A growing number of Swiss companies are experimenting with Chinese open-weight AI models, drawn by significant cost savings but wary of data security and ideological bias risks. Oxygen at Work, a Zurich-based firm that installs office plants and environmental sensors, reports saving around 40 percent of its AI costs by using Kimi, a model developed by Chinese startup Moonshot AI, while strictly barring it from any personal data, passwords, or sensitive backend systems. The release of Kimi K3 — which now rivals leading US models from OpenAI and Anthropic in benchmark tests, despite being trained on less powerful chips due to American export restrictions — has intensified interest, but industry groups such as Swiss AI note that most member companies remain hesitant, citing concerns over Chinese-influenced training data, potential propaganda embedded in models, and the technical overhead of running them securely on local servers.