Over the past two years, AI hardware investment has focused on GPUs, HBM, and advanced packaging, but semiconductor testing is becoming increasingly important as chips grow more powerful and complex. Nomura believes AI chip upgrades are increasing the value of back-end manufacturing, with longer test times, tighter capacity, outsourced orders, and higher equipment requirements improving testing suppliers' strategic position. Using NVIDIA’s AI GPUs as an example, Nomura estimates that if Hopper’s final-test time is set at 1, Blackwell could require around 4 times as long and Rubin could require around 7 times as long. Testing costs as a share of total manufacturing costs could rise from approximately 1.9% for Hopper to 2.5% for Blackwell and 3.3% for Rubin. Advanced packaging creates a "yield cliff"; if each die has a 90% yield, combining ten dies would produce a theoretical module yield of only about 35%, making early testing critical. Co-packaged optics (CPO) adds complexity by combining electronic and photonic chips, requiring testing of electrical performance, optical performance, and communication between the two, potentially creating new demand for automatic test equipment, probe cards, optical instruments, sockets, and temperature-control systems if CPO moves into volume production.