AI Barcelona

 The Post-Chinchilla Era: The Fundamental Shift to Test-Time Compute For the first half of the 2020s, the trajectory of artificial intelligence was governed by the Chinchilla scaling laws, a set of empirical observations suggesting that intelligence was a direct function of three variables: parameter count, dataset size, and training compute. This paradigm fueled the "bigger is better" arms race, leading to the creation of monolithic models that required massive, energy-intensive training runs spanning months. However, as we move through 2026, the industry has hit a point of diminishing returns in the pre-training phase. Data exhaustion—the depletion of high-quality human-written text—and the escalating costs of compute have forced a pivot. We are no longer seeing the same exponential gains from simply adding more layers to a transformer architecture. Instead, the frontier of intelligence has shifted from the pre-training phase to the inference phase. This shift is character...

Sovereign Cloud, Hyperscalers, and EU Rules in 2026: Building the Infrastructure for Europe’s AI Future In 2026, “sovereign cloud” is no longer a slogan but the main lens through which Europe thinks about data infrastructure, AI capacity, and strategic autonomy. At the same time, US hyperscalers are racing to rebrand parts of their footprint as sovereign‑ready, testing how far EU rules can bend without surrendering core political control over data. For executives and policymakers, the question is less whether cloud will be sovereign, and more who gets to define what sovereignty means in practice. The answer will shape where future AI models are trained, who can audit them, and which jurisdictions hold leverage over the most valuable datasets in the world. What “Sovereign Cloud” Means in 2026 At EU level, sovereign cloud has crystallised into a three‑layer idea: technical controls, operational autonomy, and legal insulation from non‑EU jurisdictions. It is not just about where ...