Apple discontinued the 512 GB config — still rentable here

Run your AI agents on a
dedicated Mac Studio
for 7 or 14 days.

Keep Claude Code, Codex, Ollama, and MLX running around the clock — without leaving your own Mac on. A dedicated Apple Silicon machine for local LLMs, AI agents, and remote development. The Mac is hosted in Kobe, Japan — nothing is shipped to you. Pay once, get a secure access page, connect over Tailscale.

The 512 GB unified-memory configuration is no longer sold by Apple. Here you can rent one — a whole machine, yours alone, with connection details on your access page right after checkout.

  • Remote-only — nothing shipsThe Mac is hosted in Kobe, Japan. You connect from your own computer.
  • From $149 / 7 daysOne-time payment. Your rental starts immediately on payment. No subscription, no auto-renewal.
  • Encrypted via TailscaleDirect WireGuard tunnel — no open inbound ports.
  • SSH / VNC / Screen SharingVS Code Remote SSH works out of the box.
Mac Studio M3 Ultra
MEMORY512GB
BANDWIDTH819GB/s
GPU80cores
SSD4TB
ssh kizai@mac-ultra-01.ts.netonline
How it works

From checkout to SSH, in three steps.

Nothing is shipped and there is nothing to set up on our side — provisioning is instant and automated.

STEP 011
512 GB · 7 daysUSD
$499one-time
Pay

Pick a plan and pay

7 or 14 days, 96 GB or 512 GB. Stripe Checkout — no account creation before purchase. The moment payment completes, one machine is reserved for you alone.

STEP 022
ACCESS PAGE
SSH
ssh kizai@mac-ultra-01
Tailscale
https://login.tailscale.com/…

Open your access page

Provisioning is automated, with no human in the loop. Your access page shows the Tailscale invite, SSH command, VNC address, and password right after checkout.

STEP 033
$ ssh kizai@mac-ultra-01
Last login: ...
$ ollama run llama3:70b
>>> ready

Connect and build

SSH from your terminal, or use VNC / Screen Sharing for the desktop. VS Code Remote SSH, Cursor, Ollama, Docker — your usual tools work as-is.

What fits in 512 GB

Frontier open-weight models, in one box.

With 512 GB of unified memory at 819 GB/s, the weights stay resident in memory — no sharding across GPUs, no offloading to disk.

GLM-5.2 (744B MoE)4-bit · ~390 GB
FITS
DeepSeek-V3 / R1 (671B MoE)4-bit · ~380 GB
FITS
Kimi K2.7 Code (1T MoE)2-bit · ~325 GB
FITS
DeepSeek V4-Flash (284B MoE)4-bit · ~145 GB
FITS
Qwen 3.6 / 70B-class4-bit · ~40 GB each
SEVERAL AT ONCE

Quantized sizes are approximate and depend on format and context length. The 96 GB configuration comfortably holds one 70B-class model at 4-bit. We don't publish our own benchmark numbers — bring your workload and measure what matters to you.

Spec comparison

Two configurations, picked by workload.

The 96 GB model covers most experiments. For large open-weight models — or several models resident at once — take the 512 GB configuration that Apple no longer sells.

SPEC
STANDARD
96 GB unified memory + 1 TB SSD
Solo dev · 7B–70B experiments
HIGH-END · APPLE DISCONTINUED
512 GB unified memory + 4 TB SSD
Up to 1T-class models
Memory
96 GB unified
512 GB unified
CPU cores
28 cores
32 cores
GPU cores
60 cores
80 cores
Storage
1 TB SSD
4 TB SSD
Model scale
7B–70B class
405B–1T-parameter class
Concurrent models
1–2 mid-size resident
Multiple large models resident
Network
1 Gbps fiber
1 Gbps fiber

Full 512 GB spec sheet — hardware, model capacity, and rental details →

Full control

A real Mac, entirely yours.

