China AI Monitor: Why Chinese AI Is Going All-In on Open Source as the U.S. Pulls Back
Funding gaps, user stagnation, and political trust are driving China’s open-source gamble while U.S. giants lock down their models.
A split is emerging in global AI. While U.S. giants are increasingly locking their technology behind closed doors, Chinese companies are flinging theirs wide open. The contrast was on full display last week: just two days before unveiling its controversial GPT-5, OpenAI unexpectedly released gpt-oss—its first set of open-weight models, a rare move in an otherwise closed-first strategy. OpenAI co-founder Greg Brockman told Wired, that these models weren't undercutting OpenAI's proprietary models with a free option, but rather complementing OpenAI's existing services. The gpt-oss series quickly shot to the top of Hugging Face's trending models list, though before this, the vast majority of trending open-source models came from China.
This shift is deeply symbolic. Just last Wednesday, Mark Zuckerberg hinted that Meta might pivot from open to closed source while sharing his vision for personal superintelligence. With Meta retreating and OpenAI not prioritizing open-source models, this has opened a strategic window for Chinese players to step in and set the tone for global open-source AI. Chinese AI companies from tech giants to startups have almost universally embraced the open-source path. Chinese companies are rewriting the open-source AI narrative.
How Far Along Are Chinese Companies with Open Source?
Tech Giants Taking Different Approaches
The first half of 2025 saw Chinese tech giants collectively hit the accelerator on open source. On the last day of June, both Baidu and Huawei simultaneously announced major open-source releases. Robin Li, who had previously championed closed-source strategies, finally led Baidu to open-source 10 models from its ERNIE 4.5 series. Huawei also open-sourced multiple versions of its Pangu models. This marked a turning point: even giants who had been hesitant about open source could no longer ignore the trend.
Alibaba stands out as perhaps the most committed executor of open-source strategy. Since 2023, its Qwen team has open-sourced over 200 models, releasing multiple Qwen3 series models in July 2025 that dominated Hugging Face rankings. Alibaba's calculation is clear: attract developers through open-source models, then drive Alibaba Cloud sales. This strategy has proven effective, generating significant revenue growth for their cloud division.
Tencent open-sourced its first Mixture of Experts (MoE) model, Hunyuan-A13B, on June 27, followed quickly in July by HunyuanWorld-1.0: the industry's first interactive 3D world generation model. This showcases Tencent's ambitions in specialized application domains. ByteDance has recently open-sourced various developer-focused technologies and frameworks, though its flagship Doubao model remains closed-source.
Even smartphone manufacturer Xiaomi, while disrupting Tesla with its electric vehicles, released the open-source voice model MiDashengLM-7B in August to power its automotive and smart home ecosystem.
Increasingly Open Startups
While large companies open-source models primarily for strategic positioning within their broader business empires (especially given most have cloud services), Chinese AI startups beyond DeepSeek have also embraced open source. Moonshot AI open-sourced its highly praised Kimi K2 model on July 11, with its powerful coding capabilities once again boosting community confidence in Chinese large language models. Zhipu AI has repeatedly open-sourced its GLM series models in April and July, and prominently adopted the Z.ai domain, signaling its global and open-source ambitions.
Why Have Chinese Companies Chosen a Path Opposite to America?
DeepSeek Shows the Way
DeepSeek R1's explosive success served as a crucial catalyst. Before this, the value of open source was hard to quantify and couldn't directly translate into investor-recognized returns. But DeepSeek proved that a world-class open-source model could itself become the most powerful marketing tool, capturing global developer attention and adoption. This showed Chinese AI companies a clear path to technological leapfrogging and global influence.
Subsequently, models like Kimi K2 and Qwen3 topped authoritative rankings like LMArena, further validating this approach's viability. Today, the world's best open-source models are almost entirely dominated by Chinese offerings: Kimi, DeepSeek, and Qwen. As one Moonshot AI employee put it, DeepSeek's success proved that their belief in "high-quality open source" was indeed a viable path forward.
Stagnant User Growth Drives Search for New Ecosystems
China's internet development has been heavily mobile-focused rather than PC-centric, leading to AI usage being primarily concentrated on mobile rather than desktop. However, Chinese AI products are hitting user growth bottlenecks, pushing AI companies to look toward B2B and developer ecosystems—and open source is the most effective way to build ecosystems and capture attention.
QuestMobile's "2025 Domestic AI Application Mid-Year Report" clearly indicates that active user numbers for AI products have declined on both mobile and PC platforms, with native app growth "completely stalled." Well-known applications including Kimi and ChatGLM have become "disaster zones for user churn."
Meanwhile, Chinese AI companies lag significantly behind their American counterparts in fundraising capabilities. Stanford's 2025 AI Index Report shows that "In 2024, U.S. private AI investment grew to $109.1 billion—nearly 12 times China's $9.3 billion."
With insufficient capital reserves and commercialization revenues not yet fully materialized, purely pursuing DAU has become an outdated metric. Therefore, AI companies need new ways to demonstrate their technical value and potential, making open source an inevitable choice.
Through open source, companies can draw developers and users into their technical ecosystem, creating vibrant communities. As Moonshot AI's infrastructure lead explains: "The more you open source, the more people actually follow your stuff, and the cost for you actually becomes lower. Lots of people will build many things on top of our foundation." This not only helps models iterate and improve rapidly but also brings community recognition and resource support, opening new development paths when user growth stagnates.
Building Global Developer Trust and Mitigating Political Risk
In today's suspicion-filled geopolitical environment, any closed-source high-tech product from China may be viewed as a potential "black box," with its security and intentions easily questioned. Open-source strategies completely expose model architecture, weights, and even training code, allowing any global developer or researcher to conduct deep "dissection" and examination, significantly reducing external suspicions.
More importantly, Chinese models generally adopt permissive open-source licenses, contrasting sharply with Meta's Llama series restrictive custom licenses. The latter's ambiguous terms, unstable access experiences, and numerous commercial use restrictions deter many developers and enterprises hoping to build commercial products. In comparison, the open attitude of Chinese open-source models has won them greater developer favor.
The Great Divergence: Open vs. Closed
While OpenAI locks its most advanced models behind APIs, Chinese companies have chosen the completely opposite path: fully opening their technical capabilities to global developers. This divergence reflects China's unique position and suggests AI competition may evolve along two completely different trajectories. This “Great AI Divergence” is reshaping the global landscape. In the short term, open source has given Chinese firms unprecedented visibility and influence. Whether it can also deliver lasting commercial and technological dominance—that’s the question the next chapter of the AI race will answer.