AI ASIC shipments projected to surpass GPU volume by 2028

The AI hardware landscape is undergoing a significant shift as custom AI Application-Specific Integrated Circuits (ASICs) are projected to overtake GPUs in shipment volume by 2028. According to Omdia, AI ASIC shipments are expected to reach 10.5 million units by 2028, surpassing the 10.1 million units projected for GPUs. While GPUs will likely maintain higher revenue through 2032 due to high pricing, the gap in unit volume is narrowing as major tech firms develop large-scale custom chip projects.

Key players driving this transition include Google with its TPU, as well as Microsoft, Meta, OpenAI, ByteDance, Alibaba, and AWS, many of whom are expected to enter mass production phases for their own chips between 2027 and 2028.

Simultaneously, the AI supply chain bottleneck is shifting from wafer production to advanced packaging and substrates. The demand for ABF (Ajinomoto Build-up Film) substrates is surging due to the increasing size and complexity of next-generation AI chips, such as Nvidia’s Rubin architecture. This demand has led to capacity shortages, with some suppliers seeing lead times extend to 12–14 months and customers negotiating long-term capacity commitments through 2029. While glass substrates are being explored as a potential future alternative, ABF remains the primary solution for the foreseeable future.