Signal Training GPT-4 and Gemini Ultra cost a reported $78 million and $191 million
Summary
This signal puts a price on training the most advanced AI models. The report states that OpenAI's GPT-4 required an investment of $78 million and Google's Gemini Ultra $191 million. Such costs are a major barrier for smaller organisations and regions with little funding, and research into cheaper training, hardware optimisation and sustainable computing is lagging. By contrast, DeepSeek built high-performing models for around $5.5 million, showing that resource-conscious innovation is possible. The report uses this contrast to argue that cost, more than capability, shapes who can take part in frontier AI. It also ties the figure to the widening gap between industry and academia.
Classification
Evidence 1
- The impact of Artificial Intelligence on our societies European Commission, Directorate-General for Research and Innovation (Directorate D - People: Health and Society, Unit D3 - Health and Societal Transitions); edited by Ewa Luger and Marianthe Stavridou page=9;section=2.2 Identifying gaps in existing research 2026 accessed 2026-07-26
Part of trends 0
No objects.
Directly linked issues 1
- IssuePublic research cannot follow frontier AI, and industry contracts bind the academics who do1 trends · 2 signals
Relation types: direct_urgent
Public id: fm-08af111a9079
