Cloudflare says its serverless database D1 now manages more than 5.5 million individual databases, and getting there required rebuilding the control plane, not just adding bigger servers.
D1 is Cloudflare's SQLite-based serverless database, where each database runs on its own Durable Object, making it cheap to spin up at scale. In a technical post, Cloudflare detailed how it scaled D1's control plane as usage exploded: live databases grew 15x over the past year and 65x over two years, with daily active databases up 18x, as of October 8, 2026. The fix was architectural separation - query traffic ("data plane") stays at the edge, while slower management operations like creation and deletion ("control plane") can touch central Postgres clusters through a connection-pooling layer called Hyperdrive. Cloudflare also leans on Workers KV, a distributed key-value store, to cache routing metadata so most requests never reach a central database at all.
The real story here is what's driving the growth. AI-assisted coding and vibe-coding platforms are pushing developers away from one shared database per app toward a database per user, per tenant, or per preview environment - multiplying database counts by orders of magnitude. Centralized architectures built for a handful of big clusters buckle under that pattern. Cloudflare's answer borrows the same playbook serverless compute used to scale: push metadata to the edge, keep any single-region component off the hot path.
Cloudflare cites OpenAI's ChatGPT Sites as a customer using D1 this way, for things like leaderboards and comments - a reminder that the "many small databases" pattern is becoming the default for AI-built apps, not an edge case.