BlockBeats News, July 1st. The Asian AI hardware market is entering a more decentralized and discerning phase. Over the past year, the market's mainstream focus has been relatively clear: GPU, HBM, TSMC, advanced packaging, and complete servers have been the most easily understood AI infrastructure investment directions for investors.

However, recent reports from Nomura, UBS, and JPMorgan Chase show that this focus is now expanding deeper into the supply chain. The demand brought about by AI data center construction is no longer just driving orders for core chips but is also starting to impact various components in the supply chain such as PCBs, CCLs, substrates, power supplies, cooling systems, passive components, HDD-related parts, silicon carbide, and optical interconnects. In other words, the market's focus is shifting from "who can get hold of GPUs" to "who can address system-level bottlenecks."

Nomura

In its Asian AI Semiconductor and Server report released on June 30th, Nomura defined the recent semiconductor stock pullback as a "healthy digestion" phase rather than the end of the AI cycle. The firm pointed out that the Philadelphia Semiconductor Index has risen by about 85% since March, with a surge of over 200% since revisiting the AI theme in May 2025. After such rapid growth, a short-term pullback is not surprising, especially as the market begins to factor in pressures on free cash flow for cloud providers, rising interest rates, and execution risks in high-end technology roadmaps. However, Nomura believes that a price correction does not equate to a fundamental reversal. On the contrary, the capital expenditure of hyperscale cloud providers may continue to rise until 2027, as the competition in AI infrastructure is turning into a "either invest on a massive scale or lose ground" race. Even if some cloud providers face cash flow pressures, they will find it difficult to voluntarily slow down their construction pace.

Key to Nomura's analysis is the supply side. The report suggests that the new capacity additions that started at the end of 2025 will take a considerable amount of time to come online, indicating that the tightness in some parts of the supply chain may not ease but rather intensify in the second half of 2026 and into 2027. Bottlenecks are no longer limited to GPUs or HBMs. Advanced packaging capabilities such as CoWoS, WoS, and CoW are still in short supply, while AI PCBs, CCLs, ABF substrates, server power supplies, cooling systems, testing, and storage could also become new pain points.

Therefore, Nomura has raised the target prices of several AI semiconductor and hardware companies, including TSMC, ASE Technology, MediaTek, Unimicron, Flex, Samsung Electronics, Foxconn, Lenovo, and others. The implicit logic is that the AI hardware market has transitioned from a shortage of core chips to a shortage in the systemic supply chain. The next round of profit upgrades may come from those components and processes that were previously not in the spotlight but are indispensable in server architecture.

UBS

UBS's research on China's semiconductor and hardware industry adds another piece to the puzzle from the perspective of the Chinese supply chain. In a report dated June 26, the bank stated that after discussions with five H-share Chinese technology companies, a common signal observed was that the demand for Chinese AI infrastructure remains strong. Local IC design companies are trying to increase AI computing efficiency through scale-up interconnect, silicon photonics, advanced packaging, and system-level optimization, particularly in the context of constraints in advanced semiconductor processes. This reflects the reality of China's AI hardware chain: when access to state-of-the-art processes and high-end overseas GPUs is limited, the performance of individual chips is difficult to catch up entirely. Therefore, manufacturers can only enhance system efficiency through interconnect, packaging, photonics, and cluster architecture.

UBS mentioned that some companies are using photonic interconnects to address data transmission bottlenecks in large-scale GPU or ASIC clusters, while others are strengthening local supply chain collaboration from wafer fabrication to packaging and testing.

