Jensen Huang’s First X Post Unveils a 25‑Institution Open‑Weight AI Letter

Following Mark Zuckerberg, Jensen Huang has opened an X account, and his first post is a public letter signed by 25 institutions that spans almost every layer of the AI industry.

The letter groups the signatories into four categories:

Models & research: Meta, Mistral, NVIDIA, Arcee AI, Black Forest Labs, Reflection

Infrastructure & enterprise software: Microsoft, IBM, Dell, ServiceNow, Box, CrowdStrike, Palantir, Telnyx

Open ecosystem: Hugging Face, Linux Foundation, Mozilla, Arena

Investment & entrepreneurship: Andreessen Horowitz, Emergence Capital, Y Combinator

Notably, OpenAI, Anthropic, and Google’s flagship closed‑source models are absent, likely because they prefer not to be named.

First, open weights bring AI into everyday business

Relying on the most powerful frontier models for every task is economically infeasible. The optimal approach is to reserve cutting‑edge models for truly difficult problems while assigning routine tasks to smaller, cheaper, and more specialized models.

Factories, hospitals, farms, classrooms, and small businesses often need a cost‑controlled, data‑controlled system that can run reliably for years, rather than the top‑ranked model on a leaderboard.

A model that can run stably on an enterprise intranet for three years may be more valuable than a marginal leaderboard gain.

Second, open weights spread competition across the whole stack

When models can be downloaded and modified, competition extends beyond model vendors to chip makers (seeking deployment efficiency), cloud platforms (contending on hosting costs), tool providers (improving fine‑tuning and evaluation), and application companies (offering domain expertise).

Open weights enable new products to emerge on top of the model layer, but the letter warns that a few closed‑source models could become the sole paid gateway of the AI era.

As foundational capabilities concentrate, startups lose bargaining power and enterprises face higher switching costs.

Third, enterprises need to keep knowledge in‑house

Companies investing heavily in AI fear vendor lock‑in. Prompt engineering, fine‑tuning data, evaluation sets, business processes, and user feedback become valuable internal knowledge assets.

Open weights let enterprises control their data, adjust models to their needs, and deploy them in environments that meet business requirements—crucial for sectors such as healthcare, manufacturing, finance, government, and security.

In many scenarios the primary concern is not model intelligence but whether data can stay off‑network, whether the system works offline, and whether the organization can decide when and how to upgrade the model.

Why the letter was issued

The United States’ early lead in open‑source models is being rapidly closed by Chinese models. According to the ATOM Project, Qwen‑derived models now account for over 40% of new monthly derivatives on Hugging Face, while Meta’s Llama has fallen from a peak near 50% to about 15%.

When a model becomes the foundation for extensive research, fine‑tuning, and applications, its influence spreads throughout the developer ecosystem, shaping which models become de‑facto standards.

25 institutions also have their own interests

NVIDIA sells compute, Microsoft, IBM, and Dell provide infrastructure and enterprise services, Hugging Face handles model distribution, startups and enterprise software vendors seek to reduce dependence on a single model provider, and investors look for more startup opportunities.

The more open the weights, the more active the entire industry chain becomes, expanding business opportunities for hardware, deployment, fine‑tuning, security, evaluation, and application layers.

For ordinary enterprises this diversified structure is healthier, offering choice and making it easier to switch suppliers.

The author, a long‑time advocate of open‑source models, notes that closed‑source models suit chasing performance ceilings, while open weights are better for spreading capability across countless industries.

A truly healthy AI ecosystem will likely see both routes coexist long‑term, preserving user choice.

Model weights are becoming the new main battlefield of global AI competition.