THOUGHT OF THE DAY
AI Capital Spending Is Broadening Into the Factory Layer
Lam Research and KLA are gaining as investors focus on the equipment required to manufacture, inspect, and scale advanced AI semiconductors. Their strong order flows and roughly 51.7% gross margins show that process control and yield improvement are becoming strategic constraints as chip complexity rises. The next beneficiaries of AI capex may be suppliers that enable wafer throughput and defect reduction, not only companies selling finished accelerators.
Semiconductor Valuations Are Splitting Between Growth Quality and Expectations
The AI memory cycle remains powerful, but the market is differentiating sharply between Micron—already valued for near-perfect execution—and SK hynix, which reportedly trades near 7.4x earnings despite strong HBM exposure and a long-term NVIDIA agreement. This creates a more selective semiconductor trade: demand can remain structurally strong while individual stocks still face downside if valuation outruns shipment and pricing assumptions.
Geopolitical Manufacturing Footprints Are Becoming a Valuation Variable
TSMC’s planned $265 billion global manufacturing footprint is increasingly viewed as a strategic asset, not merely a capacity expansion. Diversified production can reduce customer exposure to Taiwan-related disruption while preserving access to leading-edge process technology. Foundry resilience and geographic redundancy should command a larger share of investor attention as governments and hyperscalers treat advanced chips as critical infrastructure.
COMPUTE & SEMICONDUCTORS
- Broadcom’s AI semiconductor revenue is reported to have grown 221% year over year, supported by custom ASIC programs for Google, Meta, and Anthropic. Its ability to deliver application-specific chips at scale reinforces the competitive threat to general-purpose accelerators, particularly where hyperscalers prioritize inference efficiency and workload-specific economics.
- Micron remains a major HBM and DRAM beneficiary, with long-term customer commitments and supply constraints reportedly extending through 2030. However, the stock’s elevated expectations create asymmetric downside if HBM pricing, qualification schedules, or AI infrastructure spending moderate.
- SK hynix combines strong HBM exposure, a long-term NVIDIA relationship, and a reported valuation near 7.4x earnings. The valuation gap versus Micron suggests the market is pricing memory exposure unevenly rather than treating all HBM suppliers as equivalent.
- Lam Research and KLA are capturing investor interest as AI demand increases spending on etch, deposition, inspection, and process control. Their order momentum indicates that wafer-fab bottlenecks and yield economics are becoming investable parts of the AI supply chain.
- TSMC remains the central advanced-node foundry, while its global manufacturing investment offers customers additional geopolitical and supply-chain resilience. The foundry moat is expanding from process leadership to trusted, geographically diversified capacity.
POSITIONING IDEAS
Bullish
- KLA (KLAC) and Lam Research (LRCX): Favorable exposure to the factory layer of AI spending, with rising process complexity supporting inspection, metrology, and wafer-fabrication equipment demand.
- Broadcom (AVGO): Custom ASIC growth across hyperscalers supports a credible second engine of AI silicon demand. Its reported 221% AI revenue growth and lower valuation than NVIDIA strengthen the relative-value case, although execution expectations remain high.
- SK hynix (000660): HBM demand, long-term NVIDIA supply relationships, and a low reported earnings multiple provide a more attractive risk-reward profile than fully re-rated memory peers.
Bearish
- Micron (MU): The company remains fundamentally exposed to the AI memory upcycle, but its valuation leaves less room for execution misses. Any weakening in HBM pricing, customer inventory, or AI capex could trigger a sharp multiple reset.
- Broad semiconductor equipment momentum: Strong orders and margins may already discount a prolonged AI fab-investment cycle. A delay in advanced-node capacity additions or tighter customer budgets would expose the sector’s cyclical sensitivity.