1. For most of 2026 and 2027, AI chips (GPUs, CPUs, memory, storage devices) will remain the primary bottleneck in AI computing infrastructure. However, semiconductor production capacity is rapidly increasing, and the industry has significant expansion flexibility given sufficient lead time.
2. Sometime within the next 18–24 months, due to delays in data center construction and deployment, as well as limitations in power supply (including grid power and on-site/independent power solutions), chips may no longer be the core factor limiting AI computing power.
3. Although data center deployment has exceeded expectations so far and has not hindered AI cloud deployment and token generation, power and infrastructure bottlenecks are expected to become key variables that the market needs to focus on in the second half of 2027 and 2028. (The above views are from a JPMorgan report dated August 5th.)
