Philadelphia Semiconductor Index Drops Over 3% as SanDisk Tumbles 5%; Citi Says AI Capex Still Expanding but Profit Outlook Diverges

TradingKey - On October 8 Eastern Time, AI hardware stocks generally faltered, with the Philadelphia Semiconductor Index dropping over 3%. Although AI infrastructure investment continues to expand, the earnings prospects of hardware companies have begun to diverge, and the narrative of "continued growth in AI demand" may not be enough to drive stock prices further up.

In terms of major market indexes, as of press time, the Dow Jones Industrial Average fell 0.19% to 51,080.39; the Nasdaq Composite Index dropped 1.24% to 27,196.81; and the S&P 500 Index slipped 0.65% to 7,751.09.

At the sector level, optical communication stocks, memory stocks, and chip stocks were all sold off. SanDisk (SNDK) fell 5.5%, breaking below the $1,600 mark; Corning (GLW) dropped 5.42%; Marvell Technology (MRVL) fell 5.03%; SK Hynix (SKHY) lost 4.94%; Intel (INTC) dropped 4.92%; Micron (MU) fell 3.88%; AMD (AMD) slid 3.44%; Qualcomm (QCOM) declined 2.55%.

Citi's latest industry research indicates that global AI-related capital expenditure will grow by about 106% in 2026 and continue to expand by about 56% in 2027. Although the growth rate is slowing down, the overall spending scale is still expanding, meaning a growth deceleration should not be equated with shrinking demand.

However, against the backdrop of substantial gains previously accumulated by the sector, relying solely on "continued growth in AI demand" may not be enough to support further upward movement in stock prices. Third-quarter earnings reports will test companies' revenue growth, profit margins, and cash conversion, serving as a key window to distinguish the quality of growth among hardware companies.

Interconnect components are a key area of optimism for Citi. The firm upgraded its preference ranking for Amphenol (APH) and Corning (GLW) to first and second place, respectively, among component targets. Its core rationale is that as the data rate, power density, and architectural complexity of AI systems increase, the value of interconnect components required per system also rises. Consequently, revenue growth for relevant suppliers stems not only from server volume expansion, but also from higher procurement value per system and market share gains.

Latest data cited by Citi shows that Amphenol's IT Datacom segment achieved 63% organic growth, with AI-related sales growing faster than traditional business, while Corning benefited from higher optical fiber density and increasing demand for inter-rack connectivity. However, the evolution of interconnect technology is not simply a case of "optical replacing copper." Citi believes future systems will adopt hybrid architectures, where copper connections continue to serve short-distance and high-power scenarios, while optical connections expand alongside higher transmission rates and distances. For suppliers, whether new orders convert into revenue and whether product upgrades drive margin expansion deserve more attention than merely scaling up sales volume.

In the storage sector, Citi prefers hard disk drives (HDDs) over NAND flash memory. The firm believes that the HDD industry features lower competitive intensity and stronger capex discipline, leaving room for margin expansion. Demand outlooks from Seagate (STX) and Western Digital (WDC) both point to at least around 25% annual exabyte capacity growth, while long-term procurement agreements have enhanced demand visibility. Citi noted that Western Digital's per-terabyte pricing recently increased by roughly 18% to 19% year-over-year, indicating that capacity demand and pricing together support revenue; however, customer demand still needs to be validated through actual deliveries.

By contrast, NAND faces a divergence between enterprise and consumer demand. Citi expects NAND demand to grow by about 29% in 2027, driven by incremental demand from AI inference, context storage, and offloading cache to SSDs, while consumer and PC-related demand remains weak. This implies that strong data center investment does not automatically translate to simultaneous prosperity across all hardware end-markets. For AI hardware companies, the key focus in the next stage will be whether they can preserve profits alongside revenue expansion and convert those profits into cash.