It may be niche, but it's a big niche in a data-driven world.
TurboQuant is a compression algorithm introduced by Google Research (Zandieh et al.) at ICLR 2026 that solves the primary memory bottleneck in large language model inference: the key-value (KV) cache.