# Statementdog (財報狗) EP538 Notes: While Everyone Cheers the Compute Boom, I'm Asking Who Got Left Out
> Personal notes from Statementdog (財報狗) Podcast EP538: anti-bubble thinking, Korea's memory deleveraging, memory prices slowing but supply still tight, firms starting to rein in AI spend, and compute going tiered-and-cheaper. Plus how the episode echoes two things I've been practicing: separating noise from structure, and watching cross-market leverage. Educational notes, not investment advice; disclaimer at the end.
Published: 2026-07-10
Locale: en
Tags: podcast-notes, market-view, ai-compute, memory, education
TL;DR: This Statementdog episode flips the question: while everyone cheers AI compute, who got squeezed out by the hype (anti-bubble)? And when firms start saving on AI, has compute demand really peaked? Key points: anti-bubble thinking to dodge short squeezes; Korea's memory crash is deleveraging, not a fundamental reversal; price gains slowing but supply still tight; firms reining in AI spend; compute going tiered-and-cheaper yet total demand still rising. I use it to sort out two things I'm practicing: split noise from structure first, and put cross-market leverage into my risk model.
> *Pursue learning, and each day something is gained;*
> *pursue the Way, and each day something is let go.*
> — Laozi, "Tao Te Ching," ch. 48 (Spring and Autumn period); translation mine
> These are my **personal notes** from Statementdog (財報狗) Podcast **EP538** (published 2026-07-09). This is not a transcript and not official content. If you want the full thing, please support the original show. Below I've combined what the episode sparked for me with my own reflections.
## What the episode is about (one line)
While everyone is celebrating AI compute, this episode turns around and asks two questions: **who got squeezed out by the hype (anti-bubble)?** And **when firms start to "save" on AI, has compute demand really topped out?** For me, the most valuable thing here isn't any single stock — it's a way of thinking that refuses to stand in the loud center and instead looks off to the side.
## What I took from it
- **Anti-bubble thinking, to avoid the short-squeeze risk**: rather than shorting a hot bubble whose reversal is impossible to time (we've all heard the stories of shorting AI and getting squeezed to death), look for companies that got mispriced precisely because all the money got sucked into the hype. The show's example: traditional energy stocks that were left behind during the EV craze. The other side of a mania often has something undervalued standing on it.
- **Korea's memory crash is "deleveraging," not "a fundamental reversal"**: Korean retail investors piled into 2x-long memory ETFs; one regulatory warning triggered a deleveraging cascade. That's a **regional capital phenomenon**, not necessarily global memory fundamentals turning down.
- **Price gains are slowing, but supply is still tight**: research firms estimate Q3 DRAM/NAND contract prices still up double digits QoQ — just a slowdown from last quarter's spike, which the stocks front-ran. But new fabs take time, so the real supply gap is still there.
- **Firms are starting to rein in AI spend**: Uber, Meta and others are setting usage caps and removing the internal leaderboards that encouraged AI use. The honeymoon of "burn tokens without limit" is fading.
- **Compute is going "tiered" and cheaper**: send simple tasks to cheap models, use local or in-house chips (ASICs) to dodge the "Nvidia tax." Per-unit cost drops — but because it's cheap, usage grows, so **total compute demand is still climbing.**
## My own extensions
**First: anti-bubble thinking is really the same thing as the valuation discipline I keep practicing.**
My own logic has always been to say things clearly before the consensus forms, then watch bottlenecks and valuation rather than chase the narrative. "Anti-bubble," plainly put, means: when a story gets so hot that all the money crowds in, the things pushed out may already be unreasonably cheap. This isn't about catching every falling knife — it's a reminder that **the loud center is usually the most expensive, and the real mispricings sit in corners nobody watches.** The precondition is that the company itself holds up — orders, moat, cash flow — not cheap for cheapness's sake.
**Second: the Korea-deleveraging point lands right on something I've been building.**
Lately I've been building a "see-through-fake-diversification" risk check for my own portfolio — computing the **correlations** between my holdings to see whether a basket that looks diversified is actually one bet when it crashes, and watching cross-market leverage as a variable. So "Korean retail deleveraging may spill into Taiwan small-caps" isn't a headline to me — it's something to put into the model. **You can't watch Taiwan by watching Taiwan alone; when leverage loosens next door, sentiment travels across the sea.**
**Third: I love this episode's rebuttal of "compute has peaked."**
Firms saving money is easily read as "AI demand has topped." But the episode reminds us: unit cost falls, usage rises, total demand may still be climbing. That's the same reflex I keep practicing — **when bad news hits, the first move isn't to rewrite the conclusion, it's to ask "did the structure really change, or did they just find a cheaper way to do it?"** Cheaper isn't less demand; optimizing isn't peaking. Tell those apart, and you won't sell the trend halfway down the mountain during a transition.
## A note on my inner state this half-year
Honestly, the heaviest lesson this half-year wasn't any theme — it was **slowing down.**
I used to scramble for a script at the first market tremor, wanting to prove I could call the direction. Now I'd rather sort it out first: is this the index selling off (beta), or did this company actually break (single name)? Did the structure change, or is noise being amplified? "Anti-bubble" and "cheaper isn't peaking" are, at bottom, the same discipline — **not led by the noise, not scared off by the panic.**
What actually gives me security has never been calling one thing right; it's laying the foundations of life first, then letting research slowly grow into something that stands on its own. Once that clicked, every big market swing feels more like it's handing me a problem to solve than a grade on my report card.
Pursue learning and gain a little each day; pursue the Way and let a little go each day. Learning more chart tricks is the "gaining"; the hard part is "letting go" of that urge to prove myself right.
## Worth a look
- **Statementdog Podcast (original show)**: EP538 — for the full content, please listen on Statementdog's official podcast and support the creators.
- To make "separate noise from structure" a habit, the best drill is: on every big down day, write down "is this beta, deleveraging, or fundamentals" first — then read the news, not the other way around.
---
*This is a personal, educational reflection after listening to a podcast; it is not Statementdog's official content, and it is not a buy/sell recommendation on any security, offers no price targets, and does not target any current position. Companies mentioned are for illustration of the episode or a concept only. Investing carries risk; make your own decisions through your own research or consult a qualified professional.*