At the company-specific level, UBS focuses on hardware companies such as BlueStar, Silead, SiFlower, and Tanyu Advanced. BlueStar is no longer just a consumer electronics appearance parts company; it is also involved in photonics, liquid cooling, glass substrates, and HDD-related materials. The report states that BlueStar is expected to have a significant year of value-added growth in consumer electronics by 2027, with server liquid cooling, photonics, and glass substrates possibly becoming new growth areas. Silead benefits from AI PCB demand, and the company has seen relatively smooth cost transmission of next-generation products without significant delays in Rubin Ultra-related backplane solutions. Regarding the market's focus on Q-glass and PTFE routes, the company believes customers are still evaluating them, and technological changes themselves could bring new material and process opportunities. Tanyu Advanced, in the SiC industry chain, is benefiting from the increasing penetration of electric vehicles and the expansion of data center applications, with selective price increases already appearing due to urgent orders.

The significance of UBS's report lies in expanding the AI supply chain narrative from the core chipsets between Taiwan and the U.S. to China's hardware chain capability. Chinese companies may not directly control the highest-end GPUs, but they can find their place in PCBs, silicon photonics, liquid cooling, SiC, glass substrates, and domestic computing interconnect architectures.

JPMorgan

JPMorgan's research on the electronic components industry further delves into more mature and fundamental components. In a memo dated June 30, the bank's discussions with investors in Hong Kong and Singapore in late June focused on MLCCs, aluminum electrolytic capacitors, HDD-related components, and Japanese electronic component companies such as Murata, TDK, Sunlord, Nichicon, Nippon Chemi-Con, Ibiden, Rohm, MinebeaMitsumi, and Alps Alpine. The key question the market is truly concerned about is whether AI server demand is sufficient to change the supply-demand dynamics of these traditional component industries.

In the past, MLCCs, capacitors, inductors, and certain power components were more tied to the cycles of smartphones, automobiles, and industrial sectors, and their growth potential was not prominent. However, AI servers have higher requirements for power stability, power density, heat dissipation, storage, and high-speed interconnects. The value of passive components used per server may increase. If the demand continues to rise, even mature industries may regain the ability to raise prices.

J.P. Morgan noted that investors have mixed views on MLCCs. Some believe that companies such as Murata, Taiyo Yuden, Nichicon, and Nippon Chemi-Con have already priced in the demand from AI servers and HDDs. On the other hand, some believe that inventory levels are low, and if downstream demand continues to exceed expectations, the price hike cycle may just be beginning. TDK is seen as a potential beneficiary in the AI server power chain, and components like thin-film inductors may benefit as server power densities increase. Nichicon and Nippon Chemi-Con may benefit from high-voltage direct current data center architectures, as these architectures require higher-spec aluminum electrolytic capacitors. The HDD industry chain is also back in focus due to the rising AI data storage demand, which may support some HDD component and material companies. Compared to Nomura and UBS, J.P. Morgan is more cautious in its tone: it acknowledges that the AI narrative is spreading to electronic components but also warns that some stocks have already priced in price hike expectations. In the future, if orders, price negotiations, or earnings realization fall short of expectations, stock prices could decline.

Reviewing the three reports together, the Asian AI hardware cycle has not stopped at the GPU level but is continuing to expand along the data center construction chain. Nomura emphasizes server and advanced packaging bottlenecks, UBS sees the Chinese hardware chain taking on demand in directions such as silicon photonics, PCBs, liquid cooling, and SiC, while J.P. Morgan captures the price hike expectations that may occur in mature component industries. They collectively point to a shift: AI capital expenditure remains the main focus, but the market is no longer only rewarding the most prominent core chip assets. In the next stage, investors will need to assess which links are truly supply-constrained, which companies have the ability to convert scarcity into price, orders, and profit margins.

This also signifies that AI trading has entered a more challenging phase. In the first stage, buying into core chips and advanced packaging leaders was almost a bet on AI infrastructure expansion; in the second stage, opportunities are more diversified, and validation is more complex. PCBs, CCLs, power supplies, heat dissipation, MLCCs, aluminum electrolytic capacitors, HDDs, SiC, and optical interconnects all stand to benefit, but their demand elasticity, price hike ability, and profit transmission are not the same. For investors, the existence of AI demand is no longer the biggest debate. The real disagreement lies in which supply chain constraints are true bottlenecks and which are just narratives of stock price preemptive trading.