Open-weight and open-source AI aren’t just technical artifacts; they are geopolitical tools. You can have both a thriving proprietary model ecosystem and a flourishing open ecosystem. Deepseek proved that just releasing a strong open-weight model can have disproportionate psychological and market impact. Most of the capital is currently flowing into closed, private models, but open-weight models are already punching above their weight in real-world performance. The catch? Most of the best ones are coming out of China. It’s important to distinguish open-weight from open-source. We often don’t know what data went into training these models. That might not matter for 98% of use cases, but for sensitive domains, it absolutely does. Like everything in tech, security lags behind deployment. In areas like code generation, where subtle backdoors could be catastrophic, trust becomes paramount. That’s why genuinely open-source models with verifiable data provenance should be prioritized in sovereign and critical infrastructure use cases.