US startups are increasingly building their artificial intelligence applications on Chinese open-source AI models, a trend highlighted by prominent venture capitalist Martin Casado. This development underscores a significant shift in the global AI landscape, where Chinese models are gaining traction due to their performance and cost-effectiveness.
"It's just remarkable how many US startups are being built on Chinese OSS AI models," stated Martin Casado, a General Partner at Andreessen Horowitz (a16z), on social media. He added, "I'd say the majority that are building custom models via post training." This observation points to a growing reliance on foreign foundational AI technology within the American startup ecosystem.
Chinese open-source models, such as Alibaba's Qwen and DeepSeek, have achieved high rankings on global leaderboards, often surpassing Western counterparts like Meta's Llama 3 and Microsoft's Phi-3 in performance benchmarks. For instance, San Francisco-based Abacus AI has reportedly adapted and fine-tuned Alibaba's Qwen model, demonstrating practical adoption. These models offer competitive performance at significantly lower costs, making them attractive to startups with limited resources.
Casado, a seasoned investor and former founder of Nicira, a networking infrastructure company acquired by VMware, is a vocal proponent of AI innovation. His commentary often emphasizes the need for robust technological advancement and criticizes what he perceives as overly restrictive regulations. His recent remarks reflect a concern that the US is lagging in the foundational open-source AI space.
"The US needs to step up, make it a national priority and back with a huge investment," Casado urged in his tweet. This call aligns with broader geopolitical competition in AI, where both the United States and China view AI leadership as crucial for economic prosperity and national security.
While the US government has initiatives like the National AI Initiative Act and the CHIPS and Science Act to boost domestic AI research and development, its approach is largely private-sector driven. In contrast, China employs a state-led strategy with substantial government funding and national programs. US export controls on advanced semiconductors, intended to slow China's AI progress, have inadvertently spurred Chinese companies to innovate more efficiently and embrace open-source solutions, further accelerating their global influence in this critical technology domain.