Samsung Electronics became the first chipmaker to mass-produce and ship sixth-generation high-bandwidth memory chips in February — the component that Nvidia, AMD, and Google are competing to secure for their next-generation AI accelerators — while negotiations with more than 50,000 workers collapsed this week, leaving a threatened 18-day strike starting May 21 as the single biggest near-term risk to the global AI chip supply chain. Quarterly financial disclosures released this month show Samsung poured 11.34 trillion won ($7.61 billion) into research and development in Q1 2026, up 25.5% year over year, while rival SK Hynix lifted its own R&D spending 68.3% to 2.55 trillion won ($1.71 billion). Samsung has made a public announcement of commenced HBM4 mass-production shipments; SK Hynix has confirmed HBM4 is progressing per customer agreements but has made no comparable public announcement, and its next-generation HBM4E mass production is targeted for 2027. Anyone planning GPU deployment timelines or enterprise memory procurement on the assumption that both South Korean giants are shipping HBM4 at identical scale is working from a false premise.
Talks Collapse, Strike Clocked for May 21
Government-mediated negotiations between Samsung Electronics and its unions broke down this week after 17 hours of final talks ended without agreement in the early hours of May 13. The National Samsung Electronics Union has scheduled an 18-day general walkout from May 21 to June 7, demanding that Samsung scrap its performance-bonus cap and redirect 15% of operating profit to worker incentives — a figure that could reach 45 trillion won ($30 billion) based on analysts' full-year estimates. More than 50,000 employees are expected to participate, and South Korea's Prime Minister convened an emergency ministerial meeting after the breakdown. The country's Labor Minister met directly with union chair Choi Seung-ho on May 14; the union rejected Samsung's subsequent written offer the same day, saying it would not resume talks before June 7.
The scale of the potential production hit is not theoretical. A single half-day rally at Samsung's Pyeongtaek campus on April 23 reduced daily memory output by 18.4%, according to the Seoul Economic Daily, while contract foundry output on the affected shift fell 58%. About 1,700 Samsung suppliers have drawn up contingency plans. JPMorgan Chase estimates the 18-day walkout could reduce Samsung's annual operating profit by more than 40 trillion won. Samsung holds roughly 40% of global DRAM market share and is currently shipping the HBM4 chips at the center of Nvidia's Vera Rubin AI accelerator program. Any delivery interruption risks pushing Nvidia and other hyperscalers toward SK Hynix, which already controls roughly 57% of the HBM market. The Suwon District Court is expected to rule on Samsung's injunction request any day. In its court filing, Samsung argued that "even a lawful walkout could cause massive damage" — a statement the union has noted directly contradicts the company's reassurances to investors on its April 30 earnings call.
Samsung Announced HBM4 Shipments First; SK Hynix's Timeline Differs
Samsung announced on February 12 that it had begun mass-producing HBM4 and shipping commercial units to customers, calling it a world first. The chip is built on Samsung's sixth-generation 10-nanometer-class DRAM process — internally designated "1c" — and incorporates a 4-nanometer logic base die, a departure from the conventional practice of adapting proven designs for HBM base dies. It delivers 11.7 gigabits per second of operating speed, 46% above the JEDEC industry standard of 8 Gbps. Samsung EVP and memory development head Sang Joon Hwang said the approach allowed the company to "satisfy our customers' escalating demands for higher performance, when they need them." Samsung expects its total HBM sales to more than triple in 2026 compared with 2025, and at its April 30 earnings call, memory chief Kim Jaejune told analysts that customer demand fulfillment rates are at a record low — with buyers already lining up 2027 supply.
SK Hynix's Q1 2026 earnings call, held April 23, confirmed a different timeline. HBM sales head Kim Ki-tae said HBM4 is "under development with key customers from early stages" and that the company is "preparing for timely ramp-up." SK Hynix's next-generation chip, HBM4E, is targeted for sample shipments in the second half of 2026 and mass production in 2027, using the 1c nanometer process for its core die. SK Hynix has separately said that customer demand for HBM4 over the next three years already exceeds the company's supply capacity. Where SK Hynix does lead is in market position: Counterpoint Research analyst Jeongku Choi has placed SK Hynix's share of Nvidia's HBM orders at approximately 70%, with UBS projecting the company will capture a similar share of the HBM4 market for Nvidia's Rubin platform.
Samsung's Record $7.6 Billion R&D Quarter and What It Funded
Samsung's 11.34 trillion won Q1 R&D budget covered completion of HBM4 volume production and the start of HBM4E development, with first samples planned for Q2 2026. The quarter also saw the launch of the Exynos 5410, a chip integrating three satellite communication technologies on a single die, and the mid-range Exynos 1680 application processor. Total capital expenditure for Q1 reached 11.23 trillion won ($7.54 billion), of which Samsung's Device Solutions division — which oversees semiconductors — accounted for 10.19 trillion won, or roughly 90.7% of the total. For the full year, Samsung plans to invest more than 110 trillion won (approximately $73.8 billion) in facilities and R&D, focused on AI semiconductor technology and new businesses including robotics. At its March 18 shareholders' meeting, CEO and Vice Chairman Jun Young-hyun told shareholders that Samsung is "the world's only semiconductor company capable of providing one-stop solutions from logic to memory, foundry, and packaging" and committed to building the "technological competitiveness to secure leadership in the AI semiconductor market."
