A year after DeepSeek R-1's launch, China has fundamentally shifted the global open-source AI landscape. Chinese-developed models now dominate downloads on Hugging Face, officially surpassing those from the United States. This marks a significant turning point, underscoring a rapid rebalancing of power in artificial intelligence.
Hugging Face, in collaboration with the Massachusetts Institute of Technology, recently published an analysis titled "The One-Year Anniversary of the DeepSeek Moment." This report revealed that Chinese open-source AI models now account for 17.1% of global downloads, edging out the US share of 15.8%. Giants like Baidu, ByteDance, and Tencent have dramatically increased their open-source contributions, with Baidu alone moving from zero to over 100 releases on the platform.
The DeepSeek Effect: Breaking Barriers
DeepSeek R-1's release in January 2025 is identified as the catalyst for this transformation, dismantling three critical barriers:
- Technical: It democratised advanced reasoning capabilities, making them accessible as downloadable, fine-tunable assets.
- Adoption: Its permissive MIT licence allowed for swift integration into production environments, accelerating real-world application.
- Psychological: Crucially, it reframed the industry's mindset, shifting the conversation from "can this be done?" to "how can we do this effectively?"
These ripple effects are clear. Even prominent American models, such as Deep Cogito v2.1, released in November 2025, are now built upon Chinese foundations, specifically a fine-tuned version of DeepSeek-V3. Adding to this, Alibaba Cloud's Qwen family has achieved over 700 million downloads, outstripping Meta's Llama to become the most popular open-source AI system globally. We've seen similar moves in the region with Qwen launching to take on Google's Nano Banana.
Geopolitical Implications and Market Response
Microsoft President Brad Smith recently voiced concerns that US AI companies are falling behind their Chinese counterparts outside Western markets.
He told the Financial Times^:
We have to recognise that right now, unlike a year ago, China has an open-source model, and increasingly more than one, that is competitive.
Smith specifically highlighted DeepSeek's rapid expansion in Africa and other emerging markets, where Western platforms often face affordability challenges. A Microsoft research report estimates DeepSeek holds an impressive 89% market share within China.
In response, the US launched the ATOM (American Truly Open Models) initiative in July 2025, with backing from industry leaders including Hugging Face CEO Clement Delangue and OpenAI's chief strategy officer Jason Kwon. This initiative explicitly cites Chinese AI momentum as its primary motivation. The push for open-source alternatives could also influence future discussions around topics like OpenAI to test ads inside ChatGPT and Google bets on ad-free AI, shuns ChatGPT model.
Investment and Future Outlook
The "DeepSeek shock" has spurred significant investment in Chinese AI startups. MiniMax raised $619 million in its Hong Kong IPO, with shares doubling on their first trading day. Zhipu AI followed suit, raising $558 million and becoming the first major Chinese large language model developer to go public.
While DeepSeek holds 4% of the global chatbot market share, compared to ChatGPT's 68% and Google Gemini's 18% (according to Similarweb data cited by Malay Mail), its trajectory is undeniable.
DeepSeek's forthcoming V4 model, anticipated in mid-February 2026, is reportedly outperforming both Claude and GPT series in code generation based on internal benchmarks. As Hugging Face concluded, "The world is still reacting, sparking a new wave of open-source enthusiasm." This shift indicates a broader movement towards diverse AI development, moving beyond a Western-centric model.
What are your thoughts on this significant shift in the global AI landscape? Share your predictions for the future of open-source AI in the comments below.
YOUR TAKE
We cover the story. You tell us what it means on the ground.
What did you think?
Share your thoughts
Join 3 readers in the discussion below
Latest Comments (3)
It’s interesting to consider the "psychological" barrier DeepSeek R-1 broke. Beyond the technical specs, how much of this download shift is genuinely about superior models versus a strategic framing and cultural narrative being pushed? Especially when we see US models like Deep Cogito v2.1 also building on these foundations, it suggests a deeper convergence than just raw numbers.
The rise of Chinese models, particularly Alibaba Cloud's Qwen family surpassing Llama in downloads, certainly points to a significant shift. However, I'm keen to understand the deeper implications for trust and safety in a regulatory landscape. The UK AI Safety Institute's focus on responsible deployment means we need to look beyond download figures. What assurances do these open-source models offer in terms of data privacy, bias mitigation, and overall ethical governance, especially as they become foundational for other systems? The "how can we do this effectively" question needs to encompass thorough safety assessments from the outset.
It's interesting how Deep Cogito v2.1 uses DeepSeek-V3. We're always looking at how to integrate powerful AI into our platforms at Tokopedia, especially for things like product recommendations or customer service. The question is, how much infrastructure investment is needed to run these large models efficiently in Southeast Asia? We don't always have the compute power of bigger markets.
Leave a Comment