Source: European Parliament
Question for written answer E-000512/2025
to the Commission
Rule 144
Alex Agius Saliba (S&D)
A shock wave went through the tech markets on 27 January 2025 with the release of Chinese-made DeepSeek AI, which seems to perform similarly to US tools such as ChatGPT.
DeepSeek’s claim that it used significantly fewer resources and microchips led to a drop in share prices. Enormous investments in AI hardware might be less essential to AI development than had been assumed.
It was reported that DeepSeek used distillation to learn from the OpenAI model and that this could be a breach of intellectual property (IP) rights – an ironic claim by a company that used swathes of IP-protected material to train its own model.
These developments need to be followed closely to be used as lessons for the upcoming proposals for the ‘AI Gigafactories’ and ‘Apply AI’ initiatives.
- 1.Has the Commission analysed DeepSeek’s claims that it operates an efficient AI system at lower cost and with fewer resources, including fewer advanced microchips, and does it deem these claims to be credible?
- 2.What are the lessons to drawn from DeepSeek’s approach when developing an AI ecosystem in the EU with a fraction of the budget available to the US big tech firms and how will these lessons be incorporated in the upcoming ‘AI Gigafactories’ proposal?
- 3.What is the Commission’s view on the method of distillation used to train AI models within the European IP framework?
Submitted: 5.2.2025