Following its 2026 technology conference, Deutsche Bank pointed out that there are currently no clear signs of a cooling down in AI computing power demand. Investors' expectations for Hyperscaler capital expenditure in 2027 have risen to $1.2 trillion to $1.5 trillion, significantly higher than the approximately $800 billion in 2026. Meanwhile, the sources of demand are expanding from large cloud providers to Neocloud, enterprises, sovereign AI projects, and emerging AI labs, and the breadth of AI infrastructure spending continues to expand.
Currently, the main constraints in the industry chain are concentrated on memory, and to some extent, power supply. Regarding next-generation processors, the focus of discussions among companies has shifted from simply increasing computing power to how to obtain more memory supply, optimize chip architecture, and continuously reduce the cost per token for inference.
This means that the next stage of incremental growth in the AI industry chain may shift more towards HBM/DRAM, power infrastructure, and next-generation architecture. As long as capital expenditure continues to expand, market judgments on AI need to simultaneously observe the supply side: whether memory prices, supply capacity, and power supply can keep up may be more important than simply discussing whether GPU demand has peaked.
