AI Deployment

Own your stack. No per-seat SaaS.

AI Deployment / Hardware Guide

Local AI Hardware Guide

A plain-language guide to running AI on your own hardware — what each tier is for, what it can actually run, and what's coming next. This is a resource inside our AI Deployment service, not a separate product: the hardware is cheap and commoditized; the value is the integration, configuration, and management around it. No jargon, no hard sell.

Your data stays put

The model runs on a box in your office. Nothing you type leaves the building — no cloud, no third party reading it.

No monthly cloud fees

You buy the hardware once. There's no per-seat subscription or usage meter ticking in the background every month.

It augments your team

A private assistant for drafting, research, and routine work — sized to your business, owned by your business.

Entry · Solo operator or small office
Quiet, low-power, and private — your own AI model running offline on a small desktop.
Box cost only
~$1,500–$3,000 CAD
Compact silver mini-desktop render — entry-tier private AI box
Apple Mac mini (M4 / M4 Pro)
A tiny, silent desktop that fits in your hand. A 48GB M4 Pro is the sweet spot for private AI.
What this is forOne person who wants a quiet, always-on private assistant on their desk.
What it runs8B–30B models comfortably; a 48GB config handles ~70B at a modest pace. 1–2 agents.
DeliveryLead time confirmed at scoping — varies by configuration.
Compact black mini-PC render — entry-tier private AI box
AMD Ryzen AI mini-PC
A compact Windows/Linux box (e.g. a Minisforum-class Ryzen AI 9) for teams that prefer not to run Mac.
What this is forA small office wanting an affordable, standard-OS private AI box.
What it runsFast on 8B–14B models; 30B at a moderate pace. 1–2 agents.
DeliveryLead time confirmed at scoping — varies by configuration.
Box cost only. What we charge to configure, install, and support your AI system is the conversation — book a call.
Standard · The workhorse
A shared AI resource for a small team — more memory, more speed, more people on it at once.
Box cost only
~$3,000–$6,000 CAD
Brushed-metal cube desktop render — standard-tier private AI box
Apple Mac Studio (M4 Max)
A silent, powerful box that sits in the office and serves the whole team's AI from one place.
What this is forA small team sharing one capable, quiet private AI machine.
What it runs~70B models comfortably; 3–5 agents at once.
DeliveryHigh-memory Apple configs are supply-constrained in 2026 — roughly 6–10 weeks.
Mesh-front compact desktop render — standard-tier private AI box
Strix Halo mini-PC (Framework Desktop / GMKtec EVO-X2)
Ryzen AI Max+ 395 with 128GB — a lot of AI memory packed into a very small footprint.
What this is forA team wanting large-model capability without a full workstation or the Apple wait.
What it runs~70B models comfortably; 3–5 agents at once.
DeliveryLead time confirmed at scoping — varies by configuration.
Box cost only. What we charge to configure, install, and support your AI system is the conversation — book a call.
Performance · Heavy lifting
The largest models, many agents at once, and full data sovereignty — for shops that run real volume.
Box cost only
$6,000 and up
Brushed-metal cube desktop render — performance-tier private AI box
Apple Mac Studio (M3 Ultra)
The maximum-memory Mac — quiet enough for an office, strong enough to run many agents together.
What this is forA team running large models and many agents without a noisy server room.
What it runsLarge models with many concurrent agents.
Delivery96GB is the current ceiling (128/256GB withdrawn May 2026); long lead times — confirmed at scoping.
Deskside AI compute unit render — performance-tier private AI box
NVIDIA DGX Spark (GB10, 128GB)
A personal AI supercomputer — NVIDIA's deskside box built specifically for AI development and inference.
What this is forHeavier development and inference on a dedicated, purpose-built AI machine.
What it runsLarge models with strong, sustained throughput.
DeliveryLead time confirmed at scoping — varies by configuration.
Full tower workstation render — performance-tier private AI box
RTX-Pro / Threadripper-Pro workstation
A custom-built workstation from a Canadian integrator (Puget Systems ships to Canada) for the most demanding work.
What this is forData-sovereign shops running multiple large models — training and inference together.
What it runsMultiple large models concurrently for data-sovereign training and inference workloads.
DeliveryCustom Canadian build (Puget Systems ships to Canada); lead time confirmed at scoping.
Box cost only. What we charge to configure, install, and support your AI system is the conversation — book a call.
On the horizon

What's coming next

What's next in local AI hardware — clearly labelled so you know what's real and what's rumour.

Rumored

Mac Studio M5 Max / M5 Ultra

Expected later in 2026 (possibly October; RAM-shortage delays are likely). The M5 Ultra is rumoured to support up to 256GB of unified memory.

Confirmed

NVIDIA DGX Station (GB300 Grace Blackwell)

Announced. A deskside supercomputer with very large coherent memory, trillion-parameter class. Tens of thousands of dollars; shipping later in 2026.

Confirmed

AMD Ryzen AI Max 400 "Gorgon Halo"

Announced · systems Q3 2026. AMD has detailed the Max 400 family, with OEM systems expected to start in Q3 2026; not broadly buyable yet.

Plain language

How to read this guide

Three things to keep in mind — no technical background required.

Specs = brains + memory

The chip is the brains; the memory (RAM) is the workspace. More memory means the machine can hold a larger, smarter model at once.

Bigger memory, bigger model

"70B" or "30B" is the size of the AI model. Larger models are more capable but need more memory — which is what separates the tiers.

Lead time matters

Chip shortages and custom builds mean some machines take weeks to arrive. We factor delivery time into every recommendation.

Not sure which box fits?

That's the conversation. On a free discovery call we'll size the right hardware to your workload and budget — then configure, install, and support it with you.

Book a Free Discovery Call →