#삼성전자

With Google's Gemini 3.0 in the spotlight, the computing power behind it, Google TPUs, is also in the spotlight. While Nvidia GPUs have been at the center of AI training and services, a new battleground is emerging: TPU vs GPU. Google's TPU is a customized chip designed for deep learning computation, so it has a high power-to-performance ratio and can reduce the cost of building large-scale AI data centers, which is why big tech companies including META are considering adopting TPUs. This change is also an important signal for domestic semiconductor companies. As the number of AI accelerators like Google TPU increases, the demand for high-bandwidth memory like Samsung Electronics HBM and SK Hynix HBM3E also increases. In the end, the rise of Google's TPU is likely to be the beginning of a rewriting of the rules of the entire AI infrastructure market, beyond the competition with Nvidia. #GoogleTPU #TPUvsGPU #NvidiaCompetition #Gemini3 #AIsemiconductor #SamsungSamsungElectronics #SKHynix

The AI semiconductor supercycle started with HBM, but it's D-RAM that's fueling the fire. Demand for both learning (HBM) and inference (DDR5-eSSD) has exploded simultaneously as big tech makes long-term investments in AI infrastructure for survival. This was coupled with a "perfect storm" of memory vendors shifting their lines to the more profitable HBM, structurally reducing the supply of commodity D-RAM. As a result, D-RAM prices have risen dramatically, and demand is spreading from the data center to electric vehicles, robots, and smart factories. This is not a cycle, but the beginning of an industry restructuring. #AI #Semiconductors #HBM #DRAM #Supercycle #Samsung Electronics #SK Hynix