In a remarkable turn of events for the global semiconductor industry, South Korea’s SK hynix is now at the center of a shifting power dynamic, as the world’s largest technology companies scramble for access to advanced memory chips essential for artificial intelligence (AI) data centers. According to 녹색경제신문 and ZDNet Korea, the memory market has tilted decisively in favor of suppliers, with SK hynix exercising newfound leverage as demand for high-bandwidth memory (HBM) soars.
Traditionally, tech giants like Microsoft, Meta, and Nvidia held the upper hand in negotiations with chipmakers, dictating prices and volumes. But the AI boom has upended this order. With AI data center expansion driving an insatiable appetite for memory, suppliers such as SK hynix are now fielding a barrage of financial support proposals from these very customers—offering everything from hefty prepayments to investments in dedicated production lines.
Insiders say the proposals on the table are unprecedented. Global big tech firms have recently approached SK hynix with options including investment in exclusive memory production facilities, funding for advanced ASML EUV lithography equipment, prepayments covering 30–40% of contract value, and even price band mechanisms to set annual price ceilings and floors. Some of these offers specifically target SK hynix’s new fab in the Yongin semiconductor cluster, focusing on DRAM production.
Yet, rather than leaping at these offers, SK hynix is proceeding with caution. The company is carefully weighing structural risks and business flexibility before accepting any terms. As SK hynix told 녹색경제신문, “We are comprehensively reviewing various approaches and structural alternatives that differ from traditional long-term contracts.” This measured approach is a stark contrast to the past, when memory suppliers were eager to lock in long-term deals to buffer against volatile market cycles.
The reason for this change is simple: leverage. Thanks to its dominant position in the HBM market and the ongoing shortage of AI memory, SK hynix is generating robust cash flows. The company’s strengthened bargaining power means it no longer needs to rely on customer funding, and accepting such funding could even become a liability—potentially tying up supply and pricing flexibility for years to come.
An industry insider explained to 녹색경제신문, “Previously, memory makers were more aggressive in securing long-term contracts to mitigate price decline risks. Now, customers are the ones moving first to ensure supply stability. As AI server investment competition heats up, the strategic importance of memory suppliers grows in tandem.”
This new reality is reshaping contract structures across the industry. During its April 2026 earnings call, SK hynix revealed, “There are limits to accommodating all customer requests under current supply constraints. Requests for long-term contracts to secure volume are increasing rapidly.” Samsung Electronics echoed this sentiment, noting that some recent long-term deals with customers are now “binding,” underscoring a shift from simple price haggling to high-stakes battles over supply priority and guaranteed volume.
The backdrop to these negotiations is an all-out investment race among big tech companies to build out AI infrastructure. Microsoft, for example, has announced capital expenditures of $190 billion for 2026, with $25 billion earmarked to cover rising memory and component costs. Meta, too, is pursuing deals across its supply chain to secure future capacity. As a result, the focus of negotiation power is moving away from traditional CPU and GPU components and toward HBM and memory technologies.
But even as tech giants dangle large sums and creative contract structures, uncertainty remains over whether these proposals will translate into finalized deals. Price band contracts and massive prepayments are rare in the memory sector, and issues such as accounting complexities, revenue recognition, customer concentration, and potential regulatory scrutiny still need to be addressed. Some industry voices warn against becoming too closely tied to any single customer, emphasizing the need for supply flexibility and a diversified client base.
Meanwhile, SK hynix is not standing still on the technology front. In a bid to further secure its position in the AI memory supply chain, the company is collaborating with Intel on the research and development of 2.5D packaging technology—a crucial innovation for next-generation AI accelerators. According to ZDNet Korea, SK hynix is currently testing the integration of its HBM memory with system semiconductors using Intel’s Embedded Multi-die Interconnect Bridge (EMIB) technology.
2.5D packaging, which involves inserting a thin interposer between a semiconductor and its substrate to boost chip performance, is widely used in AI accelerators from companies like Nvidia and AMD. Until recently, Taiwan’s TSMC has almost exclusively supplied this technology, but soaring AI semiconductor demand has led to severe supply shortages. The SK hynix–Intel collaboration could help diversify the 2.5D packaging supply chain, reducing reliance on TSMC.
Intel’s EMIB technology offers a different approach from TSMC’s chip-on-wafer-on-substrate (CoWoS) system. Instead of a large interposer, EMIB uses small silicon bridges to connect chips only where necessary, allowing for more flexible and efficient chip layouts. SK hynix, for its part, has already set up a small domestic line dedicated to R&D in 2.5D packaging, and is now sourcing materials and components for potential mass production using EMIB.
“Although it’s still in the early R&D phase, SK hynix is actively testing 2.5D packaging implementation using Intel EMIB,” a source familiar with the matter told ZDNet Korea. “They are also scouting for candidate materials and components needed for actual mass production.”
This partnership is mutually beneficial. Intel stands to expand its advanced packaging business, while SK hynix can improve the yield and stability of its HBM products by tailoring them to the unique demands of 2.5D packaging. According to a semiconductor industry insider cited by ZDNet Korea, “Intel is actively promoting EMIB technology to SK hynix and major OSAT (outsourced semiconductor assembly and test) companies. In the mid to long term, Intel EMIB is expected to become an additional supply source for AI accelerator 2.5D packaging.”
As the AI revolution accelerates, the world’s leading memory suppliers and their partners are finding themselves in an unfamiliar but powerful position. The days when customers called all the shots appear to be over—at least for now. Instead, as AI data centers proliferate and the hunger for advanced memory grows, companies like SK hynix are rewriting the rules of engagement, balancing the promise of lucrative deals with the need to maintain flexibility and technological edge.
For the broader industry, the implications are profound. As supply chains diversify and contract structures evolve, the next chapter in the AI memory race promises to be every bit as dynamic—and unpredictable—as the technology it supports.