# A Car Stuck to a Wall, and a Number That Came In Three Times Too Small — Notes on Gooaye EP706

Source: Realpha Blog (blog.getrealpha.com)
Original article and charts: https://blog.getrealpha.com/en/blog/gooaye-2026-10-17-ep706/

> Listening notes on Gooaye EP706 (2026-10-17): how the Tesla Roadster demo is a jumping car rather than a flying one, and why a NT$5 billion margin-debt drop where the host expected NT$15-20 billion pushed him defensive. Educational commentary only, not investment advice; no tickers recommended and no price targets.

Published: 2026-10-18
Locale: en
Tags: Gooaye, podcast notes, risk management, semiconductors, AI infrastructure


![A sports car pinned at ninety degrees against a tall track wall, dust pulled up beneath it, the track receding into night toward the horizon](/covers/gooaye-2026-10-17-ep706-cover.png)

> See an opening and advance; know the difficulty and withdraw.
>
> —— *Zuo Zhuan*, 12th Year of Duke Xuan (Spring and Autumn period; translation mine)

In Gooaye EP706, released 17 October 2026, host Hsieh Meng-kung talked about two things that look unrelated. One: after watching the Tesla Roadster launch, he upgraded his order from Performance to Frost on the spot, wired another NT$6 million, bringing his deposit to NT$7.5 million — while warning listeners that the thing is a jumping car whose pressure tank fires once, with production slated for 2028. Two: in the same week he turned defensive on Taiwan equities. Thursday saw a heavy selloff; intraday he told friends margin debt would fall NT$15–20 billion that day. The number printed at NT$5 billion, and that gap convinced him the shakeout was unfinished. Put together, the two stories show someone willing to pay up for a product and willing to step back from a tape he can't read.

## 16 October: a car pinned to a wall

He called it the most shocking footage he has seen, roughly level with the first time SpaceX caught a booster with the chopsticks on a sea platform — maybe a notch above it, because rockets are far away from him and this car he had already ordered.

He expected a pretty concept car, something like the Porsche Mission S, and wanted to know what the buttons looked like. Instead the demo had a track like an amusement-park coaster. The car drove up, hit ninety degrees, kept climbing until it hung roughly parallel to the ground. He assumed he was watching animation. The mechanism: high-pressure fans in the body, side skirts dropped to pull the underfloor toward vacuum, roughly two atmospheres of pressure differential holding the car to the surface. Then it reversed, charged again, launched off the end, rotated upright mid-air on thrusters, and blew gas downward to land on the reaction force.

![A car is stuck to a vertical wall with three arrows pointing from the car into the wall to represent suction and one arrow pointing down for gravity; on the right, two bars of very different heights show that the pressure difference is far larger than the weight.](/figures/roadster-wall-pressure-en.svg)

His reaction was to call the whole family into the living room. His wife watched and said *what the f***.* That bit made me laugh, because I had the same blank "what am I looking at" moment watching a booster land.

Then he pulled the excitement back, and this is the line worth keeping: it's a jumping car, the tank fires once and it's gone, there is no sustained output. So the Taiwanese media interviewing tech executives about whether road rules need rewriting for flying cars are aiming at the wrong target. He added a point from an interview he liked: humans already have the flying car, it's called a helicopter, and helicopters need air clearance.

Why bother with any of it? Because a car's acceleration ceiling sits in tyre friction. However much power you have, tyres that can't hold the ground just spin. Downforce presses the tyres in, rear thrust adds push, and the demo showed 0 to 100 km/h in 0.8 seconds. The downward-firing part he reads as a capability demo — what you get when a rocket company and a car company have a child.

![The upper curve rises to a single tall spike and then drops straight to zero, while the lower curve holds a flat, sustained level, contrasting a one-shot burst with continuous output.](/figures/one-shot-vs-continuous-en.svg)

The first thing he did after the stream was upgrade, hoping an early wire means an early car. He guesses the Performance version lands at NT$12–14 million and Frost at NT$17–20 million. Above NT$25 million he'd call it expensive, and would buy it anyway, since a mid-to-high Ferrari sits in the same range and this car pushes one spec to the limit. For his NT$7.5 million deposit he got no thank-you note and no flowers, consistent with a service centre where you fetch your own drink from the mini fridge.

