GERMANY: AI DEMAND CONTINUES TO EXPAND, MEMORY AND ELECTRICITY BECOMING NEW BOTTLENECKS

Following the 2026 Science and Technology Conference, the Deutsche Bank noted that there were still no clear signs of cooling in AI ' s computing needs. Investors' expectations of capital expenditure in Hyperscaler in 2027 have risen to 1.2 trillion-1.5 trillion United States dollars, significantly above about $800 billion in 2026. At the same time, the sources of demand are spreading from large cloud manufacturers to the Neocloud, Enterprise, Sovereign AI and the Emerging AI Laboratory, and the breadth of AI infrastructure expenditure continues to expand. The main constraints of the current industrial chain have been concentrated in memory and, to some extent, in electricity supply. In the context of the next generation of processors, the focus of business discussions has also shifted from a simple expansion of computing power to more memory supply, the optimization of chip structures and the continuous reduction of the cost of reasoning per Token. This means that the next phase of the AI industrial chain may be subject to more migration to HBM/DRAM, electricity infrastructure and next generation structures. As long as capital spending continues to expand, the market's judgement of AI needs to look at the supply side at the same time: memory prices, supply capacity and the ability of electricity to keep pace may be more important than simply discussing GPU demand. Kim Xian