Good for ↑
Memory chip pricing is recovering as AI training clusters demand High Bandwidth Memory at unprecedented scale. The market is buying Micron directly, but the constrained supply chain for HBM production — deposition, etch, advanced packaging equipment — is the less-crowded trade. Long US large-cap semiconductor capital equipment names with HBM exposure as the upcycle accelerates.
Source: MarketWatch ↗
The article highlights a growth ETF strategy that blends value with momentum to play the memory chip rally, using Micron as the poster child. But the deeper trade is one step removed: when DRAM and HBM pricing inflects upward in a supply-constrained cycle, the companies that MAKE the machines to produce these chips — capital equipment suppliers — see a simultaneous surge in orders AND margin expansion, with less headline risk than the chip makers themselves. Here is the transmission chain: AI training clusters (think NVIDIA H100/H200/B200 GPUs) require HBM3e stacked memory. HBM is produced via a radically different process than standard DRAM — it requires through-silicon vias (TSVs), advanced deposition layers, and thermal compression bonding. This means Micron, SK Hynix, and Samsung MUST invest in specialized capital equipment to expand HBM capacity. Micron has publicly guided for HBM3e to be a significant revenue contributor in FY2025. That capex flows directly to a small number of US-listed equipment suppliers: companies specializing in etch (LRCX), deposition (AMAT), and metrology/inspection (KLAC). These names are already showing momentum — XLK is up +3.5% on the week, leading all sectors — but the HBM-specific capex cycle is still in early innings. The market context is supportive: VIX at 16.7 and falling (-4% on the week) signals risk-on appetite is returning. The 10Y yield is stable at 4.48% after a -4% weekly move, reducing the discount-rate headwind for growth names. S&P 500 is up +4.7% on the month. This is a regime where semi equipment names with strong earnings visibility can re-rate. The key structural tailwind: HBM capacity cannot be built overnight — lead times for advanced deposition and etch tools run 12-18 months, meaning equipment order books are being filled NOW for 2025-2026 production ramps.
The strongest bear case is that HBM capacity is being added faster than consensus expects — SK Hynix is already the dominant HBM3e supplier to NVIDIA, and Samsung is aggressively ramping. If the HBM market becomes a three-way race with oversupply emerging by late 2025, DRAM pricing could roll over before the equipment upcycle fully materializes, causing customers to delay or cancel tool orders. The specific assumption most likely to be wrong: that Micron's HBM ramp is large enough to drive incremental equipment orders at LRCX/AMAT/KLAC beyond what is already in their backlogs. If Micron is sourcing HBM tools from non-US suppliers or if the ramp is slower than guided, the equipment thesis weakens considerably. The data point that would flip conviction: a Micron earnings call where HBM capex guidance is cut or deferred, or LRCX/AMAT guiding to flattening memory equipment orders in their next print.
The 2016-2018 DRAM upcycle is the closest historical analog. When DRAM pricing inflected in mid-2016 after years of oversupply, LRCX and AMAT outperformed Micron itself on a risk-adjusted basis over the subsequent 18 months — LRCX rose ~180% from mid-2016 to early 2018 vs. MU's ~300%, but with roughly half the drawdown volatility during corrections. The equipment names benefited from the capex commitment being locked in regardless of spot price fluctuations. A second analog: the 3D NAND transition in 2017-2019, where NAND producers had to massively re-equip fabs, drove AMAT and KLAC to multi-year highs even as NAND spot prices were volatile. The pattern: when a structural process change (3D NAND then, HBM now) forces fab re-tooling, equipment suppliers capture durable multi-year revenue streams that are less correlated to spot memory prices than the chip makers themselves.
Memory chip pricing recovers as AI cluster buildout drives insatiable HBM demand, directly benefiting Micron and other DRAM producers. The market buys MU and memory-adjacent ETFs as the most obvious trade.
Micron and competitors must rapidly expand HBM3e/HBM4 production capacity to meet hyperscaler demand from NVIDIA, AMD, and custom ASIC programs at Google and Amazon. This requires massive, specialized capex commitments to a narrow set of equipment suppliers — particularly those with etch, deposition, and advanced packaging tool expertise. Equipment order books lengthen, pricing power improves, and backlog visibility extends 12-18 months forward.
US large-cap semiconductor capital equipment companies — specifically those with disproportionate HBM/advanced DRAM process exposure — see a compounding earnings revision cycle: higher tool volumes, better ASPs on HBM-specific tools, AND improving gross margins as utilization rises. The market is currently pricing these names as 'general semi equipment' plays, missing the HBM-specific revenue mix shift. As Micron's HBM revenue guidance becomes more concrete through 2025 earnings calls, the equipment suppliers get re-rated on a higher-quality, longer-duration revenue stream. This is the third-order trade: the machines that make the chips that power AI are less crowded, less volatile, and have more durable earnings power than the chip makers themselves in an upcycle.
No comments yet.
Sign in to comment
Earlier connections in the same sector that share a theme with this one.