## The same week's tape: a Thursday selloff, a Friday that wouldn't bounce

The market half of the show opens with him saying the tape keeps getting more brutal, and that even his own shift toward an anti-momentum approach feels uncomfortable now. The signal was Thursday's drop followed by a Friday rebound that failed.

His picture of a late-stage bull run is the Fall Guys level where the floor tiles disappear one by one and fewer places remain to stand. You can still find limit-up names, but big money can't fit into them. Small accounts conclude the tape is fine. Mid-size and large accounts start running out of places to go. His own handling: keep sixty to seventy percent invested, but pull cash back from momentum names, theme names, and anything not on the long-hold list, then wait and re-identify what is workable.

![On the left is a complete grid of floor tiles; on the right only a few scattered tiles remain, overlaid by a dashed circle far larger than any single tile, showing that big money no longer fits.](/figures/shrinking-floor-tiles-en.svg)

## The number that came in three times too small

This is the part I keep coming back to, because it's a complete forecast-versus-outcome cycle.

Intraday Thursday he told friends margin debt should drop NT$15 billion, NT$20 billion if the market slid further. His basis: the OTC index had hit minus five percent, the weighted index was down hard, and a pile of names were heading limit-down, which in his experience scares a cohort out. The print arrived after 9pm: NT$5 billion.

![Two bars differ sharply in height: a tall bar on the left stands for the intraday expected range of margin reduction, while the short bar on the right is less than a third of it, with a bracket marking the gap between them.](/figures/margin-expectation-gap-en.svg)

A friend offered an explanation — people have been beaten up so often they've gone numb and stopped moving. He rejected it, because in his experience nobody sits numb through the whole thing; positions that don't get washed out get washed out later. With the index back at a cluster of moving averages, he leaned toward one more flush.

What I like here is the sequence: write the forecast, check the answer, and when the answer disagrees, adjust the position rather than the story. The friend's explanation sounded reasonable, and he declined it by comparing it against shakeout patterns he has watched before.

## After TSMC spoke, the good news didn't lift the tape

TSMC reported in the same week. He read the results as a beat, noted C.C. Wei saying they'll take whatever business they can, and said the company's tone was positive — he saw no crack in the armour. The problem was the stock couldn't rally afterward, and the index stayed flat on the floor.

So he pushed one layer down: if news this good can't lift the market, what comes next that can? Running with friends that morning the only candidate anyone could name was sentiment relief if the US war authorisation lapsed into a stop. Nothing else. When you can't name the next card, you take some chips off the table.

He also offered a cross-check: the worst-hit market that week was Korea, where roughly half the index weight sits in memory. Memory is the tightest, hottest corner of this cycle. If that corner can't hold and the index is setting up to break lows, components and hardware broadly deserve caution. His expectation is specific: revenue and earnings keep coming through on a long view, while multiples get compressed first — unavoidable, as he sees it.

![Four blocks stacked top to bottom represent the supply chain, with the topmost layer labelled as the tightest in supply showing a crack, and an arrow on the right pointing downward to show every layer below must be revalued.](/figures/tightest-layer-cracks-en.svg)

## Where the cycle sits: infrastructure, platform, software

He brought back the three-act sequence he has used for years: infrastructure first, then platform, then applications — software. Not his invention, he says; past cycles ran this way and he just says it out loud. For years the market argued this time is different and software is dead. Software didn't die.

![Three humped curves hand off one after another from left to right; the first has already rolled over and the third is still rising, with the money marker sitting on the third and a dashed line above marking the prior peak.](/figures/infra-platform-software-baton-en.svg)

This leg, he notices money rotating into software, with most of the US names making new highs sitting there, so he's looking at software again. He also marks his own error: his US book didn't make a new high this leg because too much of it sits in hardware, AVGO included. The decision point ahead is whether hardware can keep up on the next bounce. If it can't, the habit of handing a well-positioned hardware name thirty times earnings may not come back.

