Why We Switched from Prisma to Drizzle — Type-Safe SQL Without the ORM Tax

TL;DR: Prisma’s hidden N+1 queries, 2.3MB client bundle, and opaque query plans cost us 3 production incidents in 6 months. Drizzle 0.33 gave us explicit SQL, 900KB bundle, and zero surprises — but migration took 2 weeks and we lost prisma studio for non-technical stakeholders. Full comparison →

The Context

Three-dev team, Next.js 14 + PostgreSQL (Supabase) + tRPC. Prisma 5.x since Jan 2025. Pain points: (1) N+1 in include: { posts: { include: { author: true }}} on dashboard — 47 queries for 20 users, 2.1s response. (2) prisma migrate deploy timeout on 500k-row table (lock contention). (3) Bundle analysis: Prisma Client 2.3MB gzipped (18% of client bundle). (4) Type-safe but query-unsafe — runtime surprises.

What We Tested

Tool	Use Case	Verdict	Why
Drizzle ORM 0.33	Full repo migration (8 models, 47 relations)	✅	Explicit joins, 900KB bundle, query plan visible in code, zero N+1 since
Prisma 5.18 (latest)	Same queries, optimized with select	⚠️	Can fix N+1 but requires deep query plan knowledge; bundle unchanged
Kysely	Type-safe SQL builder alternative	✅	Closer to raw SQL, excellent types, but no migration tool (use golang-migrate)
Prisma + ZenStack	Auth/RLS layer on top	⚠️	Adds complexity; still Prisma underneath

The Pivot Point

March 2026: Dashboard API p99 2.1s. EXPLAIN ANALYZE showed 47 queries from single findMany with nested includes. Prisma’s query event logging confirmed N+1. Fix: hand-written SQL in $queryRaw — defeated Prisma’s purpose. April: prisma migrate deploy hung 18 min on 500k-row ActivityLog table (adding index). Supabase advisory lock contention. June: Decided to migrate. 2 weeks: drizzle-kit introspect:pg → generated schemas → rewrote 247 tRPC procedures → updated tests. Pain: prisma studio gone — product manager used it for quick data checks. Workaround: built admin panel with drizzle-orm + tanstack-table.

What We Use Now

Drizzle ORM 0.33 as primary database layer:

• Schema: db/schema/*.ts — single source of truth, TypeScript types inferred
• Queries: db.select().from(users).innerJoin(posts, eq(users.id, posts.authorId)).where(...) — explicit, readable, no surprises
• Migrations: drizzle-kit generate:pg → drizzle-kit migrate:pg — SQL files in drizzle/, reviewable in PR
• Bundle: 900KB gzipped (was 2.3MB) — 60% reduction
• CI: pnpm drizzle-kit check validates schema against Supabase staging

Drizzle Studio (experimental) for internal data browsing:

• Runs npx drizzle-kit studio — local only, no auth, not for stakeholders
• Replaced PM’s prisma studio use with custom admin panel (Next.js + TanStack Table + Drizzle)

When You’d Choose Differently

• Prisma if: team includes non-SQL-fluent devs, need prisma studio for PM/ops, rely on Prisma’s middleware/extensions (soft deletes, multi-tenancy), or have massive existing Prisma codebase where migration ROI is negative.
• Kysely if: you want raw SQL control with TypeScript types, don’t need a migration tool (bring your own), and prefer query builder over ORM patterns.
• Raw pg + Zod if: maximum performance, minimal deps, team writes SQL daily — but lose type-safe schema evolution.

Tool Crucible Rating

Dimension	Rating (1–5)	Notes
Overall	4	Query predictability + bundle size win; migration cost + studio loss are real
Ease of Use	3	Learning curve: SQL knowledge required; drizzle-kit CLI is excellent
Value	5	Free, MIT, no runtime overhead, replaces Prisma’s $0 but with better DX
Support	4	Active Discord, weekly releases, Supabase/Neon/PlanetScale first-class

---

This is part of our Database Tooling series. See full comparison: Drizzle vs Prisma vs Kysely: Type-Safe Database Access in 2026