A16z’s Martin Casado reassesses AI risks as scaling laws refuse to break

For years, skeptics have predicted that AI scaling laws would hit a wall. Martin Casado, general partner at Andreessen Horowitz, has a different read: the wall keeps moving, and that’s not entirely good news. Casado has spent much of 2026 making the case that continued torrents of capital flowing into AI have kept scaling laws intact. The basic idea behind these laws is straightforward: pour more compute, more data, and more money into training larger models, and you get predictably better results. That predictability has turned AI development into what Casado calls a “scale-up capital game,” where dollars invested correlate more directly with outcomes than in almost any other technology sector. The resource concentration problem Because this relationship has held up under sustained investment pressure, a small number of pioneering AI labs have been able to translate massive capital reserves into capabilities that smaller competitors simply cannot match. Compute power, talent, and funding have all concentrated among a handful of firms, creating what amounts to an oligopoly in frontier AI development. Casado has flagged this concentration as one of the more tangible risks in the cur...