The race for AI infrastructure is beginning to reshape the relationships among major Asian manufacturers. Conversations between SK Group and Foxconn, reported by some media, point to a potential deeper partnership between the Korean memory and semiconductor materials chain and Taiwan’s manufacturing muscle for servers, racks, and complete systems for AI data centers.
This goes beyond a one-off alliance. South Korea controls a critical portion of the advanced memory that AI accelerators require, especially HBM. Meanwhile, Taiwan hosts a significant part of chip manufacturing, servers, boards, cooling systems, assembly, and the physical deployment of so-called AI factories. If SK and Foxconn strengthen their cooperation, the result could be a more integrated supply chain— from wafers and memory to finished racks.
This move comes at a time of tight supply pressures. SK Hynix, a key part of the South Korean group, has become a major HBM memory supplier for NVIDIA. Its CEO, Chey Tae-won, stated at Computex that the company plans to double its wafer capacity over the next five years to meet AI demand, warning that memory bottlenecks could persist through 2030. According to data cited by Reuters, SK Hynix held a 58% share of the global HBM market in Q1, ahead of Samsung and Micron, each with 21%.
From Memory to Full Rack Systems
The connection between SK and Foxconn makes industrial sense. SK Hynix provides one of the most scarce layers of the system: high-bandwidth memory for GPUs and accelerators. Foxconn brings manufacturing scale, system assembly, AI server integration, and the capacity to produce complete racks for clients like NVIDIA and large cloud providers.
Foxconn has been moving beyond simply “the iPhone manufacturer.” The Taiwanese company has positioned itself as a central player in AI servers and AI factories. In 2025, it claimed it could produce 1,000 AI racks per week and expected to increase that capacity by 2026. It is also developing, alongside NVIDIA, a $1.4 billion supercomputing cluster in Taiwan, with 27 MW of power and Blackwell GB300 chips, scheduled for completion in the first half of 2026.
For SK, forging closer ties with Foxconn may mean better visibility into downstream demand. In a market where customers buy not just chips but complete AI systems, understanding the evolution of racks, servers, power shelves, liquid cooling, and NVIDIA’s designs is increasingly important. Memory sales are no longer abstract; they are designed and delivered for specific platforms, with tight schedules and increasingly technical requirements.
For Foxconn, a stronger relationship with SK could help secure memory supply at a time when HBM has become one of the most limiting components of the AI supply chain. The Taiwanese manufacturer needs to coordinate with GPU, memory, board, network, power, and cooling suppliers to deliver large-scale systems. Any delay in one layer affects the entire system.
SK Siltron and the Value of Controlling Critical Materials
Potential deepening of ties with Taiwan coincides with another strategic debate within SK Group: what to do with SK Siltron. The South Korean conglomerate was reconsidering selling this silicon wafer manufacturer, after negotiating a possible deal with Doosan. Information from BigGo Finance indicates that SK has halted the process, considering wafers a strategically relevant asset within its AI value chain—especially due to their ties to semiconductors, data centers, and energy.
If confirmed, this decision would make sense within a broader strategy. For years, many tech conglomerates sold off industrial assets to reduce debt or focus on higher-margin businesses. But AI is altering that logic. Materials, manufacturing capacity, substrates, wafers, memory, energy, and cooling are now viewed as strategic elements, not just production costs.
While SK Siltron doesn’t manufacture GPUs or AI models, it operates at a fundamental phase of the semiconductor supply chain. In a cycle dominated by capacity shortages, keeping a wafer supplier close offers security, bargaining power, and industrial coordination. For SK Hynix, whose expansion into HBM depends on intensive investments and detailed capacity planning, that stability may carry more value than short-term financial divestment.
Available information suggests a shift within SK: the group is no longer solely focused on capital liberation through asset sales, but on which components it needs to retain to maintain a more integrated AI supply chain. In this context, Foxconn appears as a natural partner in the final system, while SK Siltron and SK Hynix strengthen the core of materials and memory.
Taiwan and Korea: More Allies than Rivals in the AI Era
Taiwan and Korea have competed for decades in semiconductors, electronics, and components. But AI architecture requires closer collaboration. NVIDIA designs accelerators and platforms; TSMC fabricates the most advanced chips; SK Hynix and Samsung supply memory; Foxconn, Quanta, Wistron, and others assemble servers and racks; and hyperscalers purchase capacity to train and deploy models.
No single entity controls the entire chain. That’s why events like Computex or GTC Taipei are no longer just product fairs but spaces where industrial coordination for the coming years is negotiated. Jensen Huang has recently strengthened his relationships with Taiwanese and Korean partners through meetings, dinners, and public messages directed at companies like SK Hynix, Samsung, LG, Naver, TSMC, Foxconn, and Quanta.
The thesis is clear: AI demands less fragmented supply chains. Future systems like Vera Rubin and its successors will require more memory, power, density, cooling, and faster deployment cycles. This necessitates parallel design of chips, memory, servers, racks, and data centers. The old model of supplier-by-supplier no longer suffices when the final product is a complete AI factory.
For Korea, the opportunity lies in turning its memory leadership into a broader position within AI infrastructure. For Taiwan, the challenge is to consolidate its role not only as a chip manufacturing hub but as a platform for integrating systems, servers, racks, and data centers. The discussions between SK and Foxconn fit exactly into this trajectory.
If this alliance translates into deeper agreements, it could impact the entire supply chain: HBM supply contracts, joint capacity planning, system designs compatible with future NVIDIA platforms, memory and server integration, and even AI factories across Asia and the U.S. It would also reinforce both companies’ positions before NVIDIA and large cloud providers who seek partners capable of delivering volume, quality, and on-time delivery.
It’s too early to overstate this. Public details are limited, and no formal agreement between SK Group and Foxconn has been announced. But clearly, the market trend is evident. AI rewards those who can better coordinate the physical supply chain— wafers, memory, chips, servers, racks, energy, and data centers. In this new economy, Korea-Taiwan relations may shift from sectoral competition to much deeper interdependence.
Frequently Asked Questions
What are SK Group and Foxconn exploring?
According to DigiTimes, contacts between the two companies could indicate a deeper collaboration in the AI supply chain between Korea and Taiwan, although no finalized agreement has been announced.
Why is SK Hynix so important for AI?
Because it is one of the leading providers of HBM memory, a critical component for GPUs and AI accelerators. Its capacity influences how quickly high-performance systems can be deployed.
What role does Foxconn play in AI?
Foxconn has become a key manufacturer of AI servers, racks, and complete systems, with close ties to NVIDIA and the ability to produce infrastructure at scale.
Why is SK Siltron relevant in this strategy?
SK Siltron makes silicon wafers, a fundamental material for semiconductors. Keeping this asset could help SK Group strengthen control over a critical part of the AI value chain.
via: finance.biggo