Can AI capex extend the semiconductor cycle?

AI demand is broadening beyond GPUs, with specialized memory chips, CPUs, and ASICs now playing a larger role in supporting both training and inference.

That wider chip mix, along with continued hyperscaler capex and in-house silicon efforts, suggests the semiconductor upcycle could last longer than many investors expect.

Memory remains a major source of upside, as tight HBM and DRAM supply should support stronger pricing and earnings through 2027.

At the same time, power components for data centers are becoming a key bottleneck, which keeps analog chipmakers well positioned.

The main risk is that supply eventually catches up to demand, but for now constrained capacity still favors leading compute, memory, and power names.