Two bear narratives were circulating and he walked through both. First, CoreWeave's debt printing at junk-like yields, which may mean nobody wants it, with people asking whether the company blows up. He has never liked the neocloud layer: AI exposure runs from power to clouds to component manufacturing to software, so why pick the riskiest layer, especially with SpaceX as a possible entrant into that business.

Then he revised half his own view on air. A friend told him to look at the actual capacity at CoreWeave and Nebius — not that large, so if things break, a hyperscaler or NVIDIA likely absorbs them and the company survives. He accepted that, and drew the boundary: a company surviving and a stock holding up are separate questions, since price collapses don't track operations. The second narrative was a sequel to July's blowup, with claims that another AI fund got caught. His stance: in a drawdown every bad headline gets recruited as the reason, and six months later nobody remembers which one it was.

On positioning he got concrete. Friends are buying the dip through SOXL; he avoids it. His ordering for dip-buying is S&P over Nasdaq over SOXL, and in a real crash his circle tends to buy the index first to capture the beta snap, then pick names. He also admits he never catches the low — he buys and it falls another twenty or thirty percent, and that's the cost. His leverage warning: even if your direction is right, even if this is only a technical pullback, you don't know how deep a technical pullback goes, so anyone running 1.8 to 2.0 times should price the worst case first.

## Good news, flat stock — so what is actually falling?

If you watched the TSMC call this week, thought the numbers looked fine, and then watched the stock sit there, the question in your head is probably "then what is this selling off on?" I was asking it too.

Unpacked: good news confirms expectations, while price moves on changes in expectations. A strong call means the expectation was validated, not that anyone wants to add at this price. A good report against a flat tape says the good news is already in the price and no new marginal buyer showed up.

![A four-quadrant chart with news quality on the horizontal axis and whether the share price responds on the vertical axis, with the lower-right cell for good news and a flat price highlighted.](/figures/good-news-no-bid-en.svg)

Two things make this usable next time. First, keep "was the news good" and "did the news bring new buyers" as separate columns — one report can tick one and fail the other. Second, ask his question out loud: if this can't lift it, what's the next catalyst? If you can name it, hold and wait for that card. If you can't, cut the leverage. The Korea cross-check copies over too: watch whether the tightest link in the chain holds, because once it breaks, every layer queued behind it needs its multiple redone.

## Stuck at the highs — cut or hold?

One listener in the Q&A was loaded with CoreWeave, Oracle and Nebius, down fifteen percent from the high, saying his nerve only breaks below the July low. Another asked where to set a dip-buying zone on the index. Both are the same question: what decides cut versus hold.

His answer has two layers. Classification first: momentum names, theme names and anything off the long-hold list come back to cash, while names whose fundamentals you track can be held underwater because you know what you own. Then a hard line: breaking your name's prior swing low is a serious warning, and that is not the moment to white-knuckle it, because sometimes the whole thing disappears and some names never come back.

What I took from this is to make holding a conditional act. I'm allowed to hold if I can say where next quarter's business comes from. If I can't, it isn't long-term investing, it's refusing to book the loss. He also made a fair point worth copying: plenty of these companies have performed enormously well, so taking a pullback hurts, and the thing to examine is why the entire position was built at the highs — that's a scaling problem on your side.

## The one thing to take away

The idea I'm keeping from this episode: the moment an expectation fails carries more information than the news itself. He forecast a NT$15 billion margin-debt drop and got NT$5 billion, and that gap told him more than any headline — it told him the participants hadn't been washed out. Same with a strong TSMC call and a flat stock: the gap is the answer. We all collect news. Almost nobody keeps a record of their own expectations failing.

Here's something I've tried that needs no stock market: next time you're about to judge something, write the forecast as one sentence with a number in it, then add a second line — "if the result comes in at a third of what I expect, what judgement of mine changes?" That second line is the whole exercise, because it forces you to decide your response before you know the answer. I use it on waiting for my kid's test results, on a colleague's delivery date, on my own weight, and writing that second line is uncomfortable every time, since it tends to reveal that I wasn't planning to change anything whatever the result.