Hetzner CX32 vs CX33 in 2026: Should You Migrate?

Short Answer

CX33 replaces CX32 for new Hetzner Cloud deployments, but it is not automatically a performance upgrade. Both tiers provide 4 shared vCPU, 8GB RAM, 80GB NVMe storage, and 20TB of included traffic. The practical difference is lifecycle: CX32 is deprecated, while CX33 is the current cost-optimized plan.

If an existing CX32 server is stable, there is no reason to rebuild it purely for the newer number. For a new server, choose CX33. For a resize or migration, benchmark first and keep a rollback path.

CX32 vs CX33 Specs

Feature	CX32	CX33
Lifecycle	Deprecated Gen2 plan	Current cost-optimized plan
vCPU	4 shared vCPU	4 shared vCPU
RAM	8 GB	8 GB
Local storage	80 GB NVMe	80 GB NVMe
Included traffic	20 TB	20 TB
CPU selection	Not selectable	Not selectable
Best decision	Keep if stable	Use for new deployments

Hetzner’s deprecated-plan reference confirms the CX32 allocation. The current Hetzner Cloud catalog lists CX33 with the same headline resource amounts.

CX33 Pricing After June 15, 2026

Hetzner’s official price adjustment lists CX33 at €8.49/month excluding VAT and IPv4 for Germany and Finland. The hourly price is €0.0136, with the monthly cap applying while the server exists.

Cost item	Germany/Finland reference
CX33 server	€8.49/month before VAT
Primary IPv6	No additional charge
Primary IPv4	Additional paid option
Billing	Hourly with a monthly cap

The exact invoice depends on location, currency, VAT, and network choices. Use the official Hetzner pricing notice as the source of truth, then verify the total in Console.

For historical CX32 pricing and workload guidance, see the updated Hetzner CX32 pricing guide.

Is CX33 Faster Than CX32?

The plan name alone does not prove a speed increase. Hetzner classifies CX as shared-resource compute and states that CX instances may run on Intel or AMD CPUs. Customers cannot choose the CPU type, and the assigned hardware may change during a rescale.

That means a useful comparison requires workload-level measurements:

1. Record API latency, requests per second, CPU steal time, and disk latency on CX32.
2. Take a snapshot or application-level backup.
3. Test CX33 in the same location with the same image and application configuration.
4. Run the benchmark several times instead of trusting one burst result.
5. Compare sustained performance and monthly cost before moving production traffic.

Shared vCPU tiers are best for web applications, bots, development environments, and bursty background work. Sustained CPU-heavy workloads should be compared with CPX or dedicated CCX plans.

Should Existing CX32 Users Migrate?

Keep CX32 when

• The server is stable and has enough CPU, memory, and disk capacity.
• Current pricing remains acceptable for the account.
• Rebuilding would introduce more operational risk than value.
• Monitoring shows no sustained resource pressure.

Move to CX33 when

• You are creating a new server rather than preserving an existing instance.
• Hetzner requires a current plan during a rescale operation.
• You are already rebuilding the stack or changing regions.
• A controlled benchmark shows equal or better results at an acceptable price.

Hetzner warns that some changes to legacy servers can move them onto current pricing. Read the pricing notice and verify the confirmation screen before applying a rescale.

Safe Migration Checklist

1. Export application configuration and secrets to a secure backup.
2. Back up databases independently; do not rely only on a VM snapshot.
3. Record DNS TTLs, firewall rules, cron jobs, and systemd services.
4. Create a snapshot or a parallel CX33 server.
5. Deploy the same application version and run health checks.
6. Benchmark CPU, disk, latency, and memory under realistic load.
7. Switch traffic only after database and application verification.
8. Keep the old instance until logs and monitoring confirm a clean cutover.

For the base Ubuntu, SSH, UFW, PM2, Nginx, and TLS steps, use the Hetzner server setup guide. The process applies to CX33 as well.

Workload Recommendations

Workload	Recommendation
One Node.js API	CX23 is often enough; benchmark before paying for CX33
Multiple Node.js apps	CX33 provides useful 8GB headroom
n8n + PostgreSQL + Redis	CX33 is a practical starting tier
Ollama 7B model	8GB is a minimum, but CPU-only inference remains slow
Sustained builds or encoding	Compare CPX or CCX instead of assuming CX33 fits

For memory-heavy AI workloads, pair this comparison with the Ollama VPS sizing guide. For Node.js provider-level performance, see the Hetzner vs DigitalOcean benchmark.

Verdict

Choose CX33 for a new 8GB Hetzner server. Keep CX32 when an existing server is stable and migration has no measurable benefit. Treat CX33 as the supported successor, not as a guaranteed speed upgrade, because both tiers use shared resources and CPU assignment varies.

The best migration decision comes from current pricing, a reproducible workload benchmark, and a tested rollback plan—not the plan number alone.

FAQ

Is Hetzner CX32 still available?

It is listed as a deprecated plan. New deployments should use CX33 where available.

What changed from CX32 to CX33?

The lifecycle and current product generation changed, while the headline allocation remains 4 shared vCPU, 8GB RAM, and 80GB NVMe.

Should I resize a working CX32 immediately?

No. Keep it if it is stable. Resize when there is a concrete operational, availability, or pricing reason and after creating a backup.

Is CX33 suitable for Ollama?

Its 8GB RAM can fit some quantized 7B-class models, but shared CPU inference is better suited to personal automation and low-concurrency APIs than interactive production chat.