C.W.K.
← EssaysMarket

A Confession of Fear — AI Boom, Shipyard Hopeium, and Honey-Butter Chips

A Confession of Fear — AI Boom, Shipyard Hopeium, and Honey-Butter Chips

Here’s why battle scar tissue matters: you can immediately draw parallels from the battlefields you survived.

It’s not about right or wrong. It’s about the gut tell that something’s off. If it is, you don’t rush Omaha Beach for the last drop of loot.

TSMC’s CEO can post record numbers and still sound anxious, because the real story isn’t the quarter—it’s the capital cycle.

On TSMC’s Q4 2025 earnings call (Jan 15, 2026), C.C. Wei admitted he was “very nervous” about the possibility of an AI bubble, precisely because TSMC is committing tens of billions in capex—and a misstep would be “a disaster for TSMC.”

That single quote captures the modern semiconductor paradox: the world’s most strategically important foundry is trapped between the fear of overbuilding and the fear of not building enough.

The trap is structural. When demand looks infinite and everyone else is scaling, “prudence” starts to look like surrender.

The uncomfortable part is that TSMC and its customers are making decisions from the same vanishing point: the belief that AI reliably translates into durable profit pools. Wei says he met customers and even checked that they were “very rich.” But richness doesn’t prevent coordinated error—it only funds it.

This is why the Korean shipbuilding episode is such a clean parallel.

In the super-cycle before the 2008 GFC, yards expanded docks and capacity on the assumption that the forward orderbook represented stable, monetizable demand. When the macro regime flipped, the orderbook proved to be a mirage of timing, financing, and optimism—leaving excess capacity and brutal restructuring. Coverage from the period emphasized that “body expansion” through overinvestment fed destructive competition and pain even for large players.

The lesson wasn’t “don’t grow.” It was: don’t treat a boom orderbook as a permanent state of the world.

Some of their customers went out of business.

The hopeium was a loop: listening to wishful customers doesn’t count as evidence of demand.

The honey-butter-chip (허니버터칩) craze in Korea is the consumer-scale version of the same mechanics, almost cruel in its clarity.

In late 2014, Haitai (해태) launched a sweet-and-savory potato chip that became an overnight sensation. Within weeks, stores couldn't keep it stocked. Black-market resellers flipped bags at 3–4x retail. Social media turned "I found one" into a flex. The craze peaked in early 2015—a snack had become a status symbol.

Haitai initially resisted dedicated expansion and tried to satisfy demand within existing constraints—instinctive risk management. But scarcity amplified hype; demand surged partly because demand couldn't be met. Eventually, the company committed to expansion—reports described plans to double production. When the new capacity arrived, the market had already peaked; later reporting pointed to extremely low utilization.

Scarcity had been part of the product.

Once, when I was lucky enough to get one, all the kids in sight surrounded me asking where I got it. That was the magic.

Personal tell: I lost interest entirely when the shelves were full of its varieties. The hunt was the product. Once supply caught up, the magic died. And no kids groupies.

First cockroach intuition. Always works.

Put these together and the common failure mode becomes visible: the “signal” used for forecasting is contaminated by the system’s own constraints.

Ship orders rise because everyone expects freight growth. Chip demand looks insatiable because AI projects race to secure scarce capacity. Snack demand explodes because shortage itself becomes marketing.

In each case, planning under scarcity produces behaviors that make the future look more certain than it is. Then the moment capacity catches up, the illusion breaks: freight rates collapse, product hype fades, or AI spend faces scrutiny when ROI disappoints.

TSMC’s fabs are world-class. The question is whether the customers’ customers ever show up.

TSMC is trying to thread the needle: invest aggressively enough to remain indispensable, but carefully enough to survive the day the story stops compounding.

The punchline is that nobody can opt out. If every actor waits for certainty, they lose by default to the actor who builds first.

Bubbles don’t require stupidity—only incentives, synchronization, and a shared thesis.

That’s why the most honest line in this whole saga isn’t a bullish forecast.

It’s a confession of fear.

Chuck Prince, CEO of Citigroup, Financial Times interview, July 9, 2007:

“When the music stops, in terms of liquidity, things will be complicated. But as long as the music is playing, you’ve got to get up and dance. We’re still dancing.”

A few months later he was out. The following year, Citi needed a $45 billion rescue.

The kicker: Paulson later recounted that before that interview, Prince had essentially asked regulators if there was anything they could do to stop banks from taking such large risks. He knew. He still danced.

That’s the trap. Not stupidity. Incentives, synchronization, and a shared thesis.

So, are they still dancing?

Yup, they are.