One Year Since the “DeepSeek Moment”

Source: HuggingFace Date: 2026-01-20 URL: https://huggingface.co/blog/huggingface/one-year-since-the-deepseek-moment

Summary

Research retrospective: one year after DeepSeek R1 (January 2025), HF analyzes the ecosystem shift. R1 lowered three barriers simultaneously — technical (shared reasoning paths and post-training methods), adoption (MIT license), and psychological (reframed the question from “can we?” to “how well?”). Chinese organizations exploded on the Hub: Baidu went from 0 to 100+ releases, ByteDance and Tencent 8–9x. Chinese models now surpass U.S. models in new download share; top-liked Hub models are no longer majority U.S.-developed.

Implications

Thread: open-weights ecosystem health. The DeepSeek moment’s durable impact wasn’t the model — it was the normalization of sharing training methodology alongside weights. Post-training recipes for reasoning became community property rather than proprietary differentiation. The Chinese ecosystem growth numbers are stark: the Hub is no longer a primarily Western-origin resource. The downstream tension (Western organizations seeking non-Chinese alternatives for commercial deployment) creates a market gap that European and other regional open-weight efforts are trying to fill. Competition has shifted from benchmark comparisons to ecosystem and system-level capabilities — a harder moat to measure.