Comparison · different shapes of GPU cloud

AxForge vs RunPod

Q: Where does AxForge process data?

In the EU only: eu-se-1 (Stockholm) and eu-es-1 (Málaga) are live, with more EU regions in deployment. The region is pinned per API key and echoed on every response, and prompts are processed with zero retention.

RunPod is a large self-serve GPU cloud built for instant scale; AxForge is a small EU operator renting named machines by the month, with a zero-retention hosted model API.

eu-se-1 · Stockholm eu-es-1 · Málaga DGX Spark — available now

Get an API key Talk to an engineer Keys are allocated from the waiting list.

At a glance

Self-serve scale vs named machines

Dimension	AxForge	RunPod
What it is	A small EU infrastructure operator: dedicated machines plus a hosted open-model API, on hardware AxForge owns.	A large self-serve GPU cloud: rent GPU capacity on demand, at scale.
How you get compute	Request a system and talk to an engineer; API keys are allocated from a waiting list. Engineer-led onboarding.	Sign up and launch in minutes, fully self-serve.
GPU menu	Short and named: NVIDIA DGX Spark (GB10, 128 GB unified memory) available now; RTX 6000 Pro, RTX 5090 and RTX 3090 tiers on the way.	A very large menu of GPU types across on-demand pods and serverless workers.
Billing model	Monthly rental for dedicated machines; per-token billing on the hosted API.	Per-second, on-demand billing — spin up and down as you need.
Hosted inference	A managed OpenAI-compatible model API — Qwen3.8 27B served now, plus embeddings, image and speech — with zero prompt retention.	Bring your own containers and models; serverless workers scale them for you.
EU data residency guarantee	EU-only by construction: eu-se-1 Stockholm and eu-es-1 Málaga live, region pinned per key and echoed on the response.	Datacenters in many regions worldwide; you choose where a workload runs.
Hardware transparency	Named, owned machines with published serving details and performance we measured ourselves.	You pick a GPU SKU from the menu; the platform operates the fleet.

Honest routing

When RunPod is the better choice

A large self-serve GPU cloud wins on these requirements:

Compute in minutes	You need a GPU now, self-serve, with no conversation and no waiting list.
Burst capacity	Spiky training or batch workloads where per-second billing and instant scale-up beat a monthly commitment.
A specific GPU SKU	You want a particular card from a very large menu of GPU types.
Serverless workers	You want your own containers scaled up and down automatically, including to zero.

Honest routing

When AxForge is the better choice

EU sovereignty requirements	Processing must stay in the EU and you must be able to prove it — named machine, named region, pinned per key, echoed on every response.
Predictable dedicated capacity	One named machine, your traffic only, a flat monthly rental — no bidding, no preemption, no noisy neighbours.
A hosted model API	You want open-model inference as a managed API with zero prompt retention, instead of operating your own serving stack.
Auditability	You need to tell an auditor exactly what hardware ran the workload and how it was configured — and get answers from the engineers who run it.

The substance

What AxForge actually runs

Regions	eu-se-1 · Stockholm and eu-es-1 · Málaga, both live. More EU regions in deployment.
Data handling	Zero prompt retention — in-memory processing in Sweden, no disk, no logs, no training. Only billing metadata is kept. Full policy: axforge.ai/privacy.
API	OpenAI-compatible at https://api.axforge.ai/v1 — TLS 1.3, terminated in the EU.
Pricing	Qwen3.8 27B: €0.29 / 1M input · €1.77 / 1M output (launch pricing). GPU rentals: €495 / month for a dedicated DGX Spark — full price list published.
Measured speed	15.9 tokens/s single-stream decode for Qwen3.8 27B with speculative decoding.
Dedicated systems	NVIDIA DGX Spark (GB10, 128 GB unified memory), monthly rental, available now in Sweden. RTX tiers on the way.

Speed measured on our production DGX Spark node, single-stream, 2026-08. We publish only numbers we measured ourselves.

FAQ

AxForge vs RunPod — common questions

Is AxForge a RunPod alternative in Europe?

For EU sovereignty requirements and predictable dedicated capacity, yes: AxForge rents named machines monthly from EU regions and runs a zero-retention model API. For instant self-serve scale, burst workloads and a huge GPU menu, a large GPU cloud like RunPod is the better fit.

Can I rent a GPU machine from AxForge today?

Yes. NVIDIA DGX Spark systems (GB10, 128 GB unified memory) are available now as monthly rentals in Sweden (eu-se-1). RTX 6000 Pro, RTX 5090 and RTX 3090 tiers are on the way. Onboarding is engineer-led — you talk to the people running the hardware.

Does AxForge have per-second billing like RunPod?

No. Dedicated machines are monthly rentals, and the hosted model API is billed per token. If you need per-second billing and scale-to-zero, a self-serve GPU cloud is the right tool.

What does the AxForge model API cost?

Qwen3.8 27B is €0.29 per million input tokens and €1.77 per million output tokens (launch pricing). GPU rentals carry published launch pricing too — a dedicated DGX Spark is €495 per month, and the full price list is public. Talk to an engineer to scope a system.

Where does AxForge process data?

In the EU only: eu-se-1 (Stockholm) and eu-es-1 (Málaga) are live, with more EU regions in deployment. The region is pinned per API key and echoed on every response, and prompts are processed with zero retention.