Neon brings realtime queries to Postgres as Planetscale widens its cloud map

Reactive Postgres becomes real

The week's headline is Neon's push to make the database itself the reactivity layer. Neon Realtime entered public beta as a system that runs any deterministic SQL query as a live query, so a subscribed result stays current as rows change — no polling loops, no refetch logic in the client. It runs inside Lakebase Postgres rather than as a bolt-on service, which means scale-to-zero and branching apply to realtime workloads the same way they do to ordinary ones. A TypeScript SDK exposes subscribe() and useLiveQuery, with SSR support and instant local writes so optimistic UI doesn't have to wait on a round trip. Agents get in on it too, via the Neon MCP server or the neon-realtime skill.

The broader Neon changelog entry frames that launch alongside a pile of adjacent work: Realtime landing in four AWS regions, eleven new AI Gateway models, CJK full-text search in Lakebase Search, and Managed Better Auth moving to Better Auth 1.7.6. Read together, the theme is Neon positioning itself less as a Postgres host and more as the backend layer an app talks to directly.

Stripe data lands without a pipeline to build

The second half of that story is payments data. Stripe Data Pipeline added real-time sync to Neon Postgres, so Stripe streams API data straight into a Neon database with Stripe managing backfill and ongoing delivery. No custom webhook handlers, no rate-limit bookkeeping, and Neon is now selectable as a connectable destination from the Stripe Dashboard. It's in public preview and free to start. For anyone who has written subscription-state reconciliation code, the appeal is obvious: the upgrade and cancellation events your app needs are simply rows you can query.

PlanetScale's geography and plumbing

PlanetScale's most consequential move was adding Microsoft Azure as a cloud provider for Postgres and Neki databases, with the same feature set as AWS and Google Cloud — high availability, branching, Query Insights, automated backups, and Metal clusters. Vitess databases remain unavailable on Azure, so the expansion is specifically about the Postgres side of the house.

The rest of the week was infrastructure ergonomics. The Terraform provider gained Neki router parameter management, letting router settings live in Terraform and be read back through the router data source — though configuring the block resets any omitted customer-configurable parameters to defaults, so it's worth reading the caveats. The same provider release also stops persistent drift from locked Neki parameters and lets the databases data source filter by kind. Postgres bouncer managers should upgrade before PgBouncer timeout parameters start reporting their type as time instead of seconds.

On the deploy side, aggressive cutover became a per-deploy-request flag in the CLI and API, settable any time before the schema change ships, with a new column in deploy-request show and queue. The CLI itself now flags parameters that require an update first in parameter listings, for both Postgres branches and Neki configuration profiles, admins, and sidecars, and mirrors that to JSON as requires_update. The same capability lands in the CLI's own changelog entry and in the CLI release that introduced per-deploy-request aggressive cutover, whose refresh-schema command now waits for the schema snapshot and returns poll failures instead of a misleading refresh timeout.

Tooling churn

Supabase's CLI spent the week almost entirely in pre-release: a long run of 2.121.0 betas and 2.121.1 betas, plus next-major and beta builds. The one stable cut, Supabase CLI's stack and deprecation work, makes stack destroy repeatable across multiple stack IDs, exposes OrioleDB version through config, and begins warning at runtime about deprecated flags, config, env vars, and import maps ahead of the next major. It also bundles stack reliability fixes around host CA bundles, gateway writes under load, and database helper leaks. Client-side, Supabase JS added storage upload content-encoding along with postgrest-typegen output that keeps computed fields out of select('*') and resolves no-argument rpc overloads.

Prisma's week leaned toward preview and internal work: a release candidate pushed raw SQL into literals, multi-field weighted full-text indexes, and PSL mixins, while Prisma Studio surfaced Migrations as a standalone recorded-schema component and began remembering manual table layouts in the visualizer. Turso's prebuilt CLI binaries round out a quiet week of packaging.