A quiet but profound restructuring is underway in global AI hardware trade flows. While market attention remains fixated on giants like NVIDIA or TSMC, the true pace-setters of AI infrastructure expansion often reside in overlooked intermediary layers—process metrology, storage architecture, and foundational electronic components. Three companies recently highlighted by Simply Wall St.—Nova (an Israeli semiconductor metrology specialist), Seagate Technology Holdings (a hard drive manufacturer), and Vishay Intertechnology (a discrete and passive component supplier)—collectively represent this “non-core yet critical” support triangle. They don’t design AI chips, but they are indispensable cogs enabling AI’s scaling engine.
Nova’s strategic edge lies in its near-monopoly over thin-film metrology for advanced nodes. At TSMC’s 3nm and below processes, film thickness tolerances must be controlled at the atomic scale. Nova’s optical and X-ray-based systems provide real-time feedback that boosts fab yield by 0.5–1 percentage points—a seemingly marginal gain that translates to hundreds of millions in annual revenue for a 60,000-wafer-per-month 3nm facility. By 2025, AI-related customers accounted for nearly 45% of Nova’s revenue, up from under 20% in 2022, driven primarily by TSMC’s expansions in Taiwan, China and Arizona. I judge that as CoWoS advanced packaging capacity doubles between 2026 and 2027, Nova’s tool penetration will intensify further, since its metrology solutions are essential for ensuring reliability in high-density interconnects.
Seagate’s opportunity is more counterintuitive. In an era dominated by HBM memory for AI training, why would a traditional HDD maker benefit? The answer lies in the distributed deployment of AI inference. Meta, Microsoft, and Amazon are increasingly running lightweight AI models in cost-sensitive edge data centers that cannot afford all-flash architectures. Seagate’s new Exos Mozaic 3+ platform, leveraging Heat-Assisted Magnetic Recording (HAMR), delivers 30TB per drive at over 80% lower cost per gigabyte than SSDs. In fiscal Q4 2025, Seagate’s enterprise HDD shipments grew 22% year-over-year, with nearly one-third going to AI inference clusters. Crucially, its collaboration with NVIDIA has evolved beyond hardware supply into co-optimization—for instance, tailoring data prefetching algorithms for the Llama 3 inference framework to reduce I/O latency. This “hardware-software synergy” is redefining the value proposition of storage vendors.
Vishay’s role is the easiest to underestimate. As one of the world’s largest suppliers of discrete semiconductors and passive components, it provides the “capillaries” of AI server power management: MOSFETs, tantalum capacitors, and current sensors. A single NVIDIA DGX SuperPOD system integrates over 2,000 Vishay components for voltage regulation, overcurrent protection, and power efficiency. In 2025, its industrial and computing segment revenue grew 18%, far outpacing the 9% growth in automotive. Notably, Vishay is shifting its MLCC (multilayer ceramic capacitor) production toward high-reliability, high-frequency variants to meet the transient load demands of AI accelerator cards. Though low in unit price, these components carry high replacement costs—failure can render an entire GPU card unusable—leading top-tier customers to lock in certified suppliers through long-term agreements, creating an invisible moat.
What unites these three firms is their position in AI’s “unsexy” supply chain segments, where technical barriers, customer stickiness, or cost structures confer pricing power. More importantly, their capital expenditure cycles are desynchronized from AI chip designers: when NVIDIA slows procurement due to inventory adjustments, fabs continue building, and data centers keep expanding—making demand for equipment and foundational components more persistent, if lagging.
Geopolitics amplifies this structural advantage. The U.S. CHIPS Act’s push for domestic manufacturing has spurred localized equipment needs at TSMC’s Arizona and Intel’s Ohio fabs. Meanwhile, China’s accelerated push for AI self-reliance indirectly fuels demand for Nova-like technologies through third-party integrators. Although Seagate faces export controls, its HAMR technology remains unrestricted, allowing it to capture incremental orders from new data center builds in Southeast Asia and Mexico.
The market’s understanding of AI hardware remains overly centered on compute density and bandwidth races, neglecting the “ordinary foundations” of infrastructure. The rise of Nova, Seagate, and Vishay isn’t accidental—it reflects the inevitable release of ancillary demand as AI moves from lab experiments to mass deployment. Over the next 18 months, as AI inference loads shift to the edge, advanced packaging ramps up, and global data center energy efficiency standards tighten, visibility into these firms’ revenue streams will only improve.
The question worth pondering is this: as AI hardware trade flows evolve from “chip-centric” to “system-coordinated,” are investors ready to assign premium valuations to companies that don’t fabricate transistors—but ensure trillions of them operate reliably?