The Crisis Isn't About Salaries. It's About Documentation.
May 29, 2026
The Crisis Isn’t About Salaries. It’s About Documentation.
In May 2026, a labour dispute between Samsung and SK Hynix — two of the three manufacturers whose combined output accounts for roughly 90% of the world’s DRAM production — came close to becoming a political intervention in South Korea. Unions representing specialist engineers demanded the removal of bonus caps and a restructuring of pay. The threat wasn’t a conventional strike.
It was something more precise, and more costly: the possibility that a small group of engineers holding undocumented process knowledge in 3nm and 2nm fabrication — knowledge that lives in people, not manuals — might simply walk.
The chain of potential consequences stretched from Suwon to the data centres of Amazon, Microsoft and Google, through Apple’s assembly lines. All connected by a single thin thread: the tacit knowledge of a population of specialist engineers small enough, by most industry estimates, to number only in the low thousands worldwide — engineers who know how to manufacture next-generation memory.
The dominant narrative on technology risk talks about cybersecurity and geographic concentration in Taiwan. It talks far less about something more fundamental: the concentration of critical knowledge in human populations so small that individual behaviour can destabilise global supply chains.
That isn’t a new risk. It’s the same risk the world nearly failed to see coming in 1999.
The same pattern, 25 years apart
Y2K was never a physical infrastructure problem. It was a problem of knowledge concentrated in a generation of programmers who had made a reasonable technical decision — using two digits for the year, to save memory — without anticipating that those systems would still be running decades later. When the risk became visible in the 1990s, the world discovered something unsettling: the only people who could still read that code were COBOL programmers already retired, or about to be.
The knowledge wasn’t in documentation. It was in people. And that population was extraordinarily small.
Samsung and SK Hynix reproduce the pattern with unsettling precision. The fabrication process for next-generation semiconductors doesn’t live in public manuals. It lives in the procedural memory of engineers who’ve spent 10, 15, 20 years refining processes that can’t be transferred quickly — because there isn’t enough documentation to transfer them. That transfer happens through experience and mimicry, which works well as a learning mechanism but is fragile as the only mechanism of preservation. When that’s the sole route, knowledge has the shelf life of its youngest available carrier.
- Redundancy was never designed in.
- Critical knowledge was never documented.
- No one ever asked: what happens if this core group moves on?
Low risk preparedness, a reactive culture, inertial architecture. And the system kept running on inertia, because it worked well enough not to be questioned.
Technical debt, incentive debt — the same mechanics
Technical debt is a concept engineers understand well: take a shortcut today to ship faster, and you accumulate a future obligation that grows with time. Y2K was technical debt in its purest form — the two-digit decision was rational in 1965, when memory cost thousands of dollars per kilobyte. The error wasn’t the original decision. It was never revisiting it as the context changed. No one asked: does this shortcut have an expiry date?
What Samsung accumulated is structurally identical, but in the human domain: incentive debt. A salary structure designed for a pre-AI environment worked well for decades. The error wasn’t designing it that way — the error was never revisiting it as SK Hynix began competing aggressively, as demand for engineers surged with the AI boom, and as the global talent market began offering real alternatives.
The mechanics of both debts are identical:
Reasonable decision in its original context → context shifts → decision becomes fragility → fragility grows silently → breaking point → cost of repair exponentially greater than the cost of prevention.
And in both cases, the most dangerous moment isn’t the breaking point. It’s the interval between the signal appearing and the system recognising it as a signal. That interval — where the fragility already exists but no one is measuring it — is where the real damage accumulates.
The cost that appears on no budget line
Y2K cost an estimated $300-600 billion globally. But the economic figure doesn’t capture the more relevant cost: for years, the world’s most capable technical teams were busy repairing debt from the past instead of building the systems of the future. The same engineers who would have designed what comes next were auditing 1972 code instead. The cost of organisational attention has no line item. But it exists, and it’s structurally more damaging than the economic cost, because it atrophies exactly the capability the system needs most.
