The Strike That Could Cost Samsung $20 Billion in 18 Days
More than 45,000 workers are threatening to stage the largest strike in Samsung Electronics' history from May 21, reducing production of memory chips that are crucial components in AI data centres, smartphones and laptops, as Samsung and its union struggle to find a compromise over bonus payouts. South Korea's President Lee Jae Myung has personally intervened, calling for both labour and management rights to be respected — a signal of how seriously the government views the potential supply chain consequences of a prolonged work stoppage at the world's largest memory chip manufacturer. The union estimates that an 18-day strike could cost Samsung about 30 trillion won, or roughly $20 billion. A previous rally on April 23 that mobilised 40,000 workers produced a 58% drop in foundry production and an 18% decline in memory output in a single day — a demonstration of the union's capacity to disrupt that neither management nor investors should discount.
The dispute sits at the intersection of two trends that make it structurally different from conventional labour-management conflicts. Samsung's memory chip business has swung from posting operating losses throughout the 2024 memory downturn to recording its highest-ever quarterly operating profit in Q1 2026 — nearly eightfold higher than a year earlier. By the end of 2025, Samsung reclaimed the overall DRAM market share lead after shipping High Bandwidth Memory to Nvidia and expanding legacy memory production, with HBM4 chips beginning mass production in February and the entire 2026 HBM4 production run already sold out. Workers who accepted zero performance bonuses in 2024 — when the chip unit was posting losses — are now watching the same company generate record profits while their compensation structure has not changed. The strike is the consequence of a boom that arrived faster than the company's internal reward systems were designed to handle.
The Core Dispute: Who Shares the Spoils of the AI Memory Boom
The union argues that the firm's other 23,000 workers responsible for making AI chips for Tesla and Nvidia — who often work in the same buildings as their memory colleagues — should not be left behind, despite the foundry business suffering billions in losses in recent years. This is the internal fault line that makes the Samsung dispute more complex than a simple wage negotiation. Samsung's "one-stop shop" semiconductor strategy — offering memory, logic chips, and foundry services under a single corporate roof — creates divisions with radically different financial performance in the same year. The memory division is generating historic profits. The foundry division has been losing money for years. Both sets of workers want to be compensated based on the company's overall trajectory, not the specific unit economics of their division. Management wants bonus structures that reflect divisional performance. The gap between these positions is what 17-hour negotiation sessions have failed to close.
SK Hynix, Samsung's primary DRAM competitor, settled with its own union in September to allocate 10% of annual operating profit directly to employees as performance bonuses for the next decade, with projections of average payouts approaching $900,000 per worker next year. Samsung's union is requesting 15% of operating profit be allocated to a bonus pool, removal of the current 50% cap on bonus-to-base-salary ratio, and a 7% wage hike. Management countered with roughly 13% of operating profit as a one-time payment for 2026, without committing to permanent structural changes. The delta between these positions, in the context of Samsung's Q1 2026 record profits, is not large enough to explain 17 hours of failed negotiations. The real issue is permanence: workers want a structural change to how Samsung shares profits, not a one-time payment that can be reversed when the cycle turns down.
The Supply Chain Consequence: HBM Disruption at the Worst Possible Moment
The timing of the Samsung strike is uniquely damaging for the global AI infrastructure build. High Bandwidth Memory — the specialised chip architecture required for AI accelerators including Nvidia's H100 and B200 series — is the single most supply-constrained component in the AI data centre buildout. Samsung Chairman Shin Je-yoon said he was worried about losing market leadership amid fleeing customers and falling competitiveness in the event of a strike, with JPMorgan estimating that if Samsung meets the union's demands in full, 2026 operating profit faces a 7%-12% downside from increased labour costs alone — and the total operating profit impact from both higher costs and lost production could reach 2.1 trillion to 3.5 trillion won in the base case. For Nvidia, whose data centre GPU shipments are already demand-constrained by HBM availability, a multi-week Samsung production disruption creates a direct bottleneck in the AI infrastructure supply chain at a moment when hyperscaler capital expenditure commitments are at their highest-ever levels.
Approximately 200 Samsung employees have already left for SK Hynix over the past four months, reflecting competitive pressure that has already become an issue for Samsung's talent before the strike has even begun. The talent migration dynamic adds a longer-term dimension to a dispute that is already consequential in the near term. SK Hynix's more generous profit-sharing structure is not just winning the current negotiation optics — it is winning engineers and process specialists whose knowledge of HBM manufacturing is among the most valuable in the global semiconductor industry. A prolonged Samsung strike that accelerates this talent migration while reducing production output would hand SK Hynix structural advantages that are difficult to reverse even after the dispute is resolved. The stakes of the May 21 deadline are therefore considerably higher than the near-term production disruption numbers suggest.
The Broader Implication: Labour's Claim on the AI Value Chain
The Samsung dispute is the most visible instance of a tension that is building across the technology sector: the question of how AI-driven productivity and profit gains are distributed between capital and labour. The AI boom has generated extraordinary shareholder returns across the semiconductor, cloud, and software sectors. It has not yet produced commensurate wage growth for the workers who manufacture the physical infrastructure on which that boom depends. Samsung's union members are not demanding a share of abstract AI value — they are demanding a share of the specific, auditable operating profit generated by the chips they manufacture in Pyeongtaek. The clarity of that claim, backed by the credible threat of production disruption at a moment of peak global HBM demand, makes the Samsung dispute a template for how labour in the semiconductor industry will increasingly negotiate in the AI era.