Opinion
China just dropped a 2.8-trillion-parameter bomb on the AI race
The next stage of the AI race will not be decided only by who builds the smartest model. It will also be decided by which models companies trust enough to embed in their products and operations.
Moonshot AI’s new Kimi K3 is a 2.8-trillion-parameter, natively multimodal model with a one-million-token context window. It is designed for long-horizon coding, complex knowledge work and deep reasoning. Moonshot also plans to release the model weights, potentially allowing developers and companies to inspect, adapt and deploy it outside Moonshot’s platform
Those specifications alone do not make K3 the world’s best model. Parameter count is not the same as intelligence, and company benchmarks should be treated cautiously. Nevertheless, early evaluations suggest that K3 is competitive with leading American systems like Anthropic's Claude and OpenAI's ChatGPT, particularly in front-end coding and complex agentic tasks.
That is what makes the release significant.
For years, the AI race was described as one in which the United States held a durable lead through superior chips, computing infrastructure, research talent and access to capital. China faced restrictions on advanced semiconductors and was frequently portrayed as remaining several steps behind the frontier.
Kimi K3, following the momentum created by DeepSeek and other Chinese developers, suggests a different reality. China may not lead on every benchmark, but its companies are producing models that are increasingly capable, comparatively affordable and often more openly available.
A model does not need to be the strongest in every category to reshape the market. It needs to be good enough to adopt, affordable enough to deploy at scale and flexible enough to integrate into products and internal systems.
For high-tech companies, this is not merely a geopolitical story. It is a strategic, operational and legal question.
Companies should first reconsider their dependence on a small number of American model providers. Competitive Chinese models may offer lower costs, greater customization and local deployment. They could provide leverage in negotiations with existing vendors and reduce technological lock-in.
At the same time, lower prices and open weights do not eliminate risk. Before integrating a Chinese model, companies should examine where it was developed, how it was trained, what license applies and whether the model or its provider may become subject to sanctions, export controls or procurement restrictions. A model that is technically attractive today may later become difficult to use with government customers, regulated industries or international partners.
Security and data governance also require careful analysis. Local deployment may keep sensitive information within the company’s infrastructure, reducing certain privacy and confidentiality concerns. However, it also transfers more responsibility to the company. The organization must secure the model, control access, monitor outputs, manage updates and test for vulnerabilities, bias and unreliable behavior.
High-tech companies should therefore avoid choosing a model based only on benchmark rankings or API price. They need a structured selection process that considers performance, total deployment cost, licensing, cybersecurity, data location, business continuity and regulatory exposure. Contracts with customers and investors may also need to address which models are used, where they originate and whether they can be replaced if legal or commercial conditions change.
The next stage of the AI race will not be decided only by who builds the smartest model. It will also be decided by which models companies trust enough to embed in their products and operations.
The gap between the United States and China is narrowing. For technology companies, the question is no longer whether Chinese AI matters, but how to evaluate it without trading short-term savings for long-term risk.
Efrat Shuster is the Founding Partner of Shuster Law Firm.
