Research

Thinking on the forces shaping local AI: economics, open-weight model quality, privacy, and why running models locally is becoming the sensible default.

The Case Against Always-On Internet AI

Reliability, latency, and cost are reasons enough. Privacy and sovereignty are bonuses.

May 15, 2025 · 3 min read

Privacy Is Not a Feature

Every API call is a data transfer. Local inference isn't a privacy feature — it's the absence of a privacy risk.

May 12, 2025 · 4 min read

Open Weights Are the New Open Source

Qwen3, Gemma 4, and DeepSeek-V3 prove that closed models no longer have a quality moat worth talking about.

May 8, 2025 · 4 min read

The Subsidy Cliff

The big AI providers are losing money on every API call. When the subsidies end, the prices won't stay where they are.

May 1, 2025 · 5 min read