PlanetScale is the commercial company behind Vitess, offering a hosted platform that handles petabyte-scale workloads. But this year, they launched something new: Postgres support.

Why Postgres won (it's not technical)

Sam is refreshingly blunt: Postgres isn't technically better than MySQL. In many ways, empirically, it's worse. But it's winning anyway.

The reasons are sociological. Oracle buying MySQL and being terrible stewards. Heroku making Postgres incredibly easy to use. RDS abstracting away the operational pain. Better developer ergonomics—stricter SQL compliance, better types, the extension ecosystem.

Those extensions, though, are also how people get hacked. Two weeks before re:Invent, a well-known Postgres provider was compromised partly because of extensions. Amazon and PlanetScale validate a trusted set of extensions precisely because the flexibility that developers love is also a security risk.

Postgres is also genuinely hard to operate. MySQL was built for web scale from day one. Postgres was built to learn SQL on a single server—replication was added only about a decade ago, and it's not designed for high availability. This is why services like PlanetScale exist: to hide the operational complexity.

The $5 tier comeback

PlanetScale famously killed their free tier, which annoyed developers. Sam still gets three emails a day about it. But they recently introduced a $5 tier—non-high-availability, single node, but real compute that runs 24/7.

The insight wasn't that $39 (for their HA tier) was too expensive. It's that engineers are natural optimizers. They didn't want to pay for two replica nodes they weren't using. The $5 tier solves that. It's still profitable for PlanetScale, feels fair to developers, and growing quickly.

AI companies are growing their data fast

When asked what excites him about AI, Sam's answer is simple: the numbers. AI companies are adding 25 terabytes of data per week per cluster. PlanetScale is currently migrating one of the most valuable AI companies in the world.

Sam's philosophy is picks and shovels. PlanetScale doesn't talk about AI on their website. They don't chase trends. They just build a really good database and let the wave come to them. He points to companies that threw themselves at AI branding without substance; it doesn't land. Better to focus on the fundamentals.

The AI companies that matter, the ones building real products at scale, just need a database that works. PlanetScale is happy to be that database.

This interview was conducted at AWS re:Invent 2025.