- Chinese AI models from DeepSeek and Moonshot AI have quietly become embedded in daily workflows across Silicon Valley — used by companies large and small as cheaper supplements or alternatives to OpenAI and Anthropic’s offerings — a development that has largely flown under the public radar even as Washington debates AI export controls, creating a paradox where US companies are now dependent on Chinese AI while the US government simultaneously restricts Chinese access to American chips and models.
- Beijing is considering tightening its controls on homegrown AI models following their unexpected success with foreign users, according to the WSJ — a potential move that would mirror the US government’s approach of using export controls as a tool of technological statecraft, and that could force Silicon Valley companies to rapidly find alternatives to the Chinese models they have integrated into production workflows.
- The commercial appeal of Chinese AI models in the US market reflects a genuine capability and cost gap: DeepSeek’s models in particular were trained at a fraction of the cost of comparable Western models and have performed competitively on benchmarks, offering enterprise users a cost-efficient option that is difficult to ignore when procurement teams are scrutinizing AI spending; Moonshot AI’s long-context capabilities have similarly attracted enterprise attention for document processing tasks where cost-per-token is critical.
- The situation creates a structural policy tension: the US has restricted Chinese access to advanced Nvidia chips specifically to limit China’s AI development, yet American companies are simultaneously adopting the AI models that Chinese labs built — potentially creating dependencies on Chinese AI infrastructure that could be weaponized by Beijing as a leverage point in the broader technology competition, much as Hormuz gives Iran asymmetric leverage over global energy markets.
What Happened?
The Wall Street Journal reports that American companies across Silicon Valley have integrated Chinese AI models — particularly from DeepSeek and Moonshot AI — into their core business workflows, using them as cheaper alternatives to OpenAI and Anthropic. These models have gained adoption for their cost efficiency and competitive performance despite US-China technology tensions. Beijing has noticed the foreign success of its homegrown AI and is now in consultations with companies about potentially restricting the export or accessibility of these models — a move that would effectively weaponize Chinese AI the same way the US has weaponized chip export controls.
Why It Matters?
This story reveals a significant blind spot in US AI policy: while regulators have focused intensely on restricting Chinese access to American semiconductor and AI technology, American companies have been quietly building dependencies on Chinese AI models that could now be subject to Beijing’s own export controls. The dependency is real — companies that have integrated DeepSeek or Moonshot AI into production pipelines would face disruption and switching costs if those models were restricted. Beijing’s potential controls are also a signal that China is beginning to think about AI as a strategic asset to be protected and potentially leveraged, not just developed — a maturation of Chinese AI policy that parallels the US approach.
What’s Next?
Watch for whether Beijing formalizes any export control framework for domestic AI models — the consultation phase suggests this is policy development in progress rather than an imminent restriction. American companies using Chinese AI models should be developing contingency plans and alternative sourcing now, while switching costs are manageable and before any restriction creates emergency replacement pressure. For US policymakers, the WSJ report raises an uncomfortable question: should the US government be concerned about Silicon Valley’s adoption of Chinese AI models on national security grounds, and if so, does that require domestic regulation of enterprise AI sourcing? The same scrutiny applied to Chinese networking equipment (Huawei) and Chinese social media (TikTok) could logically extend to Chinese AI models embedded in enterprise workflows.
Source: The Wall Street Journal
