TLDRs:
Contents
- Meta reduces AI open-source sharing, citing safety, as Chinese rivals accelerate public model releases.
- Chinese firms now dominate top Hugging Face rankings with open models from Zhipu, Tencent, and Alibaba.
- Analysts say Meta’s pullback reflects a competitive retreat, not just a responsible move.
- AI infrastructure costs push Meta to seek external partners, selling $2B in data center assets.
Meta Platforms has announced a strategic retreat from open-sourcing its artificial intelligence (AI) models, citing growing concerns over safety and misuse.
CEO Mark Zuckerberg said the company will adopt a more cautious approach, especially regarding its Llama family of language models. The move signals a sharp pivot from Meta’s previous embrace of open-source AI, which had positioned it as a leader in democratizing AI access.
However, this change comes at a moment when Chinese AI firms are accelerating their open-source ambitions. According to Hugging Face metrics, eight of the ten most popular open models now come from Chinese companies, including Zhipu AI, Tencent, Moonshot AI, and Alibaba. These firms are not only developing advanced AI systems but making them freely available, rapidly expanding their global developer reach.
Competitive Pressures May Be Driving Meta’s Caution
While Meta cites safety concerns, industry experts suggest a more strategic motive.
“Open-source retreats often indicate a weakening competitive position,” said Jeffrey Ding of George Washington University. “Meta’s models no longer push the frontier. The timing of this announcement speaks volumes.”
Indeed, Chinese models are attracting millions of users globally. DeepSeek’s R1 model, for instance, saw its user base triple from 33 million in January 2025 to 99 million by April, with over 2.5 million derivative versions downloaded. These figures indicate that Meta’s AI offerings may be falling behind not just in capability but also in adoption and influence.
This mirrors a broader trend, where U.S. companies maintain tighter control over foundational models while Chinese firms use open-source as a soft power strategy. By sharing powerful models openly, Chinese firms are forging connections with developers and institutions globally, especially in countries priced out of expensive, proprietary Western models.
AI Costs Force Meta to Seek Partners
Compounding Meta’s strategic pivot is the rising cost of AI infrastructure. In a separate announcement, Meta revealed plans to sell $2 billion worth of data center assets, aiming to bring in third-party partners to co-develop AI facilities.
The company reclassified land and construction assets as “held-for-sale” and may contribute them to joint ventures over the next year.
Meta’s capital expenditure forecast has been revised upward to between $66 billion and $72 billion for the year, underlining the immense financial demands of AI advancement. CFO Susan Li noted the company is actively exploring financial partnerships, acknowledging that even a tech giant like Meta cannot fund AI development alone.
China’s Open-Source Push Signals Geopolitical Shift
While Meta reassesses its AI strategy, Chinese firms appear to be leveraging open-source AI not just for market dominance, but for geopolitical gain.
Analysts point to Beijing’s recent AI Action Plan, which promotes multilateral cooperation and supports access to AI for developing nations. This stands in stark contrast to the U.S. strategy of guarding technological dominance.
By providing free access to advanced models, Chinese companies are positioning themselves as global AI benefactors. According to AI pioneer Andrew Ng, China’s AI industry operates in a “Darwinian struggle” that’s driving innovation and making cutting-edge tools available to a wider global audience.
As Meta dials back its open AI commitments, the broader AI race is rapidly evolving, with China emerging as both a technological and diplomatic force through openness, while U.S. firms hedge their bets in the name of control and security.