Sinolink: Optimistic about AI copper-clad laminates and core algorithm computing hardware, semiconductor equipment and the Apple supply chain.

Sinolink released a research report stating that it is optimistic about AI-covered copper boards/PCBs, core computing power hardware, semiconductor equipment, and the Apple industry chain. The firm predicts an explosive growth in ASICs for Google, Amazon, Meta, Open AI, and Microsoft in 2026-2027. Currently, many AI-PCB companies have strong orders, full production and sales, and are aggressively expanding production capacity, with the potential for high growth in performance to continue. There is also strong demand for AI-covered copper boards, and mainland leading manufacturers are expected to benefit actively due to slow expansion of overseas copper-covered boards. In the recent AI industry chain, many companies have exceeded expectations in performance. The demand for NVIDIA's new generation Vera Rubin platform is strong, Google's AI token processing volume grew 7 times year-on-year, Anthropic's revenue has exploded, TSMC is accelerating the expansion of advanced processes for AI chips, CSP giants are seizing storage capacity, and GPU leasing prices are rising. These factors indicate very strong short-term and mid-term demand for AI. Segment industry prosperity indicators: consumer electronics (robust upward trend), PCB (accelerating upward trend), semiconductor chips (steady upward trend), semiconductor outsourcing/equipment/materials/components (steady upward trend), display (bottoming out and stabilizing), passive components (accelerating upward trend), packaging and testing (steady upward trend). Sinolink's main points are as follows: GPU and ASIC resonances are boosting performance with potential to exceed expectations. NVIDIA's Vera Rubin has entered a stage of significant boosting, and the covered copper board and PCB supply chain have received large orders. AMD's latest generation AI GPU MI450 will be shipped in large quantities in the second half of the year, and they plan to gradually increase shipments of the Helios AI rack in the second half of the year. AMD's CEO, Lisa Su, stated that customer demand has exceeded the company's internal projections for 2027, with core customers exceeding initial plans, and new customers in talks for large-scale deployments. Google's TPU V8 and Amazon's trainium3 are also expected to enter a stage of large-scale shipping. It is predicted that the performance of core AI hardware companies in the second and third quarters will accelerate, and it is recommended to pay attention to the potential for performance to exceed expectations. SK Hynix is aggressively expanding its DRAM production capacity. SK Group Chairman Choi Tae-won stated that with the growth in demand for KV cache, memory demand will see explosive growth, and memory shortages will continue until 2030. SK Hynix's wafer capacity will double within five years, with plans to increase the monthly production capacity of DRAM wafers from the current 550,000 (including around 200,000 in its Wuxi, China factory) to approximately 1 million by 2030. This expansion will be focused on the Icheon semiconductor industry cluster. The first phase of the Icheon factory will be divided into six clean rooms, with equipment installation starting in the first clean room in February 2027, planning to increase capacity by 60,000 wafers after the equipment installation is completed. Every six months, an additional 60,000 wafers of capacity will be added in each subsequent clean room, gradually increasing capacity. By the first half of 2030, the first phase of the Icheon factory alone will add 360,000 wafers of monthly DRAM production capacity. In addition, the Cheongju M15X wafer plant is currently undergoing expansion. M15X will start production in the second half of this year with a monthly capacity of 40,000 wafers. Next year, the monthly capacity will reach approximately 80,000 wafers. With the addition of the 360,000 wafers from the Icheon wafer plant and the additional 80,000 wafers from the M15X expansion, SK Hynix's DRAM wafer monthly capacity is expected to reach around 1 million by 2030 or 2031. TSMC Chairman Mark Liu stated at the shareholders' meeting on June 4 that the demand for AI-related products will continue to be strong. Benefiting from advanced processes, special processes, and packaging demand, TSMC is very confident in its growth for the coming years. The estimated capital expenditure for 2026 is expected to be around $56 billion, indicating strong market demand. There are currently no signs of a need to reduce capital expenditures, as TSMC will continue to invest in advanced processes and packaging technology to consolidate its core position in the AI semiconductor supply chain. As AI servers move towards higher computing power while continuously reducing power consumption, the requirements for power stability and signal integrity have significantly increased. High-capacity, high-voltage, and low-loss MLCCs, as well as high-end power inductors, have seen a significant increase in demand, not only driving a multiple growth in usage per unit but also pushing up the average selling price (ASP) of products. For example, on NVIDIA's new Vera Rubin platform, the value of passive components required by a single AI server is about 4 to 6 times higher than the previous generation platform, with growth momentum shifting from shipment expansion to specification upgrades and value improvement per unit. Following the price increase of MLCCs, inductors are also expected to increase in price, especially for VRM modules and automotive DC-DC converters used in AI servers, which usually require an Isat (saturation current) of 40A or more. The delivery time for overseas major manufacturers has already exceeded 20 weeks, and it is recommended to pay attention to the benefiting industry chain. From the recent strong performance of many companies in the AI industry chain, the strong demand for NVIDIA's new Vera Rubin platform, a 7-fold year-on-year growth in Google's AI token processing volume, explosive revenue growth for Anthropic, an accelerated expansion of advanced processes for AI chips by TSMC, rising GPU leasing prices, and shortages and price increases of various materials, the demand for AI in the short and medium term is very strong. Risk warning: Risks of demand recovery falling short of expectations; risks of AIGC progress not meeting expectations; risks of further escalation of external sanctions.