The industry has been abuzz with new open-source AI models from Chinese companies, most of which are potentially competitive. The US, meanwhile, is doing the opposite.
Top labs like OpenAI, Anthropic, and DeepMind continue to guard their most advanced models. Even those who have released open versions, Meta with Llama, Google with Gemma, do so cautiously, and often at a safe technical distance from their most profitable offerings.
China is advancing in the global AI race through transparency, speed, and a deliberate effort to lead in open source, as seen with Alibaba’s Qwen3-Coder, GLM 4.5 or Kimi K2.
This isn’t just a question of capability. It’s a question of strategy. And the two largest AI ecosystems in the world are now playing by very different rules.
Why the Big Tech Won’t Go Fully Open
Wei Zhou, head of AI Utilisation Research (AIUR), and Jordan Nanos, member of technical staff at SemiAnalysis, told AIM that open source is not a moral stance — it’s a business calculation.
“In short: Best performing models stay closed for competitive advantage. Open-sourcing is often a move to gain traction when behind and to attract talent.”
They continued, “Meta and Google have open-sourced models (Llama, Gemma), with smaller ones like Llama 3 8B being useful for certain tasks. But top labs like OpenAI, Anthropic, and DeepMind don’t release their best models.”
Zhou and Nanos believe open-sourcing would hurt the company’s ability to charge premium access fees. They said, “Closed models are higher margin, which helps fund talent, datacentres, and maintain a tech lead for future funding. Open-sourcing is mainly for attracting talent or advancing the community, not profit.”
They also made an interesting remark: “User data can still be collected regardless of whether weights are open or closed.” In other words, open models don’t scale revenue, and for most US players, that may be the deciding factor. The frontier stays behind the paywall.
China’s Perspective on Openness
In contrast, Chinese tech leaders are embracing openness with a blend of ambition and ideology. Grace Shao, a Hong Kong–based business analyst who tracks China’s tech ecosystem, told AIM that this shift signals more than just competitive energy; it’s also about asserting identity.
Among Chinese founders, especially those like DeepSeek’s Liang Fan, there’s a belief that “it’s better to enable developers… to harness and nurture healthy market competition.”
She also highlighted a key emotional undercurrent, “There’s been so much criticism of Chinese researchers copying and reiterating on US innovation… but I think for this generation they’re saying, look, they want to be proud… and they want to put their technology and research out there and say we’re really innovating in engineering — here is actually proof for it.”
That pride, she notes, doesn’t necessarily come with hostility. In fact, Shao believes the West’s framing of this as a geopolitical rivalry is overstated.
When asked if there’s an incentive for Chinese developers or a sense of competition, she said that a degree of personal ego might be at play, driving the desire for open-source methods to demonstrate innovation rather than imitation.
However, she believes that one cannot simply conclude that all researchers’ efforts are motivated solely by this competitive spirit or personal ego. The competition narrative is driven more by the American media, she said.
Chinese developers, in her view, are largely pragmatic — focused on enabling more innovation and wider integration, rather than fuelling ideological tension.
Open Source in AI: Is There a Risk?
Sanchit Vir Gogia, chief analyst & CEO at Greyhound Research, sees open source in AI as a geopolitical tool, used differently in each hemisphere. “Open source in AI is no longer a binary choice — it’s a geopolitical lever,” he told AIM.
“While Chinese tech majors race ahead with publicly released large models like InternLM, Qwen, and Yi, their Western counterparts remain guarded, treating open sourcing of frontier models as a liability trap,” he said.
Gogia highlighted that, according to the Greyhound CIO Pulse 2025, 61% of global CIOs support some level of openness for auditability and domain fine-tuning. But only 27% in North America trust Chinese-origin open models, citing censorship risks, unverifiable training data, and ideological alignment concerns.
From a risk perspective, Western companies view open weights as potential gateways for misinformation, cyber threats, and regulatory non-compliance — especially in light of the EU AI Act and growing US scrutiny.
“Western tech firms view full open sourcing of foundation models as incompatible with their commercial risk frameworks,” Gogia explained. “By contrast, Chinese tech companies, with state support, use open sourcing as a means of achieving technological self-reliance and global adoption.”
Gogia said that these models act as Trojan horses — embedding Chinese LLM norms and architectures into emerging markets under the guise of open access.
“While the Trump administration’s ‘winning the race’ AI plan now champions open source, it remains cautious — encouraging vetted, defensible releases over unguarded exposure,” he explained.
“Ultimately, open sourcing in the West is not being rejected — it’s being carefully redefined to suit governance and monetisation priorities.”
Gogia cited an example where a global pharma company tested a Chinese LLM for multilingual document analysis. While technically promising, the model triggered red flags for geopolitical hallucinations and unverifiable dataset lineage. Regulatory teams deemed it unfit under EMA and FDA guidelines, and the CIO halted the rollout in favour of a Western model with traceable provenance.
The companies in China and the West seem like they both want openness, but the philosophy behind their respective approaches is wildly different. It could be a silent war that cannot be seen directly, and a clear winner out of it could emerge in the near future.
[With Inputs from Supreeth Koundinya]
The post What Makes China Go Open Source in AI And Not the US appeared first on Analytics India Magazine.
