Elon Musk said he is testing a new recommendation algorithm for X that runs on Grok 4 Mini, a compact version of the large-language model developed by his artificial-intelligence venture xAI. Musk described the updated feed as “definitely better,” marking the first public confirmation that Grok is being woven directly into the social network’s core ranking system. Rolling the algorithm out to all users will require roughly 20,000 GPUs, Musk added, and will introduce higher latency in generating posts. He argued, however, that “the juice is worth the squeeze,” signalling a willingness to absorb additional infrastructure costs in exchange for what he sees as a superior, more personalized feed. The test comes as xAI ramps up compute capacity on several fronts. A day earlier the startup said it had begun training video models with Nvidia’s latest chips, and a forthcoming X/xAI facility in Atlanta is expected to operate more than 32,000 GPUs, including H100 processors. The expanded hardware footprint underscores the growing demand for generative-AI workloads across Musk’s portfolio.