China vs the US: A 2026 Comparison of Who Is Winning the AI Race

The 2026 AI race between the US and China reveals a complex landscape where investment disparity does not translate into proportional model superiority. The US spent $285.9 billion on private AI in 2025 versus China's $12.4 billion, yet the performance gap between leading models has narrowed from 17.5–31.6 percentage points in May 2023 to just 2.7 points, per Stanford HAI. While OpenAI, Anthropic, and Google DeepMind hold the frontier in absolute model capability, Chinese models like DeepSeek-V4 and Qwen have become highly competitive at dramatically lower costs. DeepSeek-V4, with 1 trillion parameters, performs comparably to GPT-5 at 1/50th the API cost; DeepSeek-R1, developed in two months for under $6 million, beat OpenAI's o1 on mathematical reasoning. Alibaba's Qwen family dominates open-source with 942 million downloads, and Baidu, ByteDance, and Tencent have aggressively increased open-source releases. In hardware, NVIDIA's Blackwell Ultra GB300 provides 15 petaflops, while China's best chip (Huawei Ascend 950PR) delivers only 1.56 petaflops, with yields between 5–20% compared to NVIDIA's 60–80%. SMIC remains at 7nm technology, two generations behind. Overall, the US leads in private capital, chip manufacturing, and model performance; China leads in research output, patents, and open-source proliferation. Neither side has an outright victory, and the race continues with converging capabilities.