‘The gap is nearly causing chaos.’
Choi Tae-won, chairman of SK Group, didn’t just drop a warning last week in Jeju. He scripted a narrative. AI chip demand will surge 60-100% next year, he said, while supply remains ‘near zero growth’. The market reacted with the usual Pavlovian spike in memory stocks. But I don’t hunt for the truth. I hunt for the story the data refuses to tell.
Let’s rewind the tape. Every semiconductor supercycle follows the same arc: a supply shock meets a narrative of perpetual scarcity. In 2017, it was DRAM for mobile. In 2021, it was automotive chips. Now it’s HBM for AI. Choi’s speech is the latest act in a well-rehearsed play—executives amplify fear to justify massive capex, secure government subsidies, and lock in customer loyalty. The question isn’t whether demand is real. It’s whether the narrative is structurally sustainable or a carefully constructed liquidity trap.
Context: The Historical Narrative Cycles of Memory
The memory industry is a textbook case of narrative decay. Each boom is born from a genuine technology inflection (PC, smartphone, AI) that concentrates demand on a few high-margin products. The dominant supplier (Samsung, SK hynix, Micron) then feeds the narrative of ‘unprecedented demand’ to raise prices and expand capacity. The bust comes when the laggards catch up, supply floods the market, and the narrative collapses into a glut. In 2018-19, DRAM prices fell 60% after the ‘cloud computing’ demand failed to materialise at the expected pace. In 2023, NAND prices hit rock bottom after the ‘data center’ boom cooled.
Now, Choi is setting the stage for Act 4. His 60-100% growth prediction is the hook. But the real story lies in the structural constraints he only hinted at: advanced packaging capacity.
Core: The Narrative Mechanism Behind ‘Zero Supply Growth’
Choi’s claim that supply is ‘near zero growth’ is technically misleading. It’s not that wafer starts aren’t increasing. It’s that the bottleneck has shifted from front-end fab capacity to back-end HBM packaging. SK hynix’s MR-MUF process and its competitors’ TC-NCF are the true chokepoints. Adding a new packaging line takes 12-18 months and costs billions. This is not a shortage of DRAM wafers—it’s a shortage of silicon interposers, TSV etching tools, and the skilled labour to operate them.
Based on my audit experience dissecting tokenomics during the 2017 ICO mania, I recognise the pattern. When a protocol’s value is concentrated in a single, hard-to-scale component (like a vesting schedule or a yield farm), the team cries ‘scarcity’ to pump the token. Here, SK hynix is the protocol, HBM is the farm, and the packaging line is the smart contract bottleneck. The difference? In crypto, you can fork. In semiconductors, you can’t.
Sentiment-Data Synthesis
Let’s quantify the narrative. The market currently prices SK hynix at a forward P/E of 18x, assuming HBM margins stay above 40% through 2026. That implies investors believe Choi’s script: demand will double, and SK will capture 50%+ of the incremental supply. But the data tells a different story.
- Customer concentration risk: Over 90% of HBM3E revenue comes from NVIDIA. If Samsung qualifies its competing HBM3E in Q1 2025 (as analysts project), SK’s share could drop to 35% within two quarters. That’s not a linear decline—it’s a cliff.
- Depreciation gravity: Every new packaging line adds $3-5 billion in capex, depreciated over 7 years. If demand plateaus in 2026, those fixed costs will crush margin. The same dynamic killed the 2018 DRAM boom.
- Geopolitical leverage: Choi called the supply issue a ‘national security’ problem. That’s code for ‘we need government money’. But it also signals vulnerability. If US-China tensions escalate, SK’s Chinese factories (Wuxi, Dalian) could be plant shutdowns, wiping out 30% of total output.
Chaos is just a pattern you haven’t decoded yet. The pattern here is that every semiconductor bull run ends when the narrative of scarcity meets the reality of overinvestment. Choi is accelerating that clock by raising expectations beyond what the supply chain can deliver sustainably.
Contrarian: The Warning Is a Sell Signal, Not a Buy
Here’s the counter-intuitive angle: Choi’s speech is a trap for momentum traders. By framing the supply gap as insurmountable, he’s sending a signal to his own competitors (Samsung, Micron) that the market will absorb any capacity they throw at it. Expect Samsung to announce a $50 billion packaging investment within six months. Expect China to accelerate its HBM alternatives (CXMT, Huawei). The result? A supply flood in 2027 that makes the current shortage look quaint.
Decode the script before you bet on the actor. Choi’s real audience isn’t the public—it’s the Korean government and the banks funding his expansion. He needs to justify the $100 billion+ investment plan for the Yongin cluster. The ‘demand explosion’ narrative is his pitch deck. Every investor who buys SK hynix stock at 18x earnings is essentially funding Choi’s lobbying effort.
I’ve seen this before. In 2021, during the NFT utility fallacy era, founders touted ‘community ownership’ to justify token prices. I spent three months interviewing community members and found that 90% of governance was controlled by a single wallet. The narrative collapsed when a major founder sold his stash. Here, the single wallet is NVIDIA. If Jensen Huang decides to dual-source HBM or switch to a custom memory interface (like HBM4 with different pinouts), SK’s monopoly premium vanishes overnight.
Takeaway: Follow the Incentives, Not the Headlines
So what’s the next narrative? Look at the companies investing in advanced packaging equipment—ASML, Applied Materials, Tokyo Electron. Their guidance will tell you if Choi’s supply constraint is real or exaggerated. If logic die prices stay flat while HBM prices rise, it’s a packaging bottleneck. If everything rises, it’s a broad demand story. The market will decode the script within two earnings cycles.
For now, treat Choi’s warning as what it is: a strategic negotiation tactic. The true narrative is being written not in Jeju, but in the cleanrooms of Hwaseong and the procurement offices of Santa Clara. I’d rather bet on the pattern than the pitch.
— I don’t hunt for the truth. I hunt for the story the data refuses to tell.
