Global | Measuring Technological Progress in Real Time with ArXiv

Summary

This report presents a novel method to track technological progress in real time using data from ArXiv, the world's most popular preprint repository in computer science, physics and mathematics.

Key points

Key points:
Understanding the process of science creation is key to monitoring the rapid digital transformation. Using natural language processing techniques (NLP), at BBVA Research we have developed a new set of technological indicators using data from ArXiv, the world's most popular open-access preprint repository in physics, mathematics and computer science.
Our monthly indicators track ArXiv preprint publications in computer science, physics, and mathematics to deliver a granular and real-time signal of research activity, thus of technological progress and digital transformation. These metrics highly correlate with traditional proxies of research activity such as patents.
Computer science preprints have exploded from under 10% of all ArXiv submissions in 2010 to nearly 50 % in 2025, driven predominantly by AI research now representing about 70% of all computer science output. Conversely, research on physics and mathematics have grown in parallel at a steady rate.
There was a turning point in the computer science paradigm around 2014, with the research focus drastically moving from classic computing theory (e.g. information systems, data structures and numerical analysis) to AI-related disciplines (machine learning, computer vision, computational linguistics and others). Research on cryptography and security was the only subfield essentially unaffected by the AI boom.
Scientific innovation plays a crucial role in shaping productivity trends, labor market dynamics, capital flows, investment strategies, and regulatory policies. These indicators provide valuable insights into our rapidly evolving world, enabling early detection of technological shocks and informing strategic investment decisions.