Jensen Huang, Mistral AI and a16z: Building national sovereign AI and avoiding digital colonization starts with ourselves

NVIDIA CEO Jensen Huang, French AI startup Mistral AI founder Arthur Mensch, and the renowned venture capital firm a16z jointly discussed the necessity of Sovereign AI. The three stated that every country should own, train, and operate its own AI, rather than completely relying on Silicon Valley or Chinese technology. They further emphasized that AI reflects culture, carrying language and values. AI is the electrical revolution of this generation, and countries that do not develop their own AI models will ultimately become "digital colonies" of other nations.

AI is a General Technology Affecting Healthcare, Education, Agriculture, and Defense

Huang and Mensch unanimously pointed out that AI has become a General Purpose Technology, deeply integrated into every industry and social aspect, like electricity and the internet before it, including healthcare, education, agriculture, and even defense.

They stated that AI can represent and execute tasks and make judgments like humans. This also means that if a single enterprise monopolizes this technology, other countries and businesses might be forced to "wait" for others to determine the future.

AI is Culture, Carrying Language and Values

AI is not just algorithms and data, but a reflection of "culture". Mensch warned that if countries worldwide do not participate in training AI for their own language and culture, they will be imperceptibly dominated by "foreign cultural AI".

Huang added that Swedes would not want an external company to determine how Swedish should be expressed. Similarly, Middle Eastern countries need their own language and regulatory understanding. "Your AI should not be decided by others."

Building Sovereign AI from Infrastructure to Application Layer

For how countries or enterprises should approach Sovereign AI, they recommended:

1. Purchase Infrastructure: Such as chips, cloud resources, and open-source models.

2. Develop Upper Layers: Use local knowledge and regulations to train specialized language and application systems.

3. Establish AI Human Resources Department: Future enterprises will not only manage human employees but continuously recruit, train, and evaluate their "AI digital employees".

Specialized Models Represent National Strength, Training Language, Healthcare, and Legal Domains Independently

Mensch believes model specialization occurs in three layers:

1. Language Specialization: First understand what you are saying.

2. Industry Specialization: Then understand the domain in which you are speaking.

3. Local Cultural Specialization: Finally understand your values and approach to handling matters.

He emphasized that this is the value of sovereign AI, no global giant can care more about your own language and culture than yourself.

Not doing it yourself is digital colonialism, culture and wisdom cannot be outsourced

Talking about the most severe risk, the host directly pointed out: "If AI is cultural infrastructure, and we do not own it, the result will be the same as past colonization."

Huang Renxun agreed: "Your culture and preferences will directly reflect on AI behavior, this is not something that can be outsourced via API." He gave an example that McDonald's and Starbucks can be seen globally, but what truly defines a city's culture are the corner cafes and local breakfast shops.

The importance of open-source models, AI should have a "democratic foundation"

Mistral and NVIDIA are supporters of open-source models. They believe:

• Open source can drive innovation in niche industries (such as mining, energy, defense, etc.).

• Open source allows more people to participate, review, and improve safety.

• Closing weights will not enhance national security, it will only allow other countries to get ahead.

From AI to culture to economy, the key is talent and partnerships

For national leaders wanting to develop sovereign AI, they offer the following advice:

1. Strengthen local AI talent cultivation: Without technical personnel, model training and customization cannot be discussed.

2. Build infrastructure and software platforms: Including AI factories, data centers, model training environments, etc.

3. Collaborate with trustworthy tech companies: Such as NVIDIA, Mistral, who can provide general tools and support open source.

4. Utilize AI to shorten technological gaps: Not to cause unemployment, but to enable more people to solve problems with AI.

The next AI wave will move from chatting to thinking, from screens to the physical world

In the next decade, Huang Renxun predicts AI development will evolve in three directions:

1. Agentic AI: Capable of proactively handling tasks.

2. Physics AI: Able to understand physical, chemical, and manufacturing principles.

3. Robotic AI (Physical AI): Truly entering the physical world, assisting in production, manufacturing, and medical care.

This dialogue repeatedly emphasized one thing: "No global giant can care more about your own language and culture than yourself, do not wait for others to help, start with yourself."

(Sam Altman's Three Observations: Costs Dropping 10% Annually, AI Agents Becoming New Work Standard, Assets Immune to AI Replacement Will Appreciate)