[DigitalToday intern reporter Seung-a Yoo] Chinese artificial intelligence (AI) companies are releasing high-performance open-weight models for free, shaking U.S. big tech’s closed AI strategy.

On July 27 (local time), IT outlet The Verge reported that Moonshot AI’s Kimi K3 is being assessed as capable of delivering performance comparable to some top-tier models from U.S. companies at lower cost. The contest is spreading beyond a technology race into a battle for control of the developer ecosystem.

The key is the method of release rather than performance itself. Moonshot AI chose to distribute Kimi K3’s model weights for free. Open-weight models differ from open source in the traditional sense, which discloses training data and the full architecture. But they offer broader use than closed services because developers can examine how a model works, run it on their own infrastructure and tailor it. Another strength cited is that developers can build new products without being locked into a specific provider.

That is also why U.S. AI companies are on edge. If developers and companies start building tools and services around models such as Kimi K3, the industry’s center of gravity could shift away from closed platforms such as Gemini, Claude and ChatGPT. A recent trend in which U.S. labs more strictly control access to the latest models and tighten safeguards is also cited as a factor increasing preference for open weights. There are also signs that some U.S. companies are moving to cheaper Chinese models.

Open weights being released for free does not mean there is no revenue model. Chinmayi Sharma (친마이 샤르마), a professor at Fordham University School of Law, said, "Free model weights are not a free AI service." She explained that even if a company releases the model itself, it can still make money from cloud infrastructure, engineering, security, maintenance and hosting-based access services. In some cases, expanding cloud usage or rising demand for advanced AI chips could bring an even bigger payoff.

China’s push for open weights is underpinned by both industrial strategy and practical constraints. With access to advanced chips and computing resources limited, an open ecosystem becomes a channel for Chinese firms to keep innovating close to the cutting edge. It also aligns with Beijing’s industrial policy to drive the spread of Chinese-made models, tools and infrastructure. It is also advantageous for expanding overseas technological influence. Chinese President Xi Jinping (시진핑) earlier this month cast China as a more equal partner while targeting the United States’ closed approach, in an extension of this trend.

In the United States, clashes over the direction of regulation are also growing. After speculation emerged that the United States could restrict access to open-weight AI following Kimi K3, 25 technology companies including IBM, Microsoft, Meta, Nvidia, Perplexity and Palantir urged policymakers to avoid hasty limits. They argued that open-weight AI is important to protecting U.S. AI leadership and preventing the power and benefits of the technology from being "concentrated in the hands of a few."

The debate grew on July 27. Major companies including Nvidia, Microsoft and SpaceX called for stronger U.S. support for open-weight models. The trigger was concern over advanced AI safety. An example was presented in which an OpenAI model that escaped control during testing attacked another company, and the other side said it had to defend using a Chinese open-weight model because strict safeguards on the most advanced U.S. models prevented it from using them. The case lent weight to arguments that open weights should be expanded rather than restricted.

Still, it remains unclear how far large U.S. AI labs will move toward openness. Google and OpenAI later voiced positions warning against hasty regulation of open models, but they did not take part in the cybersecurity-focused joint response announced on July 27. Anthropic joined neither of the two currents.

U.S. companies have not left the issue untouched. Kyle Miller (카일 밀러), a senior research analyst at Georgetown University’s Center for Security and Emerging Technology, said pressure from Chinese companies partly influenced OpenAI’s release last year of GPT-OSS. Google’s open-weight Gemma model family is also mentioned as a response to Chinese competition. But both companies still run their most powerful flagship models as closed systems.

Sharma suggested U.S. companies may move toward a combined approach of open and closed. She said the question from U.S. firms could increasingly shift to, "How much performance must we release to prevent Chinese models from becoming the base platform of the open ecosystem?" She added that a "portfolio strategy" may be more realistic, keeping the best models closed while also releasing increasingly powerful open-weight models to preserve developer adoption and ecosystem influence.

Ultimately, Kimi K3’s significance does not stop at a performance contest between individual models. China’s spread of open-weight AI is squarely raising the issue of whether closed AI can remain the market’s basic structure going forward, beyond the question of whether the United States can stay ahead of China.