Meta CEO Mark Zuckerberg called on the United States to lower barriers to developing open-type artificial intelligence models, arguing that Washington needs to strengthen the American tech industry against the rapid progress of Chinese competitors.
His intervention came alongside the unveiling of a new Meta model, Muse Glimmer, which is smaller than rival companies’ flagship models and designed to run agentic tasks directly on Mac or PC computers equipped with a single graphics card.
Zuckerberg also signaled that Meta will soon release even larger models, arguing that artificial intelligence should not be concentrated in the hands of a few companies.
“We have even bigger models coming soon,” he said in a video accompanying his 14-page essay titled “The Future Is for Everyone.”
He argued that the idea that AI is so dangerous that the only safe path is extreme concentration of power is inherently flawed.
The Fight Over Open-Weight Models
Zuckerberg’s latest move reflects growing support for open-weight models, which are gaining ground as businesses worry about the rising cost of using AI and about cybersecurity concerns tied to large models.
Open-weight models tend to be cheaper than flagship models from companies like OpenAI and Anthropic, and offer greater flexibility since their core components are available to users and businesses. Closed models, by contrast, remain fully controlled by the companies that develop them.
Chinese startups currently lead and hold their position in the open-weight model market, with systems like Moonshot’s Kimi K3, Alibaba’s Qwen3.8-Max and DeepSeek V4-Flash rivaling top American systems in performance.
Meanwhile, the flagship models from American companies OpenAI, Anthropic and Google remain closed.
Meta Returns to Open Models
Meta also plans to release the weights of Muse Spark 1.2, its most advanced model, developed by the expensive superintelligence team the company built last year, in an effort to catch up in the AI race.
Meta has a long history of building open models, but the reception of Llama 4 last year led it to shift strategy.
In his new essay, Zuckerberg argued that US policy needs to be reconsidered if America wants to keep its lead in open-weight models. Among other things, he pointed to restrictions on data use and on the process of transferring knowledge from one model to another, in which output from a more powerful model is used to train a smaller, less computationally demanding one.
“US policy needs to reduce this added friction if we want American open-source models to lead over time,” he said.
He also spoke out against restricting access to foreign open-source models, arguing that this would not be an effective solution.
The Cost of AI and Data Centers
Zuckerberg further argued that the US faces a major disadvantage compared to countries like China: the difficulty of building the necessary infrastructure.
Meta is expected to spend up to $145 billion this year on AI infrastructure, while growing local opposition to data center construction has become a major obstacle for tech companies trying to expand their computing power.
To address community pushback, Meta also announced a $1 billion fund to support communities affected by data center development.
Finally, Zuckerberg announced that Meta will set up a corporate governance structure that will let independent experts approve safety criteria before new models are released.
The debate is taking on political weight in Washington, as the Trump administration has said it will not subject open-weight AI models to voluntary safety testing, according to sources familiar with the discussions.
