Former proponents of democratising artificial intelligence through open-source releases of Llama at Meta, engineers and executives are now contemplating a radical philosophical shift: giving up their top open-source AI model, Behemoth, in favour of creating a closed, proprietary system.
This possible withdrawal occurs at a critical juncture. Although Meta is considering shutting down, Chinese AI labs have risen to the top, becoming not just competitors in the open-source large language model competition but its clear leaders.
Meta’s Trouble with AI Continues
The SuperIntelligence Lab at Meta, established to further the company’s AI goals, is at a turning point. The business is now reevaluating the same tenet that brought it recognition for transparency and innovation acceleration.
Previously, it set itself apart from clandestine rivals like OpenAI, Anthropic, and Google by openly disclosing its most potent AI models. Alexandr Wang, the former CEO of Scale AI, was recently named Chief AI Officer by the business, which also announced plans to invest hundreds of billions of dollars in large AI supercomputing clusters called Prometheus and Hyperion.
Sources familiar with the company’s discussions claim that these actions have successfully eliminated internal opposition to limiting model access. Chinese AI labs have jumped at the chance to assert leadership in the open-source AI space as Meta re-examines its open strategy, potentially creating a long-lasting edge in the global AI infrastructure.
Chinese AI Labs Changing the Dynamics of the Business
Released under an MIT-style open-weight license, DeepSeek-R1 and DeepSeek-V3 are currently on par with Gemini 2.5 Pro in terms of performance and were apparently trained for a fraction of the price, at about $6 million as opposed to OpenAI’s anticipated $100 million.
This accomplishment is not unique. China has developed what one venture capitalist called “an arsenal of open-source models,” such as Alibaba’s Qwen 3, a family of models with 128K token context released under the Apache-2.0 license that beats Deepseek V3 on important benchmarks, and Moonshot AI’s Kimi K2, which is excellent at high-code and complex tasks.
Despite Being Popular, Chinese AI Labs a Real Threat to Nation’s Data Security
Chinese labs’ advancement and Meta’s simultaneous retreat have had a domino effect on the global tech scene. Countries like Czechia, Australia, Canada, India, and the United States have started to prohibit Chinese LLMs due to data security concerns, even as certain Chinese models gain popularity on GitHub and Hugging Face.
Prominent venture capitalists, including Marc Andreessen, have warned that China will dominate the global technology stack if Western companies don’t lead in open-source AI. This fear now seems to be coming true more quickly than expected.
There are several reasons for Meta’s reevaluation. Llama 4, the company’s most recent open-source model, has failed. Meta is under increasing pressure to monetise its AI assets after making significant investments in infrastructure, talent, and computing.
Leave a Reply