A startup claims it broke through a bottleneck that’s holding back LLMs

Miami-based startup Subquadratic has emerged from stealth claiming to have resolved a decade-old mathematical bottleneck that restricts the efficiency of large language models. While the company has provided few technical details, the development addresses the high computational costs and memory requirements that currently limit how long or complex AI interactions can be. Whether the company can validate these performance claims remains to be seen, but a successful implementation could fundamentally alter the scalability of transformer-based architectures.