Google has introduced TurboQuant, a compression algorithm that reduces large language model (LLM) memory usage by at least 6x while boosting performance, targeting one of AI's most persistent bottlenecks: memory. The breakthrough lowers inference costs and expands deployment across cloud and edge environments.
Taiwan's Legislative Yuan Internal Administration Committee held its second public hearing on proposed amendments to the National Security Act on March 26. Attendance from civil society groups and academic experts was limited, and only a small number of lawmakers spoke.
SK Hynix plans to build more than KRW100 trillion (approx. US$66 billion) in net cash to support long-term investment in artificial intelligence (AI) infrastructure, as surging demand reshapes the memory industry and drives a new round of capital spending.
According to Vic Wu, General Manager of the Specialist Team Unit (STU) of Microsoft Taiwan, the era of exploratory AI is over. As the industry enters the second quarter of 2026, it is no longer asking if AI works, but how to manage the increasing volume of digital labor now entering the global workforce.
