Supabase Pricing 2026: Free Tier Limits, Compute Costs & When to Upgrade

Supabase pricing confuses a lot of developers because the free tier is genuinely useful — until it isn't, and the jump to paid happens for reasons that aren't obvious upfront. This post breaks down every tier, the real limits that trigger an upgrade, and how the costs stack up once you're running a production workload.

What the free tier actually gives you

The free tier is not a trial. You can run a real project on it indefinitely, with these hard limits as of September 2026:

The biggest catch: free projects pause after a week of inactivity. Supabase spins down the Postgres instance, and a paused project has to be restored before it serves requests again. For a portfolio site or internal tool that gets occasional traffic this is annoying; for a production app with real users it's a deal-breaker.

The second catch is the 2-project limit. If you're a freelancer juggling multiple clients on a single organisation account, you hit this fast. Creating separate organisations per client isn't a reliable workaround — check Supabase's current rules on how the free-project limit is counted across organisations — so in practice client work usually means each client owning their own organisation, or a paid plan.

Pro plan: $25 / month per project

The Pro plan removes pausing and lifts most limits:

At $25 the Pro plan is competitive for a small production app — but note it's "from $25": the included $10 compute credit covers one Micro instance, and each additional project or larger instance adds compute cost. The Micro compute instance is adequate for most CRUD workloads under moderate concurrency. Where it breaks down: connection limits. Supabase provides a connection pooler (Supavisor, with a dedicated PgBouncer option on paid plans), but if you're running a Next.js App Router site with lots of server components firing parallel queries, you can saturate Micro's limits (60 direct connections and 200 pooler clients by default).

Compute add-ons: where bills actually grow

Compute is billed separately, hourly, and is the most common surprise in real invoices. Paid plans include $10/month in compute credits, which covers one Micro instance; anything beyond that is billed on top.

From Supabase's compute docs at the time of writing. The $10 credit applies once per organisation, so on a single project the net cost of Small is ~$5/month and Medium ~$50/month.

For a typical Next.js + Supabase content site, Micro handles maybe 50–100 concurrent users comfortably. A SaaS app with background jobs or heavy joins will want Small or Medium. The step from Micro to Small (~$5/month net after the credit) is a common first upgrade — Micro's 1 GB of RAM and shared CPU can cause latency spikes under sustained load.

Team plan: $599 / month

The Team plan is aimed at startups with a dev team rather than a solo developer. It adds:

• SSO for the Supabase dashboard
• Priority email support with SLAs
• 28-day log retention and 14-day backups
• Platform audit logs
• SOC 2 and ISO 27001 reports
• HIPAA as a paid add-on

The $599/month is an organisation-level fee; projects within it still pay for their own compute and usage, as on Pro. Unless you genuinely need SOC 2 docs for a sales process or dashboard SSO, most teams stay on Pro with a Medium compute add-on and come out significantly cheaper. (PITR is available on Pro too, as an add-on.)

Point-in-time recovery costs

PITR is an add-on available on Pro and above. It lets you restore to any second within the retention window, not just daily snapshots.

Supabase prices PITR at $100/month per 7 days of retention.

For a fintech or healthcare product, PITR is non-negotiable. For a marketing site or blog, daily backups (included in Pro) are enough.

What actually forces the upgrade from free to Pro

In practice, the triggers are predictable:

1. Project pausing — the moment a client's staging URL starts cold-starting, you move to Pro.
2. Database exceeding 500 MB — happens faster than expected if you're storing files as bytea columns (don't) or running analytics queries without pagination.
3. Bandwidth over 5 GB — one viral blog post or a client uploading a batch of product images can blow past this in hours.
4. More than 2 active free projects — forces a paid plan (or moving projects to the client's own account).
5. Auth MAUs over 50,000 — unlikely for most projects but it does bite consumer apps that grow unexpectedly.

The free-to-Pro jump at $25 is low enough that moving production workloads to Pro from day one is the sensible default. The pausing behaviour alone makes free unsuitable for anything customer-facing.

How Supabase compares to PlanetScale and Neon

If you're evaluating Supabase purely as a Postgres host rather than a full backend-as-a-service:

• Neon offers a free tier that scales compute to zero after a few minutes of inactivity and wakes on the next query (no week-long pause), and its paid plans are usage-based with branching — better for development workflows. Check Neon's pricing page for current rates.
• PlanetScale dropped its free Hobby tier in 2024 and now offers both MySQL (Vitess) and Postgres; check its pricing page for the current entry price.

If you need Auth, Storage, Edge Functions, and Realtime bundled, Supabase Pro at $25 beats building equivalent infrastructure on any of those platforms. If you only need Postgres, Neon's pricing is more competitive at the low end.

Realistic monthly cost for a production Next.js + Supabase app

A small SaaS with 500 MAUs, 3 GB database, 20 GB bandwidth, and moderate query load:

• Pro plan: $25
• Compute (Small): ~$15, minus the $10 credit = +$5
• Storage overage: $0 (within 100 GB)
• Bandwidth overage: $0 (within 250 GB)
• Total: ~$30 / month

A growing SaaS with 5,000 MAUs, 15 GB database, 180 GB bandwidth, and sustained concurrency:

• Pro plan: $25
• Database overage: (15 GB − 8 GB) × $0.125 = $0.875
• Compute (Medium): ~$60, minus the $10 credit = +$50
• Bandwidth: within limit
• Total: ~$76 / month

Those numbers are predictable. The only time Supabase invoices surprise people is when Edge Function invocations spike from a misconfigured webhook loop or when storage egress accumulates from serving large files directly through Supabase Storage instead of routing through a CDN like Cloudflare.