AFEP confirms the authority’s view of the resources needed to develop foundation models – significant computing power, large amounts of high-quality data, and highly skilled human resources – while pointing out that the AI sector is a particularly fast-moving sector, and that the recent emergence of new high-performance models seems to be based on a lower volume use of these resources. Above all, the development of AI systems is also characterised by the particularly high level of investment required.
- In terms of computing power, only a handful of companies have the resources to fully train a foundation model, whereas most companies use these models, adapting them to their requirements. Upstream, graphics processing units are essential to the development of generative AI insofar as they offer the greatest computing capacity.
- With regard to data, AFEP observes that data quality is even more important than the volume of data available for developing an AI system. On this point, companies consider that access to quality data may constitute a barrier to market entry.
- When it comes to technical skills, the major players in the digital sector benefit from a number of factors that help them attract the best talent: the reputation of innovative companies, stimulating projects, virtually unlimited resources and tools, and particularly attractive remuneration packages.
- Regarding the functioning of the market, AFEP member companies note that the French market is largely dominated by foreign and non-European players, who offer cloud services that are essential for the development of AI systems. There are still some French players, but they are suffering from structural market barriers and commercial practices implemented by vertically integrated operators.