Debits the order total from the account's USD balance and creates the server. NO crypto field — pre-fund via POST /api/v1/topup first. If the balance is short, returns 402 insufficient_balance with a topup_required block (suggested amount covers deficit + clears $30 minimum). Server is active immediately on success. Bearer optional — if absent, a new account is auto-created (it will have $0 balance → 402 → topup → retry).

type: vps, dedicated, rdp, or gpu. For rdp, os must be a Windows slug (win-2022, win-2025, win-2019, win-11); RDP locations: netherlands, romania, moldova, russia. For gpu, os must be a Linux + CUDA image (e.g. ubuntu-22-cuda); GPU locations: iceland, netherlands, romania, moldova.

GPU-only fields (ignored for other types):

• ai_stacks (string[]): pre-install AI software at deploy time. Slugs: vllm, ollama-webui, text-generation-webui, tgi, axolotl, unsloth, llama-factory, comfyui, a1111, forge, kohya-ss, comfyui-video, whisper-server, jupyterlab (always on), code-server.
• pretrained_models (string[]): pre-download HuggingFace models. Slugs: llama-3.3-8b, llama-3.3-70b, qwen3-32b, deepseek-r1, mistral-small-3, gemma-3-27b, phi-4, flux-dev, flux-schnell, sd-3.5-large, sdxl-base, wan-2.1-t2v, cogvideox-5b, whisper-large-v3, bark.
• hf_token (string): HuggingFace token, required to pre-download gated models (Llama, Mistral, Gemma, FLUX-dev, SD 3.5). AES-encrypted at rest, used once at provision.
• ssh_public_key (string): SSH public key (ssh-ed25519, ssh-rsa, ecdsa-*). Optional — for password-less login.
• public_endpoint (boolean): auto-provision Let's Encrypt + nginx reverse proxy on port 443.
• auto_shutdown_hours (integer): one of 0 (never) / 6 / 12 / 24 / 48 / 168. Server self-destructs after the timer.