gpu-io.net Request access

Bare-metal Apple Silicon in Sweden

Dedicated Mac Studio and Mac mini capacity for AI inference, coding agents, CI/CD, model testing, and private workloads. You rent the whole machine, not a slice of someone else's queue.

Reserve a Mac View machines
M3 Ultra 512 GB Bare-metal hosting Inference API
Stockholm edge ready
Mac Studio M3 Ultra 512 GB 80-core GPU · 819 GB/s
Mac Studio M4 Max 128 GB 40-core GPU · Thunderbolt 5
Customer SSH / VNC / API
Sweden Dedicated host
Exit Clean routing

Choose the right entry point

Most customers should start with one dedicated machine. API access and multi-Mac cluster work are added only when the workload needs it.

Highest memory M3 Ultra 512 GB rental For large MLX/GGUF models and coding agents. Primary offer Bare-metal Mac hosting Dedicated Apple Silicon in Sweden with SSH/VNC. Trust boundary Security and reset model Single tenant, owner recovery, and reset evidence.
Single tenant No shared workers or noisy neighbours
Root access SSH and remote desktop on dedicated machines
Swedish location Operated from Stockholm-area infrastructure
Model flexible MLX, GGUF, Ollama, LiteLLM, or custom setup

Built for direct rental, not anonymous sharing

gpu-io.net starts with a simpler security promise: one owned Mac, one customer, a documented reset, and Swedish-operated access paths.

One customer per machine

No shared queues or hidden multi-tenant workers during a bare-metal rental window.

Break-glass recovery

Owner SSH, VNC, and ARD recovery are kept separate from customer accounts for reset and abuse response.

Reset evidence

Customer users, SSH keys, caches, and temporary data are removed between rentals and checked by audit scripts.

Network isolation

Rental hosts are separated from management networks. Cluster traffic is enabled only for assigned customer packages.

Available machines

The first product is intentionally simple: rent a complete Apple Silicon machine with predictable access. Inference API is available when you do not need root.

Mac Studio M3 Ultra

Limited 512 GB fleet

$990 / month

Unified memory
512 GB
GPU
80-core
CPU
32-core
Storage
2 TB SSD
Best for
Large local models, coding agents, private inference
Request this Mac

Mac Studio M4 Max

Fast agent and CI host

$400 / month

Unified memory
128 GB
GPU
40-core
CPU
16-core
Storage
1 TB SSD
Best for
CI/CD, MLX inference, app builds, testing
Request this Mac

What customers get

gpu-io.net is not a decentralized network. It is a small, operated fleet of owned Macs with clear access boundaries.

01

Dedicated access

Each machine is assigned to one customer during the rental window. SSH and remote desktop are prepared before handoff.

02

Clean reset

Rental hosts use separate runtime accounts, break-glass admin, launchd recovery, and a documented wipe/reset process between customers.

03

Swedish routing

Traffic can be routed through controlled exit nodes when needed. Bare-metal customers can also bring their own VPN stack.

04

AI-ready setup

MLX, Ollama, LiteLLM, Python, Node, Xcode tools, model cache, and monitoring can be installed by request.

Inference API when bare metal is too much

For smaller jobs, gpu-io.net can expose an OpenAI-compatible endpoint backed by local Apple Silicon. Model availability depends on license, fit, and customer workload.

  • Third-party models remain subject to their own licenses.
  • Commercial-use checks are done before public API listing.
  • Medical, legal, and other high-risk uses require separate review.
curl https://api.gpu-io.net/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen3.5-27b",
    "messages": [
      {"role": "user", "content": "Build a launch plan"}
    ],
    "stream": true
  }'

Initial hosted model menu

Conservative defaults for Apple Silicon. Bare-metal customers can run their own stack.

Model Use Fit Status
Qwen3.5 122B Best reasoning on 512 GB M3 Ultra Private beta
MiniMax M2.5 Coding and agent workflows M3 Ultra License review
Qwen3.5 27B Fast general work M4 Max / M3 Ultra Candidate
MedGemma 27B Medical text experiments M4 Max / M3 Ultra Restricted use

Tell us what you want to run

Start with the workload, model, access level, and rental window. We will answer with the safest machine setup instead of over-selling a cluster.

Email hello@gpu-io.net