When it comes to AI, Asian hardware’s easily as important as US software players

While focus remains on US software giants, a closer look reveals that Asia’s factories and materials firms are shaping the future of AI technology, highlighting new opportunities and strategic vulnerabilities.

While investor attention has clustered on the US “Magnificent Seven”, a closer look at the physical backbone of large‑scale artificial intelligence points to a different geography: factories, materials and systems engineering across Taiwan, South Korea, Japan and Southeast Asia. According to the original report from asset manager Ninety One, the next phase of the AI boom will be determined not by headline software valuations but by the logic, memory, networking and power technologies that make massive models possible. ^[1]

Ninety One identifies a cohort of largely emerging‑market firms , the “Secret Seven” , whose products sit at the physics limits of AI performance yet whose market prices still carry the discount commonly applied to non‑US equities. The argument is straightforward: world‑class accelerators require advanced logic dies, high‑bandwidth memory, extreme‑speed switches, specialised power and thermal systems, and sophisticated assembly and packaging. These components are concentrated in Asian manufacturing clusters that combine dense supplier networks, deep engineering talent and heavy R&D. ^[1]

At the centre of the logic side is Taiwan Semiconductor Manufacturing Co (TSMC), which Ninety One describes as setting the ceiling for compute. TSMC’s foundry model, its institutional confidentiality and the way fabs plug into local ecosystems give it a durable competitive edge. Recent developments show that role evolving: TSMC is evaluating an upgrade of its Kumamoto, Japan plant to support more advanced 4nm/5nm nodes, signalling demand from regional customers for cutting‑edge capacity. Industry data and government initiatives also point to diversification of capacity, with new facilities being developed in multiple jurisdictions. ^[1][5][6]

Memory, which now often constrains AI throughput, is dominated by a small set of suppliers. SK Hynix and Samsung supply the HBM modules that sit beside GPU dies and feed data at the speeds modern models demand. According to the Ninety One analysis, supply is already tight: OpenAI’s expanded data‑centre ambitions and strategic supply agreements have allocated significant future HBM output, and market observers note DRAM prices have risen sharply in 2025 as AI infrastructure demand lifts component pricing. That pricing behaviour has begun to filter into consumer electronics, with device makers warning higher DRAM and NAND costs are affecting product pricing. ^[1]

The networking and systems layers that knit compute and memory together are similarly specialised. Accton, for example, builds the high‑speed switches that enable hyperscale clusters to move data at 400‑, 800‑ and now 1.6‑terabit rates; Delta Electronics provides the power distribution and thermal engineering required when single servers draw double‑digit kilowatt loads. Ninety One highlights instances where rapid engineering responses , such as Delta’s re‑engineering when Nvidia increased server power requirements , illustrate how system suppliers can be critical enablers of new accelerator designs. ^[1]

Bringing chips and memory into functional modules is the work of advanced packaging and assembly firms such as ASE, which co‑designs packages with major customers and therefore gains early visibility into architectural shifts. Upstream, materials suppliers like Anji Microelectronics are increasing their share as regional manufacturing expands, particularly within China, adding depth to the supply chain at stages where capability has strategic importance. Together these players form a chain whose weakest link can throttle overall AI performance. ^[1]

Geopolitics and policy are reshaping that chain. Taiwan’s recent opening of a 15‑megawatt cloud centre in Tainan , housing the “Nano 4” supercomputer built with thousands of Nvidia H200 and Blackwell chips , underlines the island’s push from pure manufacturing toward sovereign AI capabilities, with TSMC and global partners playing visible roles. At the same time, national strategies elsewhere aim to shore up domestic capacity: South Korea is planning a new foundry backed by public and private funds to strengthen local production of legacy and defence‑critical chips. These moves reflect both the strategic value of semiconductor manufacturing and the desire of governments to reduce single‑point dependencies. ^[2][4]

The market implications flow from a persistent mismatch: firms that control physical constraints on AI , fabrication, memory supply, interconnects, power and packaging , often trade on emerging‑market multiples despite their centrality to a technology whose economics are being rewritten. Ninety One argues that modest repricing, not collapse, is the likeliest near‑term outcome: risk will be re‑assessed, unit economics will settle, and the firms that secure capacity and materials will capture outsized strategic value. At the same time, recent moves by Nvidia and TSMC to expand US manufacturing, and export decisions affecting chip flows to China, underscore that supply chains and demand are being contested and rebalanced across markets. ^[1][6][7][3]

For investors and policymakers alike, the lesson is that AI’s next chapter will be as much about electrons and chemistry as it is about algorithms. Industry data and recent government programmes show the places likely to matter are those that can sustain advanced fabrication, memory production, high‑speed systems and the specialised materials and assembly that tie them together. The “Secret Seven” framing is a reminder that strategic relevance and market pricing do not always move together, and where they diverge opportunities , and vulnerabilities , can be found. ^{[1][2][3][4][5]}

📌 Reference Map:

##Reference Map:

• ^[1] (ioandc) – Paragraph 1, Paragraph 2, Paragraph 3, Paragraph 4, Paragraph 5, Paragraph 6, Paragraph 7, Paragraph 8
• ^[5] (Tom’s Hardware) – Paragraph 3
• ^[6] (Reuters) – Paragraph 3, Paragraph 7
• ^[2] (Reuters) – Paragraph 7, Paragraph 8
• ^[4] (Reuters) – Paragraph 7
• ^[3] (Reuters) – Paragraph 7

Source: Fuse Wire Services