On June 23, Citi stated that AI servers and high-speed network upgrades are reshaping the growth outlook for China's PCB industry, raising target prices for Huada Semiconductor and Shenghong Technology due to the potential for AI-PCB demand to continue expanding rapidly over the next three years. In a report released on June 21, Citi raised its forecast for the total potential market size of AI-PCB from 2026 to 2028 to RMB 152 billion, 307 billion, and 562 billion, corresponding to year-on-year growth of 86%, 102%, and 83%. The bank noted that the new model incorporates demand for CPU and optical module PCBs and provided a forecast for 2028 for the first time. Citi expects that in 2027, AI-PCB demand will be dominated by ASICs, accounting for about 34%; NVIDIA GPU-related demand will account for 24%, CPUs for 16%, switches for 14%, and optical modules for 12%. In terms of growth rates, network-related demand is the strongest, especially with the ramp-up of 1.6T optical modules and switches, which may drive optical module PCBs to grow by 135% and 178% in 2026 and 2027, respectively. The report indicated that Google TPU-related PCB procurement could become a key increment in 2028. Citi estimates that Google TPU PCB procurement could reach $16 billion by 2028, surpassing NVIDIA GPU-related PCB demand. Breaking down the new demand from 2027 to 2028, Google TPU contributes about 30%, NVIDIA contributes 25%, and switches and optical modules each contribute 14%. As AI-PCB demand expands rapidly, capacity may become a new constraint. Citi expects the next round of PCB expansion announcements to begin in the second half of 2026 to support 2028 demand. The bank stated that PCB manufacturers in mainland China typically require 13 to 15 months from greenfield projects to mass production, while upstream fiberglass and copper-clad laminate expansion cycles may take longer, potentially reaching 18 months. This means that AI-CCL supply may tighten further. Citi believes that as major AI-PCB projects launch in the second half of 2026, first-tier PCB manufacturers with the ability to secure raw materials will see improved profit visibility in 2027. Related expansion announcements may also act as catalysts for the industry, strengthening investor confidence in the long-term growth of AI-PCB. In terms of individual stocks, Citi maintains a 'Buy' rating for Huada Semiconductor and Shenghong Technology. The bank raised Huada Semiconductor's target price from RMB 119 to RMB 189, citing the company's strong mass production and delivery capabilities in the data center switch and AI server PCB sectors. Citi expects Huada Semiconductor's net profits from 2026 to 2028 to be RMB 6.7 billion, 12.1 billion, and 23.2 billion, respectively, and believes its profit CAGR could reach 86%. Shenghong Technology's target price was raised from RMB 415 to RMB 456. Citi lowered its 2026 net profit forecast, mainly reflecting lower-than-expected contributions from Rubin-related revenues, but believes that 2027 profits will remain stable and introduced a net profit forecast of RMB 23.2 billion for 2028 for the first time. The bank stated that Shenghong Technology's long-term relationship with NVIDIA, HDI capacity, and potential opportunities in data center switches and ASICs still support its valuation premium. However, Citi also cautioned that the AI-PCB market still faces multiple risks, including lower-than-expected shares of GenAI-related orders, yield issues, price competition in the automotive supply chain, reduced capital expenditures by cloud vendors, weakening macro demand, rising material costs, and geopolitical risks between China and the U.S. Overall, Citi's assessment is that AI-PCB demand is no longer solely driven by NVIDIA GPUs but is expanding to include ASICs, Google TPUs, high-speed switches, and optical modules. As demand for 2028 enters the capacity planning phase earlier, leading Chinese PCB manufacturers with high-end processes, customer resources, and material security capabilities may continue to be core beneficiaries of AI infrastructure expansion.