In the Samsung/SK Hynix dispute, the hardest cost to measure isn’t the pay restructuring or the political pressure. It’s the signal the talent market just received. Because specialist technical talent now knows — with hard data, not intuition — exactly how much negotiating leverage it holds. That information can’t be unlearned. And it permanently changes retention dynamics across the entire industry — not just at the two companies involved, but in any sector with concentrated critical knowledge.
The paradox of preventive work
Had Y2K remediation failed, we’d be looking at the largest technological disaster of the modern era. But it worked. And precisely because it worked, popular perception settled on the idea that the danger had been overstated.
Successful preventive work makes itself invisible. It leaves no evidence of what it prevented — only the absence of catastrophe, which the outside observer reads as proof that the catastrophe was never real to begin with. That invisibility carries a direct organisational cost: leaders who invest in prevention rarely get credit proportional to what they averted. Systems that stay stable look as though they never needed the investment that keeps them stable. Organisations end up disinvesting from prevention — until the next breaking point makes visible, at exponential cost, what the prevention had been quietly holding up.
If the Samsung/SK Hynix crisis stabilises, global data centres and consumers will never know how close they came to a cascading shortage. And within five years, someone will argue the risk was overstated.
Beyond semiconductors
This isn’t a story that belongs only to chip fabrication. The same mechanics apply anywhere technical judgement accumulates in a small population of specialists rather than in a system: in critical infrastructure, in MEP and HVAC design, in structural and industrial engineering. A senior engineer who’s spent 15 years learning exactly how a specific HVAC system behaves under a specific building’s load profile, or how a structural retrofit needs to account for a site’s particular soil conditions, holds knowledge just as concentrated — and just as undocumented — as the process engineers at Samsung. The industry doesn’t call it a talent shortage. It calls it experience. But the shelf life is the same.
The root problem the symptoms are hiding
This isn’t a crisis about salaries. Salaries are the symptom. The problem is that critical fabrication knowledge doesn’t live anywhere else — not in systems, not in documentation, not in transferable processes. It lives in people. And once those people discover how much power that gives them, the negotiation stops being about compensation and becomes something more structural: a question of who actually controls the system.
Y2K produced a lesson the market never fully learned: critical knowledge needs to live in systems, not only in people. In 1999, the tools to implement that at scale didn’t exist at the speed required. In 2026, they do.
Tacit knowledge — what Polanyi called “we know more than we can tell” — can’t be documented all at once. The engineer who adjusts a parameter by 0.3% because “last time the environment had this humidity, that’s what worked” can’t write that knowledge into a manual. But it can be captured at the moment it happens: systems that log the decision, its context, and its outcome, without depending on someone remembering to document it afterwards.
That doesn’t eliminate tacit knowledge. It progressively converts it into explicit knowledge, reducing dependence on single carriers with each iteration — building the redundancy the system was never designed with from the start.
What’s missing isn’t the tool. It’s the design that applies it to the right problem, before the problem becomes a crisis.
Design, or luck
Early movement and early action on a system’s internal signals are the only real competitive advantage against chaos. Not because they eliminate uncertainty — they can’t. But because they turn the response to chaos into something reproducible.
Acting early isn’t just a competitive advantage. It’s the only way that advantage can be attributed to design rather than luck. Because whoever acts late and survives can believe they survived through capability. Whoever acts early knows they survived by design. And that distinction matters — not just strategically, but for knowing whether the system can reproduce the result, or can only hope luck repeats itself.
Samsung and SK Hynix have spent decades building manufacturing systems of extraordinary precision. And at the same time, they let the knowledge that makes those systems work live only in people — without redundancy, without sufficient documentation, without an organisational immune system holding active memory of the risk.
Y2K taught them the same lesson it taught everyone. And like everyone, they forgot it once it stopped hurting.
The question that lingers: What percentage of your organisation’s critical knowledge lives only in people — and how long do you have before those people discover how much power that gives them?
If that question is worth sitting with, it’s worth a conversation. We work with engineering and infrastructure teams on exactly this problem — not after a crisis forces the issue, but before it does.