The rapid development of the global generative AI market is driving unprecedented transformation in the semiconductor industry. DIGITIMES Research's latest AI Chips Special Report highlights that generative AI models' training and inference rely heavily on substantial computing resources, making AI chips a focal point.
This trend spans high-performance AI accelerators in cloud data centers and edge AI device chips driven by privacy concerns, signaling an explosive growth period for both hardware and software opportunities in AI applications. The report emphasizes that advanced manufacturing technologies and capacities of wafer foundries will be key to realizing the accelerated computing needs of AI chips.
Cloud-based AI leads generative AI development, with wafer foundries' advanced processes, advanced packaging, HBM capacity expansion, and strong demand for AI servers in cloud data centers driving growth. Analyst Stella Weng forecasts that from 2023 to 2028, the compound annual growth rate (CAGR) of high-end GPUs for the cloud will reach 44%, while cloud ASIC accelerators will see a CAGR of 52%.
Weng further notes that 2022 to 2024 will be a high-growth period for cloud AI accelerators. However, as cloud service providers refine their AI application infrastructure, she expects the growth in AI server demand to slow down in the future, gradually shifting towards replacement-focused demand.
In response to strong production demand for cloud and edge AI chips, wafer foundries are accelerating the expansion of advanced processes and packaging capacities. Analyst Eric Chen estimates that from 2023 to 2028, the CAGR of advanced process expansion below 5 nanometers in the wafer foundry industry will reach 23%.
In advanced packaging, AI chips heavily rely on TSMC's CoWoS packaging technology, with TSMC's CoWoS capacity expansion CAGR projected to exceed 50% from 2023 to 2028. Chen also notes that to meet the shipping and performance enhancement needs of AI chips, wafer foundries are actively advancing the development of advanced processes and packaging technologies, including photolithography techniques, new transistor structures, back-end power delivery (BSPDN), 2.5D/3D advanced packaging, glass substrates, chip-on-wafer-on-substrate (CoWoS), and silicon photonics integration.
Generative AI applications are rapidly expanding from cloud-based services to edge deployments, with smartphones and PCs expected to be the first to adopt scaled production. The development and market changes of AI application processors (APs) in smartphones are particularly noteworthy.
Analyst Jim Chien forecasts that global shipments of generative AI smartphone APs will reach 260 million units in 2024 and could rise to 800 million units by 2028, with a shipment CAGR of 65% during this period. This growth will benefit AI smartphone shipments, with an estimated 240 million units expected in 2024 and a potential to exceed a 50% penetration rate for AI smartphones by 2028.
In the AI PC sector, analyst Joyce Chen observes that in 2024, PC hardware and software vendors are actively testing the AI PC market. She anticipates that by 2025, with PC processors enhancing computational power and reducing power consumption through more advanced processes, and Microsoft's planned discontinuation of Windows 10 and introduction of Copilot products to the endpoint market, a new wave of AI PC growth will emerge.
Chen predicts that AI PC processor shipments will approach 50 million units in 2024, with subsequent shipments increasing annually as PC hardware and software integration stabilizes, reaching up to 150 million units by 2028, with a shipment CAGR of 39% during this period.
Looking ahead, the deployment of AI applications promises substantial business opportunities, not only expanding existing markets but also transforming current technology and capacity demands to meet new needs. This shift is expected to further intensify competition in the chip market and lead to unprecedented changes in the semiconductor ecosystem.
For more insights into the impact of generative AI on cloud and edge computing, read DIGITIMES Research's latest AI Chip Special Report. It provides guidance on the development of technologies across the supply chain, including upstream, midstream, and downstream, as well as the emergence of various AI hardware devices, offering detailed insights into market share and development blueprints for AI chips.