Not Linux in a container — Apple Silicon macOS itself. You get an admin user, so it works exactly like the Mac you already know.

ADMIN ACCESS

Admin rights, full control.

This is the difference from cloud GPUs: you log in to macOS as an admin and run things your way, end to end.

# whoami
kizai (admin)

Sign in to your own accounts

Use your usual cloud accounts and sync. It behaves like a Mac on your desk, because it is one.

Install anything

Apps from anywhere — App Store, direct downloads, your own builds. No allowlist.

Homebrew / Docker / Python

Bring your whole toolchain. Your shell config, your version managers, your dotfiles.

Full system settings

Startup items, networking, sharing — every macOS setting is yours to change.

Note: machines come as clean macOS installs. Environment setup (Ollama, runtimes, etc.) is up to you — we don't provide setup support.

Security

A locked machine, so you can focus on the work.

Encrypted via Tailscale

Direct WireGuard tunnels between your devices and the machine. No inbound ports are ever opened.

Your own admin user

A dedicated admin account issued to you alone. Other customers and operators don't see your work.

Data stays on the machine

Your data lives on the Mac's SSD — useful for work you'd rather not upload to a cloud service.

Fully wiped between rentals

After each rental the machine is fully erased with Apple Configurator and handed over like new.

Pricing

Four plans. One payment. No subscription.

Pick a memory size and a duration. You pay once and your rental starts immediately on payment, running for the full term — it's time-limited, with no auto-renewal and nothing to cancel.

96GB · 7 DAYS
96 GB unified memory
Runs 7B–70B-class models
$149/ 7 days · one-time
≈ $21/day · dedicated, not shared
Contact us for availability
96GB · 14 DAYS
96 GB unified memory
Runs 7B–70B-class models
$249/ 14 days · one-time
≈ $18/day · dedicated, not shared
Contact us for availability
512GB · 7 DAYSAPPLE DISCONTINUED
512 GB unified memory
Runs 405B–1T-class models
$499/ 7 days · one-time
≈ $71/day · dedicated, not shared
Contact us for availability
512GB · 14 DAYSAPPLE DISCONTINUED
512 GB unified memory
Runs 405B–1T-class models
$799/ 14 days · one-time
≈ $57/day · dedicated, not shared
Contact us for availability

Need more time? Extend from your access page. Monthly use is available after your first rental.

One-time payment · Time-limited (machine wiped at the end) · No auto-renewal · No refunds (see Terms) · Clean macOS, full admin — bring your own stack

FAQ

Common questions.

Do you ship the Mac to me?

No — this is a remote rental. The machine is hosted in Kobe, Japan, and you connect to it over Tailscale using SSH, VNC, or macOS Screen Sharing. Nothing is shipped, there is no deposit, and there is nothing to return.

How quickly can I connect after paying?

Provisioning is automated with no human in the loop. Your access page — with the Tailscale invite, SSH command, and VNC details — is available right after checkout completes.

Do I need an account before purchasing?

No. Pick a plan and pay with Stripe Checkout. Your access page is tied to the email you use at checkout, and we email you a link to it as a backup.

Can I extend my rental?

Yes. From your access page you can extend by 7 or 14 days at the same price as the initial rental. Monthly use is available after your first rental — contact us from your access page.

Where are the machines located?

Kobe, Japan, on a 1 Gbps fiber line. You connect over Tailscale, which builds a direct WireGuard tunnel from your device, so day-to-day latency depends mostly on your distance to Japan.

Do you help with environment setup?

No — machines come as clean macOS installs with admin access, and you install your own stack (Ollama, runtimes, anything else). This keeps the machines predictable and the price low.

What is the refund policy?

All sales are final; there are no refunds. Details are in the Terms. If something is wrong on our side, contact support and we will sort it out.

A dedicated Mac Studio,
for exactly as long as you need.

Pay once. Open your access page. Connect over Tailscale and get to work.