Google's TurboQuant, a compression algorithm unveiled this week via a Google Research paper, can make LLMs' memory usage six times smaller, potentially reducing data center energy usage and enabling powerful AI models to run on smartphones. The algorithm follows the trend of smaller AI models like China's DeepSeek, which required less data center energy and performed well on benchmarks despite being built on Meta's Llama. TurboQuant could help LLMs use data centers more efficiently, either by running more complex models or reducing the need for new data centers. This efficiency poses a potential problem for NVIDIA, whose stock has risen on CEO Jensen Huang's assumption of "the largest infrastructure buildout in history" for data centers. However, a New York Times investigation shows that data center construction is already stumbling due to local opposition, permits, inspections, and a lack of power generation and transmission.