Op-Ed: The Rise of Open Models, Taiwan's Opportunity to Seize the AI Mainstage

NVIDIA CEO Jensen Huang made his first post on social platform X on July 24 (U.S. time), sharing an open letter titled 'Open Weights and American AI Leadership.' At launch, the letter had 25 endorsing organizations, with others including OpenAI joining afterward. This signals that AI competition has expanded beyond chips, computing power, and model performance to encompass ecosystems and rule-setting authority.

Huang argues that AI will transform every industry and drive every company, and nations will develop their own AI capabilities. The world needs not only the most advanced closed models but also the most advanced open models. 'Open-weight models' refer to models whose weights can be downloaded, inspected, modified, and deployed on private infrastructure—but do not necessarily include open-sourcing the original code, training data, or full training process, and thus are not equivalent to open-source AI.

The letter highlights that open weights allow startups, universities, enterprises, and public sectors to fine-tune and deploy models without training from scratch, enabling organizations to maintain control over their data, accumulate knowledge, and reduce vendor lock-in risks. Greater participation also fosters competition across layers such as models, cloud, chips, applications, and services, preventing AI benefits from concentrating in a few platforms.

On risks, the letter does not shy away. Once weights are released, the original developer loses control over downstream uses, and modified versions are hard to track. However, it argues that closed models are not inherently safer—concentrating advanced AI in a few vendors may create a 'single point of failure.' Open-weight models, by contrast, support model evaluation, red-teaming, vulnerability detection, and patching. 'Model distillation'—using one model’s output to train or improve another—should be addressed through legal and commercial mechanisms rather than blanket restrictions.

A nation’s AI leadership cannot be determined solely by having the strongest model, but by its ability to widely deploy AI in a trustworthy and affordable manner, building a robust ecosystem. Open and closed models are not mutually exclusive; governance should be tiered based on use cases, model capabilities, and risk levels. Over-prioritizing a few closed model vendors may stifle innovation through high-cost services and platform dependency. Conversely, pushing openness without regard for information security, intellectual property protection, and accountability may incur significant costs.

For Taiwan, this is a pivotal opportunity to leap from being a 'key supply chain player' to a 'key ecosystem player.' With strengths in advanced semiconductor processes, advanced packaging, servers, cooling, and system integration, Taiwan should go beyond hardware provision. It should integrate open models into manufacturing, healthcare, finance, and public services, developing domain-specific models deployable within enterprises or on edge devices. Supporting SMEs in AI adoption will also enhance corporate control over sensitive data and trade secrets.

The government has already launched the 'Taiwan Sovereign AI Training Corpus (Beta)' and activated the National Grid Cloud Computing Center. The next step is to integrate computing power, clearly licensed data, testing tools, and evaluation mechanisms. It should enrich local language datasets—including Traditional Chinese, Taiwanese Hokkien, Hakka, and Indigenous languages—and establish systems for model provenance and version labeling, cybersecurity checks, risk classification, accountability tracing, and third-party verification. Taiwan should also actively participate in shaping international AI standards and governance frameworks.

AI sovereignty does not mean isolation, but the ability to choose models, govern data, deploy independently, verify risks, and co-shape norms. Taiwan should not remain merely the 'hardware heart' of AI, but evolve into a crucial hub for model applications, security validation, and trustworthy AI governance—transitioning from a key supplier to a co-architect of global AI rules.