Sequoia Partners Warns US Open Model Base is Shifting to China
Sequoia Capital partners Dean Meyer and Konstantine Buhler stated that while the US maintains a lead in closed-source frontier models, Western companies are increasingly relying on Chinese open-weight models as a product base, training teachers, and synthetic data sources.
Qwen, Kimi, GLM, and DeepSeek are widely adopted, with an ATOM report showing Qwen's share of newly added fine-tuning and adaptation models rising from 1% in January 2024 to 69% in February 2026.
Market mechanisms indicate that China's open weights lower the global developer entry threshold, driving a surge in adoption, while US companies legally distill their outputs to train new models, and local closed-source vendors restrict similar usage, causing funds and talent to flow towards the lower-cost open ecosystem, benefiting Chinese model providers.
Source: Public Information
ABAB AI Insight
Sequoia has previously invested continuously in AI infrastructure and application layers, and Dean Meyer and Konstantine Buhler have long observed the global model ecosystem. This public warning continues their focus on technological sovereignty and supply chain risks.
In terms of capital pathways, US labs can achieve a balance of revenue and compliance through controlled sales of training rights, while Chinese open models rapidly accumulate data flywheels and developer networks through community contributions and low-cost access, concentrating resources on high-adoption platforms.
Similar to how the early open Meta Llama series promoted ecosystem expansion, Chinese models like DeepSeek accelerate the global fine-tuning market. The US is currently at a bifurcation stage, leading in closed-source but lagging in open bases.
Essentially, this is about technological substitution and capital concentration. China's open weights achieve low-cost diffusion through distillation mechanisms, while Western restrictions accelerate local companies' shift towards external teacher models. The mechanism is that differences in intellectual property and access rules lead to a reconstruction of training data and computing power pathways towards open sources.
ABAB News · Cognitive Laws
The higher the closed-source moat, the faster the penetration of open bases.
Distillation is a knowledge transfer tax; whoever opens the teacher model collects long-term rent.
Leading models sell answers, open models sell classrooms, and ultimately classrooms define the industry.