Samsung's main South Korean labour union is threatening an eighteen-day strike from 21 May, the largest in the conglomerate's history. JPMorgan puts the potential operating-profit hit at 21 to 31 trillion won ($14B to $21B). Chairman Jay Y Lee made his first public apology over the dispute on 16 May, and government-mediated talks resume on Monday.
The trigger is not pay levels in the abstract. It is the gap between them. Samsung has offered memory chip workers a bonus of 607% of annual salary, matching what SK Hynix has been paying for the last year. It has offered employees in its foundry and system LSI businesses (the people who manufacture and design the logic chips that go into Nvidia, AMD, and Tesla parts) bonuses of 50% to 100%. That is a six-to-twelve times spread, between people who often work in the same building, on the same wafers, with overlapping process technology.
The cross-subsidy that built Samsung's chip strategy
For about a decade, Samsung has run its chip business as a single Device Solutions Division spanning memory, system LSI design, and foundry manufacturing. The strategic pitch, repeatedly delivered by Jay Y Lee, is to be the only semiconductor company that offers a one-stop shop across all of these, in contrast with the specialists: Micron in memory, TSMC in foundry. The financial reality of that pitch is a cross-subsidy. Memory has thrown off massive profits, especially since the HBM-led AI memory boom that started in earnest in 2024. Foundry has lost trillions of won annually. Memory has funded foundry's continued capex and operating expenses, including its wage bill.
Samsung negotiator Kim Hyung-ro said the quiet part on the record, per transcripts reviewed by Reuters: "They, the logic chip business, posted losses in the trillions of won and honestly, if it had not been for our company, they probably would have gone out of business or closed down." Substitute "our company" for "the memory division," which is what he meant, and the structure is visible. Memory workers' productivity has been paying foundry workers' salaries. The new bonus proposal makes that explicit by refusing to share the memory windfall with the people who do not generate it.
Why the model is breaking now
Three things have changed simultaneously and they all point the same direction.
First, the AI memory boom has made the profitability gap between memory and foundry too large to paper over. HBM3E and HBM4 sales to Nvidia, AMD, and Google have lifted memory margins to levels Samsung has not seen in cycles. Foundry, meanwhile, has been losing money on subscale 3nm and 4nm runs and a customer book that is thin compared with TSMC's. The internal numbers are now too lopsided to share quietly.
Second, SK Hynix removed its bonus cap a year ago and started paying memory engineers roughly three times what Samsung paid for the equivalent work. That created an outside option Samsung had to match, or watch a steady defection accelerate. Samsung's 607% offer is essentially a defensive bid for memory workers against SK Hynix. The 50% to 100% foundry offer is what is left over after that defence is paid for.
Third, foundry workers have credible outside options too, and they are using them. One Pyeongtaek foundry engineer told Reuters his team has shrunk sharply over the past couple of years as people moved internally to memory or externally to SK Hynix. A thirty-year chip researcher said he has applied to Micron. The union argues, plausibly, that a six-to-twelve times bonus gap will push the rest of the logic talent out, into the memory division or out of Samsung entirely. Either outcome guts the foundry roadmap.
The "win at all cost" foundry bet
Jay Y Lee has publicly committed Samsung to being "clear No 1" in the logic chip market by 2030. That is a TSMC challenge stated as a corporate goal, with a deadline, in a market where TSMC has roughly 60% share and decades of customer-relationship and yield-learning advantage. Samsung's playbook to close that gap has been the usual one for a number-two challenger: aggressive process node bets (it pushed gate-all-around at 3nm before TSMC), aggressive pricing to win marquee customers (Tesla, Qualcomm, briefly Nvidia at various points), and patient capex through losses, funded by memory.
The bet only works if foundry can hold its talent through the loss years until the strategy matures. That is exactly what the bonus structure does not now permit. By signalling, in the most concrete way possible, that the company values memory output six to twelve times more than logic output, Samsung is creating the talent flight that kills its own 2030 target. Yonsei University governance professor Namuh Rhee, quoted by Reuters, framed it as a structural problem: "Samsung must enable foundries to become self-reliant." Translated, the conglomerate cross-subsidy is no longer politically sustainable once the profitable side's workers organise.
What the apology and the Monday talks mean
Lee's public apology on Friday and Samsung's replacement of its lead negotiator are signals that the company expects to move on the foundry side of the bonus offer. The pressure is multidirectional: South Korean President Lee Jae Myung has publicly criticised excessive union demands, the prime minister and finance minister have privately said a strike must be avoided, AmCham Korea has warned about the country's reputation as a reliable supply-chain partner, and Samsung's board chair has warned internally about capital outflows and a weaker won. None of those pressures favour the union's full demand (15% of operating profit into a bonus pool, plus removal of the 50% bonus cap). All of them favour some upward revision of the foundry bonus to narrow the gap.
The most likely landing zone is a foundry bonus in the 200% to 300% range, well below memory but well above the current 50% to 100% offer. That would let the company claim it preserved performance-based pay differentiation while giving the union enough to call off the strike. It would not solve the underlying problem, which is that the cross-subsidy model can survive a settlement but cannot survive another AI memory cycle if foundry continues to lose money.
What to watch
The Monday outcome. If talks fail again, the strike on 21 May lasts at least eighteen days and JPMorgan's $14B to $21B operating-profit hit becomes the floor estimate, not the ceiling. If they succeed, the deal terms reveal whether Samsung's "one-stop shop" strategy survives in its current form.
Customer reactions. Nvidia, AMD, Google, and Tesla are all in the affected supply chains. Any of them making contingency moves to Micron (for memory) or TSMC (for foundry) would mark the strike as the moment Samsung's customer-concentration insulation finally cracked.
Internal restructuring. The Yonsei professor's "Samsung must enable foundries to become self-reliant" is the spinout argument in polite language. A separately listed Samsung Foundry, with its own capital structure and compensation policy, has been discussed in Seoul for years and dismissed by Lee. The strike is the first event that makes the discussion politically harder to dismiss.
The 2030 target. If Samsung quietly drops the "clear No 1 in logic by 2030" framing in upcoming guidance, that is the company conceding the bet without saying so. Watch for softer language about "leadership in advanced process technology" or "strategic logic capabilities" in the next earnings cycle. The shift from numeric to qualitative goals is how multi-year strategic retreats are typically announced.
