DeepSeek V4 Intensifies U.S.-China AI Race With Open-Source Gambit
China's cheaper, open-weight model challenges American dominance as accusations of distillation attacks and chip smuggling mount.
DeepSeek V4's release has sharpened the technological rivalry between Washington and Beijing, shifting the competition from raw performance to global adoption.
Performance Gap Narrows, But U.S. Leads
DeepSeek V4-Pro features 1.6 trillion total parameters with 49 billion activated per token using mixture-of-experts architecture, plus a one-million-token context window. Expert estimates cited by the Council on Foreign Relations indicate the U.S. remains about seven months ahead of Chinese models in raw performance, though this is a contested and time-sensitive estimate.
API pricing runs approximately four to ten times cheaper than U.S. competitors—roughly $0.27 to $0.435 per million input tokens compared to $2.50 to $15 for American models. DeepSeek's weights are open-source under MIT License and available on Hugging Face.
DeepSeek has acknowledged compute capacity limitations for serving V4-Pro at scale.
Allegations of Illicit Methods
U.S. government officials and major AI laboratories including Anthropic, OpenAI, and Google have accused DeepSeek of conducting distillation attacks to extract capabilities from American models. Anthropic specifically accused DeepSeek, Moonshot AI, and MiniMax collectively of using more than 24,000 fake accounts and over 16 million interactions; DeepSeek's individual share was not separately specified in public reports.
U.S. officials have also alleged that DeepSeek used smuggled Nvidia Blackwell chips for training, part of broader Chinese efforts to bypass American export controls on high-end semiconductors.
The Adoption Race
The competition has evolved beyond an arms race of raw benchmarks into what analysts call an adoption race. The ability to deploy good-enough solutions quickly and cheaply—particularly in the Global South—has become a primary driver of geopolitical influence.
"The country that deploys AI fastest across its economy and government will shape the character of the AI era," said Michael C. Horowitz, a senior fellow at CFR.
Strategic Vulnerabilities
The situation exposes a critical weakness in American AI dominance: while U.S. labs produce the most capable models, China is leveraging open-source distribution and aggressive cost-reduction strategies to capture market share during the deployment phase.
If Washington cannot protect intellectual property from industrial-scale distillation and close chip-smuggling loopholes, its lead may remain measured in months rather than years—allowing Beijing to shape the global AI ecosystem even without technical supremacy.