AI deployment, infrastructure, and economics.
Models, compute, token economics, private infrastructure, and the systems required to deploy AI where knowledge is the asset.
The Anatomy of Private AI, Part 1: What a Model Actually Is
A model is a folder of numbers you can download. This page lets you open one, read what's inside, and switch the Wi-Fi off. Part 1 of a three-part series on why self-hosted open AI is private.
SOC 2 Doesn't Cover AI Agents. Here Is the Audit That Does
A SOC 2 report shows a vendor follows its process. It doesn't test what an AI agent can reach or send. The AI Agent Security Audit does: ten tests, run before go-live and after every change.
Ottawa Will Help Pay for Your Private AI Stack
Canada's new Productivity Mega-Deduction, the expanded SR&ED credit, and RAII's 0% financing stack. A worked example: how a $1M private AI build returns $250K in its first tax cycle.
The AI Training Question Your Vendor Can't Answer
A mathematician spent a year drafting a proof inside a vendor's AI tools. The vendor joined the race, won in 88 hours, and couldn't rule out that his data helped. Where your IP thinks matters.
The State of Open-Source Models 2026: The Big Board
The definitive ranked board for open-weight AI models — all 23, graded on six axes, filtered by hardware tier, category, and license. Updated October 2026.
AI Token Economics: Why Cheaper Tokens Made Your Bill Explode
LLM prices fell ~80% in a year, and AI spend doubled anyway. The mechanism, the math, and what to do about it.
Open-source inference
Production AI that answers for itself.
Every research note points to the same thesis: production inference is a stack problem. Providers, models, routing, cost, evidence, and operations. Northwood builds the open-source tooling and operates the whole system.