Transfer

I spoke with several product managers during my travels over the past two weeks. I'd like to share some interesting findings:

Sentiment: Overall very panicked and confused.

Positions:

1. Shanghai hedge funds are extremely bullish on memory chips, partly due to the TRS restrictions I mentioned a few weeks ago. They are rapidly selling off the entire Chinese semiconductor sector and buying DRAM ETFs to reduce margin pressure.
2. Hong Kong hedge funds have mixed and highly divergent positions; bulls and bears are evenly matched.
3. US market makers are waiting on the sidelines—not because of stock price issues, but because they still don't believe long-term agreements can transform memory chips from cyclical valuations to growth valuations.
4. Macro hedge funds have cleared all memory/semiconductor positions.

Interesting viewpoints from both sides:

Bulls:

1. Supply chain checks show that prices and margins will remain at higher levels for longer.
2. AI capex in '27/'28 will be much higher than most expectations.
3. OAI/ATH's ARR shows no signs of stagnation. Codex's WAU is currently growing wildly.
4. The technical report for Kimi K3 bluntly implies that every semiconductor component is a bottleneck 😂
5. No macro risks; CPI/PPI are far below expectations.
6. Even if spot memory prices cannot rise, these memory companies can support their stock prices through large-scale share repurchases or dividends.
7. OAI/ATH is hiring a large team of solution engineers to quickly build vertical solutions (finance/healthcare/legal), looking for the next S-curve beyond coding.
8. Even if memory chips fade, the entire semiconductor sector will remain intact.

Bears:

1. Higher memory prices are unsustainable and will destroy demand. Customers will resist further price hikes, and the government may even intervene to force large-scale capacity expansion.
2. Memory prices will peak in Q4.
3. Capex in '27 is undeniable, but the budget outlook for '28 is unclear.
4. Even with solid fundamentals, positions are extremely crowded, and Korean retail leverage is a huge hidden risk.
5. Although CSPs have aggressive capex plans for '27, most deliveries will be delayed due to power supply constraints.
6. There is no intermediate story of rapid AI application adoption like AI coding.
7. Higher prices and more long-term agreements in an upcycle do not prove that memory chips are no longer cyclical. We need to see resilience in a downcycle to prove this, such as memory chips maintaining a profit margin floor during a recession.
8. Memory is core to AI: if you trade the death of memory, then everything dies.
9. Memory is no different from other commodities like silver, oil, or lithium carbonate. When they peak, their narrative and fundamentals look fucking great.