# Good at Buying, Hopeless at Selling? Use a Dumb Line — Notes on Gooaye EP690 > Reflections on the 2026-08-22 episode of Gooaye: the Marvell–Google chip agreement and the shift toward customer-owned tooling, a market that quietly reset everyone's baseline, and why an unglamorous exit rule beats a brilliant one. Educational commentary only — not investment advice, and no individual stock recommendations. Published: 2026-08-23 Locale: en Tags: Gooaye, podcast notes, investing discipline, semiconductors, exit rules ![A golf driving range at dusk, a long row of empty bays receding into the distance, one lone figure in the nearest bay repeating a swing](/covers/gooaye-2026-08-22-ep690-cover.png) > The old man said: "Nothing to it — only that my hand is practised." > > —— Ouyang Xiu, *The Old Oil Seller* (Northern Song, c. 11th century) An old man pours oil through the square hole of a coin into a gourd without wetting the rim. Onlookers gasp. He gives them five words: I have simply done it often. In the 2026-08-22 episode of Gooaye (EP690), somewhere between a film recommendation and a wine tangent, the host answers a listener who says he has a knack for buying near lows — win rate above seventy percent — but no feel at all for selling. The answer: "Just exit on the ten-day moving average. It's that dumb." It reads like a brush-off. Listen to the whole episode and it isn't. The wine segment, the segment on going full-time, the segment on chip orders — they all circle the same point: what usually costs people money isn't a lack of insight, it's one recurring decision that never got pinned down. ## What this episode covers The first half is personal: a space film, Michelin's new wine list, and the host's own journey through Burgundy, which he describes as attacking the top of the pyramid first. The middle returns to markets and the week's biggest item — the commercial agreement between Marvell and Google, and what it means for the whole chip design chain. The back half is listener mail: divorce, training, going full-time, biotech, employee share subscriptions, and how AI made everyone busier rather than freer. ## Key takeaways **1. Google's move changes *how* it buys, not *who* it buys from.** The Marvell–Google agreement covers products attached to the TPU ecosystem — inference chips, memory controllers, network interface silicon, near-memory compute. In exchange, Google receives warrants that, if fully exercised, amount to roughly 7% of shares outstanding, vesting against revenue out to around 2030. The host's reading: Google is committing harder to COT — customer-owned tooling — pulling the high-value pieces of a chip programme, design, masks, key component procurement, in-house. **2. Moving from turnkey to COT converts markup into fees.** Under the turnkey model, Google handed over specs and a partner handled design, memory procurement, packaging and test, marking up each component and quoting one total price. Very profitable work. Under COT, the design partner's revenue falls back to non-recurring engineering charges, IP licensing and design services. Margins compress — but the rules also get explicit, to the point where the customer will tell you upfront what you're allowed to make. **3. The same headline means opposite things to two different companies.** The market read it as bearish for the whole chip design group. The host disagrees, on a specific ground: the chips Marvell picked up here — one for a search-engine-specific application, plus a few sitting alongside networking and processors — are modest in volume and, crucially, **were not taken from anyone; they are new programmes**. The company with genuine reason to worry is the one whose turnkey business was working beautifully, because the five programmes it was assumed to win may now include two or three going the COT route. Other design partners could actually benefit: a customer this determined is opening more projects, not fewer. **4. When things fall, people rush to give the decline a name.** This is the line from the episode I'd keep. The host notes that nearly everything was pulling back that week, so there's little need to insist this particular news caused the drop. When prices fall, our minds grab the most visible headline of the day and hang causation on it — while "almost every ticker retraced this week" was sitting there as the simpler explanation. **5. Greed is relative, and the reference point moves without telling you.** At the deepest point of the late-July decline, he expected it might take until Q4, maybe next year, to recover lost ground. Half of it has been recovered already — and now people complain the tape is slow, sticky, and punishing to anyone chasing strength. The same price level went from unimaginable in July to disappointing in August, with nothing objective changing except the point we measure from. **6. Fill in the bottom of the pyramid before you judge the top.