80% of a16z-Backed Startups Now Utilize Chinese Open-Source AI Models

Silicon Valley is witnessing a significant shift in artificial intelligence (AI) model adoption, with a substantial majority of startups backed by venture capital firm Andreessen Horowitz (a16z) reportedly opting for Chinese open-source models over proprietary American alternatives. This trend is primarily driven by drastic cost efficiencies, as highlighted by Aakash Gupta in a recent social media post. Martin Casado, a general partner at a16z, confirmed this observation, stating there's a "high chance" that 80% of pitching startups are leveraging these Chinese technologies.

The economic disparity is stark, with Chinese open-source models like DeepSeek costing significantly less per token. Gupta pointed out that these models are "214x less per token" compared to offerings from companies like OpenAI. For a startup processing 100 million tokens monthly, the difference translates to approximately "$1,400 versus $300,000," a critical factor that can extend a company's runway from three months to eighteen.

This financial advantage stems from vastly different development costs; DeepSeek reportedly trained its model for $5 million, while OpenAI's GPT-5 training cycles are estimated at $500 million. Gupta asserted that the real constraint in AI was never capability, but rather burn rate, a challenge he believes China addressed by prioritizing efficiency over the pursuit of Artificial General Intelligence (AGI). Chinese models are increasingly matching the performance of GPT-4 on coding benchmarks, despite their lower cost.

The reduced infrastructure costs enable a crucial second-order effect: the ability for startups to fine-tune models for specific use cases. American startups, burdened by high API rates from proprietary models, are often limited to generic applications. In contrast, users of Chinese open-source models can develop specialized variants, fostering greater innovation and tailored solutions.

Anjney Midha, another partner at a16z, acknowledged China's growing dominance in the open-source AI landscape, reportedly stating, "it’s really China’s game right now." This sentiment underscores a strategic divergence, where American AI labs focus on achieving AGI, while Chinese labs prioritize widespread distribution and adoption by making AI infrastructure affordable. Gupta noted that "all 16 top-ranked open-source models are Chinese," including DeepSeek, Qwen, and Yi, which are being deployed at scale as free alternatives to premium services.

Looking ahead, this economic imperative is expected to shape the future of AI development. Gupta predicted that "every startup that survives the next funding winter will have optimized around Chinese open-source as default infrastructure." He concluded that for seed-stage founders, "economics beats nationalism every time," suggesting that while America may still strive to build the "best model," China has already "won the race to build the one everyone uses."