The news: US startups are rapidly adopting Chinese open-weight AI models to cut costs, ship features faster, and keep data on-device—putting pressure on closed, pay-per-user systems from OpenAI and Google, per NBC News.

Startup founders say Alibaba’s Qwen and DeepSeek R1 are now getting close to the performance and quality of leading proprietary models like GPT-5 and Gemini 2.5. This sentiment was echoed at Tech Week and Tech Futures events, where early-stage teams credited DeepSeek as a major accelerator for product development.

Why it matters: Free, customizable models such as DeepSeek R1, Z.ai, and Qwen lower the barrier for early-stage builders, letting brands and teams develop advanced features or experiment with AI tools at a fraction of traditional subscription costs. 

This shift to open systems aligns with a broader enterprise trend—43% of US technology leaders plan to adopt open-source AI models in the next 12 months, per Forbes. 

That places open-source AI alongside top-tier data investments like vector databases (45%) and knowledge graphs (46%). Open ecosystems are no longer experimental and are considered necessary for the core AI stack.

Adoption is spreading beyond startups: Circlemind AI told NBC that Chinese models now lead online dev resources. Airbnb CEO Brian Chesky echoed this, calling Qwen “very good … fast and cheap,” even as the company considers integrating ChatGPT.

China’s way in: Early use of open Chinese models gives their creators rare influence over Western AI workflows. While US startups rely on these open models to build products, their real-world usage data helps refine the next wave of Chinese systems—accelerating their pace and influence.

The caveat: Chinese-built models introduce questions about data governance, IP exposure, and long-term support. Brands will need clear guardrails like vendor audits, on-device constraints, and internal red-team testing to avoid hidden risks that could risk workflows and campaigns.

What this means for brands: The next wave of AI-development will come from teams mixing closed systems with fast, inexpensive open models—many of them from China. This hybrid approach lets marketers test ideas quickly, tailor models to their data, and drive down cost per interaction.

The path forward is simple: Use open weights for rapid prototyping in sandboxed, privacy-safe environments, then move the best ideas into regulated, closed stacks to scale with confidence and maximize ROI.