AI compute demand drives semiconductor boom

Global AI token generation has surged more than 330‑fold in two years, prompting a rush for GPU capacity and prompting AI infrastructure firms to shift focus from raw hardware specs to measurable output. Taiwanese company Zettabyte introduced a “quality compute” framework that gauges the useful work – or “goodput” – delivered by GPU clusters, emphasizing reliability, readiness, observability, cost transparency and energy efficiency.

A separate industry report shows that a single AI server cabinet now houses over 4,500 chips, with semiconductors accounting for more than 95% of the cabinet’s value and over half of AI data‑center capital spending. Investment in AI data centers is projected to total $4 trillion between 2023 and 2030, with roughly $2.8 trillion flowing to semiconductor and related hardware. Semiconductor revenue tied to AI data centers is expected to jump from $27 billion in 2022 to $12 trillion by 2028, a 40‑fold increase, while the sector’s growth shift moves from raw compute power to memory bandwidth, interconnect speed and energy‑efficient designs.