The AI boom runs on memory chips. And the people who make those chips are about to walk off the job.

On May 21, nearly 45,000 Samsung Electronics workers plan to begin an 18-day strike — the largest walkout in semiconductor history. Their target: the fabrication plants that produce the high-bandwidth memory (HBM) chips without which every Nvidia GPU, every AI data center, and every large language model is functionally dead weight.

This isn’t a labor footnote. It’s the AI industry’s first real hardware supply chain crisis.

The Math Is Brutal

Samsung produces roughly a third of the world’s DRAM and a massive share of HBM. Only three companies on Earth make HBM at all — Samsung, SK Hynix, and Micron. There is no Plan B if Samsung’s 12 fabrication lines go dark.

JPMorgan estimates the strike could slash Samsung’s operating profit by $14 billion to $20.8 billion. Sales losses from the production halt alone could hit $3 billion. A one-day walkout in April already gave a preview: foundry output dropped 58% and memory fabrication fell 18% during the affected shift.

And you can’t just restart chip fabrication like flipping a light switch. Samsung has already begun “warm-down” procedures, scaling back wafer inputs, because halting mid-process means scrapping wafers that cost $20,000 each.

This Is Really About Who Gets Paid

Strip away the logistics and this strike is a referendum on a simple question: who profits from the AI revolution?

The backstory starts with SK Hynix. Last September, Samsung’s chief rival settled with its union to allocate 10% of annual operating profit to employees for the next decade, with no bonus caps. Based on 2026 forecasts, that translates to average payouts of $460,000–$477,000 per worker this year. Projections approach $900,000 per person next year.

Samsung’s workers did the math. Their union is requesting 15% of operating profit, removal of a cap limiting bonuses to 50% of base salary, and a 7% wage hike. Samsung countered with roughly 13% — but only as a one-time payment for 2026, with no structural commitment.

Here’s where it gets ugly. Samsung proposed bonuses of 607% of annual salary for its 27,000 memory chip workers, specifically designed to beat SK Hynix. But for the other 23,000 workers — engineers building AI chips for Tesla and Nvidia in the foundry division — the proposed bonus was just 50% to 100%. These workers often share the same buildings as their memory colleagues.

Union leader Choi Seung-ho put it plainly: “If the memory division gets 500 million won while the foundry division only gets 80 million won, what motivation would those employees have to keep working?”

Samsung Is Already Bleeding Talent

The strike hasn’t started and the damage is underway. Roughly 200 Samsung engineers have defected to SK Hynix over the past four months. Foundry teams in Pyeongtaek have “shrunk sharply” as colleagues jump ship.

This directly undermines Samsung Chairman Jay Y. Lee’s vision of becoming the “clear No. 1” in logic chips by 2030. Samsung is the world’s only semiconductor company attempting a true one-stop shop — memory, logic design, and foundry manufacturing under one roof. TSMC focuses on foundry. Micron focuses on memory. Nvidia designs but doesn’t fabricate.

The internal pay gap is actively sabotaging that strategy. Samsung negotiator Kim Hyung-ro defended the disparity in internal transcripts: “They 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.”

That kind of rhetoric doesn’t exactly inspire loyalty from the engineers you need to execute your most ambitious strategic bet.

The Ripple Effects Hit Everything

If 45,000 workers walk, the shockwaves won’t stay in South Korea.

The global memory market is already stretched thin. Earlier this year, Samsung entered pricing negotiations with Apple for iPhone 17 memory chips planning to push for a 60% price increase. Samsung opened with a demand for 100% — a full doubling — and Apple accepted immediately. That’s how tight supply is.

An 18-day shutdown creates cascading delays across AI hardware. GPU shipments slow because Nvidia’s processors need HBM chips stacked alongside them. Server costs spike. Data center buildout timelines — already stretched across hundreds of billions in planned hyperscaler spending — slip further.

The American Chamber of Commerce in Korea has warned that this kind of labor uncertainty could damage South Korea’s reputation as a dependable manufacturing partner.

South Korea’s “AI Dividend” Debate

The strike has ignited an even bigger political conversation. On May 12, presidential policy chief Kim Yong-beom suggested South Korea should pay citizens a “dividend” from the AI boom, comparing it to Alaska’s Permanent Fund for oil revenues.

The market panicked. The KOSPI fell 5.1% intraday, shedding over $300 billion in value, because investors read it as a potential tax on Samsung and SK Hynix — which together account for nearly half the index’s total market cap. Kim quickly clarified he meant redistributing existing tax revenue, not imposing new levies. The market recovered to record highs the next day.

But the underlying tension is real. Samsung and SK Hynix are projected to post around 500 trillion won in combined operating profit in 2026. South Korea’s entire economy is riding on the AI memory boom. The question of who benefits — shareholders, workers, or citizens — is no longer theoretical.

The Bigger Picture

Government-mediated negotiations continue, but a 17-hour session at the National Labor Relations Commission on May 13 failed to produce a deal. Samsung workers have already rejected a $340,000 one-time bonus, demanding structural changes instead. The union wants co-CEO Jun Young-hyun to personally present concrete proposals.

Here’s what makes this story matter beyond Samsung: it’s the first time the AI boom’s labor dynamics have reached a genuine crisis point in the hardware supply chain.

We’ve spent years debating whether AI will take jobs. We’ve barely discussed whether AI is fairly compensating the people who build its physical infrastructure. If AI companies are spending $725 billion on infrastructure this year, what’s the right share for the workers on the fabrication floor?

Samsung’s 45,000 workers are forcing that conversation. The answer starts May 21.