Why Preferred Networks Builds Its Japanese AI Models Entirely From Scratch
# Preferred Networks# domestic AI# sovereign AI# model development
Key Points
- Preferred Networks develops AI models it bills as domestically made, built entirely from scratch.
- The company explains why it insists on a full in-house approach and what it gains.
- Building from scratch and fine-tuning existing open models are the two dominant development routes today.
Analysis
Building from scratch costs far more than fine-tuning, and the payoff is not short-term performance but control and accumulated capability — what data goes in, how the model is adjusted, and whether supply can be cut off by someone else. These are things you only hold if you build the whole thing. Taiwan's own domestic-model debate faces the same trade-off.