Does Postgres Scale?

The question appeared on April 30th with the economy of something long overdue: "Does Postgres Scale?" Five words carrying roughly fifteen years of accumulated tension. Postgres spent the better part of the 2010s quietly annexing territory once held by MySQL, Oracle, and a rotating cast of NoSQL insurgents that promised horizontal scale and delivered operational complexity instead. It became the sensible default — the database you chose when you wanted to be right without having to explain yourself. Extensions multiplied. TimescaleDB claimed the time-series workload. pgvector arrived just in time to absorb the AI wave. Citus offered sharding without a full rewrite. Each addition seemed to answer the scale question before anyone could ask it too loudly. Then someone asked it anyway. The Hacker News thread is the kind that accumulates slowly and then all at once — engineers sharing war stories about read replicas that drifted, connection pools that became their own full-time maintenance burden, and the particular dread of watching query latency climb on a table with two billion rows. The counterarguments arrive in the same thread with equal conviction: Discord runs Postgres at extraordinary scale, Notion rebuilt around it, Stripe trusts it with the kind of transaction volumes that would make most database administrators quietly religious. The debate is not resolved. It is, if anything, more alive than it was at the start of April.

If confirmed as a genuine inflection point in how the industry thinks about Postgres, here is what this means. The companies building on Postgres today — and there are many, because vibe coding tools, AI application frameworks, and low-code platforms have all defaulted to it as the path of least resistance — face a decision they may have been deferring. Vertical scaling has a ceiling. Connection pooling via PgBouncer or Supabase's infrastructure is a pressure valve, not a solution. The emergence of pgvector as the default vector store for AI applications means that Postgres is now absorbing workloads it was never designed for, at volumes that are growing faster than the application teams managing them. The second-order effect is a probable surge in demand for managed Postgres services — Neon, Supabase, Aurora — that abstract away the scaling complexity at a cost that may only become visible later. And for the vibe coding generation building production systems without deep database intuition, that cost may arrive as a surprise.

Watch for engineering posts from high-scale Postgres users — particularly in the AI application layer — detailing either successful scale strategies or quiet migrations away. A credible case study in either direction would settle at least part of this argument.

NewsHive monitors these sources continuously. All signal titles above link to the original reporting.

Intelligence by NewsHive. Need help navigating what this means for your business? Contact GeekyBee →