
DeepSeek
Chinese AI lab releasing open-weight frontier models at a fraction of US training cost.
Last refreshed: 19 April 2026 · Appears in 1 active topic
How did DeepSeek change the calculus for European sovereign AI?
Timeline for DeepSeek
Mentioned in: UK names first Sovereign AI investees
European Tech SovereigntyBackground
DeepSeek is a Chinese AI research laboratory backed by the quantitative hedge fund High-Flyer, which released the DeepSeek-R1 and DeepSeek-V3 models in late 2024 and early 2025. These open-weight models benchmarked at or above GPT-4-class performance at a reported training cost of around $5-6m — dramatically lower than comparable US frontier models — and were released under permissive licences allowing commercial use and fine-tuning. Cosine, the UK Sovereign AI Fund cornerstone investee, has outperformed DeepSeek on independent coding benchmarks for two consecutive years .
DeepSeek's releases reset assumptions about the capital and compute required to reach frontier AI performance. For European sovereignty policy, DeepSeek is a double signal: it demonstrates that frontier AI is achievable without US-scale data centre investment (strengthening the case for European alternatives), but it also demonstrates that a non-Western actor can produce models at the frontier (heightening the strategic anxiety that drives both EU and UK sovereign AI programmes). The EU AI Act's GPAI obligations will apply to DeepSeek models deployed in the EU market from 2 August 2026.
The Cohere/Aleph Alpha merger talks in April 2026 and the UK Sovereign AI Fund investments in the same week are both partly responses to the DeepSeek moment: consolidation and public investment as answers to Chinese open-weight competition .