Methodology for EU AI Startup Market Analysis

For the value of each funding round R, we distribute the total sum equally among all participating investors N, and by summing these values across all rounds, we estimate the total investment F by each funder j in the target company.

A limitation of this methodology is the assumption of equal contribution among investors participating in a round, which may underestimate the influence of lead investors while overstating the contribution of follow-on investors. As a result, the funding structure is assumed to be more polarized than this model suggests. Typical estimates suggest that the lead investor might commit up to 20–50 percent of a particular round. For visual clarity, the visualization also excludes investments from regions with only minor stakes in the EU companies. 

For the compute infrastructure, we traced the compute providers of the startups from public sources and then explored what hardware infrastructure the respective compute providers use. The key limitation of this methodology is the lack of information of the relative weight of the different chips in computing clusters in the case of several providers. Moreover, companies might have further, alternative compute providers for which information is not available. Considering the path dependencies between hardware and the large scale AI models, we do expect these potential missing compute providers to use the same hardware as other compute providers of a target company. 

The funders were located geographically on the basis of the location of primary headquarters, and the classification of investor types was done based on the contextual information about the nature of the company.

Source Data EU AI Startup Market Analysis

Acquired by AMD in July 2024