SK Hynix's spending pattern shows the same acceleration in sharper relief. The company spent 2.55 trillion won on R&D in Q1 2026 — 68.3% above Q1 2025 and more than double the average of roughly 1.11 trillion won per quarter it maintained across 2022 through 2024. Personnel costs made up the largest single component at 1.36 trillion won ($910 million). SK Hynix's quarterly R&D budget first crossed 2 trillion won in Q4 2025, and analysts have tied the acceleration directly to the surge in demand for AI-focused memory through last year. The company's first-quarter capital expenditure totaled 7.35 trillion won ($4.93 billion), up 21.9% year over year, with assets under construction — including new semiconductor manufacturing plants — representing the largest share at 6.02 trillion won. SK Hynix is also building a new advanced packaging plant in Cheongju at a cost of 19 trillion won ($12.85 billion), with construction already begun.
Laptop Buyers and Enterprise IT Are Paying the AI Tax
The two companies' combined shift toward high-margin HBM has produced a separate crisis for the broader memory market. Every HBM wafer produced displaces approximately three conventional DRAM wafers in manufacturer capacity, according to industry analysts — a dynamic IDC has called unprecedented. Goldman Sachs raised its 2026 DRAM supply-demand gap forecast in April from 3.3% to 4.9%, characterizing it as the most severe memory shortage in 15 years. Samsung's memory chief Kim Jaejune has said supply is already falling well short of demand; SK Group Chairman Chey Tae-won has publicly stated the shortage is likely to persist until 2030.
Gartner senior principal analyst Kanishka Chauhan has confirmed that HBM capacity reallocation is directly reducing the supply of consumer and industrial DRAM, predicting DRAM prices will rise 47% in 2026 alone due to significant undersupply. Gartner separately estimates combined DRAM and SSD prices could be 130% higher by the end of 2026 than they were at end-2025 — implying a roughly 17% rise in average PC prices and a projected 10.4% decline in unit shipments as consumers delay purchases. DRAM contract prices already surged approximately 90–95% in Q1 2026 compared with Q4 2025, and Samsung and SK Hynix plan further hikes in Q2. AMD has warned investors of a 20% decline in gaming revenue in the second half of 2026 attributable to higher component costs. Silicon Motion's CEO has stated publicly that simultaneous shortages in HDD, DRAM, HBM, and NAND — "all in severe shortage in 2026" — represent something the industry "has never faced before."
Samsung's Kim Jaejune has defended the allocation on the April 30 earnings call, arguing that focusing solely on conventional DRAM "would constrain AI infrastructure build-out" and that the margin gap between HBM and standard DRAM is expected to narrow significantly by 2027. That framing offers no near-term relief to mid-market enterprise buyers, who are now competing against cloud hyperscalers — Google, Microsoft, Amazon — that have locked in multi-year supply contracts. As Gartner's Chauhan has noted, customers already anxious about access to 2026 supply are "actually bringing forward their demand for 2027."
US Export Rules and Patent Lawsuits Narrow the Margin
Both companies are navigating compressing regulatory room in Washington. The US Bureau of Industry and Security removed Samsung and SK Hynix from its Verified End User program in September 2025, ending the long-standing exemption that had allowed them to bring American chipmaking equipment into their China facilities without individual export licenses. Washington granted both companies annual licenses for 2026, but the framework now requires yearly renewal — adding material uncertainty to future China operations at a moment when demand is at historic levels.
On the legal side, Samsung faced 86 patent lawsuits in the United States in 2024, up 70% from 51 the previous year, according to Korea JoongAng Daily. Netlist, a US memory technology firm, has been among the most persistent adversaries, with an HBM-specific suit involving die-stacking technology filed in 2025. SK Hynix faces a separate action from Monolithic 3D, which has accused the company and Japan's Kioxia of violating Section 337 of the US Tariff Act through sales of NAND and DRAM products allegedly infringing its patents.
Five Variables in the Next 90 Days
The most immediate: whether the Suwon District Court issues an injunction before May 21, and whether Samsung's workers carry through with the walkout regardless. Beyond the labor dispute, the market will watch whether SK Hynix delivers HBM4E samples on its H2 2026 schedule; whether Samsung's foundry division can ramp 4-nanometer base-die production fast enough to sustain its HBM4 assembly lines at scale; whether either company's China operations face disruption from the annual US licensing cycle; and whether the strike threat drives any near-term procurement shift from Samsung to SK Hynix or Micron at Nvidia and other hyperscalers.
For enterprise buyers, both companies' 2026 HBM capacity is already committed. Mid-market firms waiting for spot pricing are competing against Nvidia, Google, and Microsoft for residual supply in a book that SK Hynix's own management has described as sold out. For consumers, the most direct consequence is hardware pricing: analysts from Gartner, IDC, and TrendForce all agree that meaningful DRAM price relief will not arrive before late 2027 or 2028. Anyone who needs a laptop, server, or consumer device with significant DRAM between now and then is buying into a market where the price direction — shaped entirely by two companies' decision to prioritize AI infrastructure over consumer supply — runs in one direction only.
ⓒ 2026 TECHTIMES.com All rights reserved. Do not reproduce without permission.
