The U.S. and China are locked in a high-stakes contest over artificial intelligence, with open-source models emerging as a potential significant development. While American firms like OpenAI and Anthropic guard their code behind paywalls, Chinese labs are flooding the internet with freely available models. The divide reflects deeper philosophical differences: one prioritizes control, the other collaboration. Tiezhen Wang, a former Hugging Face executive, has watched this shift unfold. He helped Chinese labs launch open-source projects, noting that the trend often blurs competition lines.
Wang said during a Rest of World event that Chinese models are already fueling U.S. research. Open-source weights from Chinese labs run on U.S. hardware, creating a strange symbiosis. Both sides benefit as the AI pie grows, but the question lingers: can open-source models truly challenge closed-source giants like OpenAI?
Related: China Launches Rival Satellite Network
Distillation, a practice where models are simplified to mimic larger ones, has sparked controversy. Wang calls it “like summarizing a book.” He argues it’s common in research, even among U.S. companies. Anthropic and OpenAI have accused Chinese firms of stealing knowledge. Wang sees irony in that: “Those who didn’t generate knowledge are trying to stop others from reusing it.” He stresses that AI-generated content should be free of copyright, or else powerful entities could exploit it.
Monetizing open-source models is tricky. The logic? Once a model is open, users must invest time to run it. The lab that launched it gains an early advantage. Another strategy: release fine-tuned models for free while keeping base models proprietary. This builds brand recognition, making it easier to attract top researchers.
Some Chinese labs are tightening licenses. Wang sees this as fair. “If you’re a cloud provider, you can’t run my model for free and pocket all the revenue.” This shift aims to prevent free-riding, ensuring labs are compensated for their work. But risks remain: if open-source labs can’t monetize, they may retreat from openness, weakening the ecosystem.
Related: India Sees Opportunity in US AI Gap
Wang warns of a possible pivot. If Chinese labs can’t find sustainable revenue, they might abandon open-source. However, he’s optimistic. Capital markets could keep these labs afloat, preserving the open-source momentum.
For U.S. startups, the advice is pragmatic: find a model that fits the market first. Closed-source models often serve as launching pads, helping startups gather data and refine products. Once they’ve built a user base, switching to open-source could cut costs dramatically. “You save hundreds of times on tokens,” Wang said, highlighting the long-term benefits.
Related: Party Games To Play During Covid-19 Lockdown
Recent trips to China reveal rapid AI maturation. This aggressive adoption contrasts with the U.S., where high token costs stifle experimentation. Uber and Microsoft have both voiced concerns about expenses. In China, open-source models are cheaper, making mass adoption feasible.
Wang predicts a surge in AI innovation over the next two years. If Chinese open-source models cross a usability threshold, growth could be exponential. The U.S. leads in AI exploration, but China’s lower costs and faster adoption may reshape the field. Whether open-source can rival closed-source giants like OpenAI depends on how well labs balance collaboration with profitability—a challenge neither side has fully solved yet.