** Starting at the summit left him convinced the category was absurdly expensive. Only after working back down did he find bottles at a few thousand NT dollars that, poured blind, get mistaken for something ten times the price — and not just one of them. His conclusion isn't that expensive things are a scam. Above a certain price you are buying history, culture, the label itself, and that's a real value to the person paying. Mocking someone's hobby mostly advertises your own narrowness. The flip side: the mid and lower tiers are full of serious producers, and that dividend goes to whoever bothers to look. ## Going deeper ### "The news was good. Why is my stock down?" A clean way to take this apart, straight out of the episode. First: is this *capture* or *creation*? "Company X won an order" reads very differently depending on whether that order previously sat with someone else — zero-sum, someone gets hurt — or never existed before, in which case the pie is growing and there's simply one more slice-taker. The host's read on these particular chips is the second. The market that day priced the first. Two arithmetic paths, opposite conclusions, and the headline won't tell you which applies. Second: does this change *this year's numbers* or *the shape of the future*? COT doesn't dent anyone's next quarterly report. It changes how many programmes get awarded three years out, and in what form. Markets react to that fast and hard, because imagination doesn't have to wait for data — which is also why it can reverse entirely on the next headline. Third, and most often skipped: what was the broad market doing? If eighty percent of names retraced that week and yours fell 4%, the news may account for less than a point of it. Subtracting the market's move from your position's move is arithmetic, not judgement — and it's exactly what people skip when emotion is running. Worth saying plainly: these three steps reduce misreadings. They don't forecast prices. What they buy you is not acting on a story that turns out to be irrelevant. ### "I know when to buy. I never know when to sell." The listener's situation is specific: sharp entries, a win rate above seventy percent, no feel for exits, and positions that ride a whole hill up and back down again. The answer offered: the ten-day line — exit on the touch, try a third or a half first. Its value isn't precision. It's that it converts a feeling into a rule. Buying is an open question: you can think it over, sit it out, wait for the next one. Selling is nothing like that. The position already exists, the profit and loss moves in front of your eyes daily, and every "let's wait a bit" is itself a decision being made. With a position on, unrealised gains, and no rule, stable judgement is nearly impossible. So the first job of an exit rule isn't maximising return — it's **moving the decision from during market hours to before them**. The ten-day line will leave money on the table. It also means the question "do I sell today?" already has an answer before you open the app. A dumb rule executed a hundred times beats a brilliant judgement executed three times. One thing this listener already did right: he went back and reviewed his own trade history, so he knows where his edge is and where the hole is. Skip that step and you never learn that you can't sell — you just conclude you're unlucky. Measure your own shape first, then patch the hole. Not the other way round. A boundary worth drawing: mechanical exits fit the case where you've confirmed your entries have an edge and your exits don't. If you've never measured your entry win rate, what you need is a record, not a rule. ### "Should I just quit and do this full-time?" A lawyer in the mailbag is making exactly that move. The host's answer contains no romance: it isn't easier, it's a different job — instead of your boss grinding you, the market does. A salary at least arrives. He gets concrete about it: if July's slide had run through August and September, he says, he'd have been out holding a sign outside a property development, because income has to come from somewhere and you don't want to sell into a hole. The real point underneath isn't courage, it's that **cash flow contaminates judgement**. Once living costs must come out of the portfolio, you lose the option of doing nothing — and in most frameworks, doing nothing is a meaningful source of return. Letting a position play out requires not needing to touch it meanwhile. Cut the income and you'll be forced to sell your best holding at your worst moment, for reasons that have nothing to do with your analysis. So the thing to prepare isn't stock-picking skill; it's a long runway of living expenses kept entirely separate from the portfolio. That's also why he says the pressure exceeds a normal job: an office grinds you through people, while full-time connects every decision directly to the rent — and the market has no interest in whether you're short this month. Two other moments in the episode are the same lesson in different clothes. He recalls a 2018 biotech position where the trial passed, the readout was clean, the drug reached market — everything he needed happened — and then it simply didn't sell well. Every step of the reasoning held and the conclusion was still wrong, because one link in the chain wasn't visible from where he stood. And when asked about fundamentals versus charts, he's candid: he holds that fundamentals are the core, yet several people in one of his group chats — neighbours in some of Taipei's most expensive buildings — don't use fundamentals at all. Volume, momentum, the picture. He tried borrowing a little and found it worked nicely, but he didn't switch, because his skill points went into fundamentals from the start. That's more honest than any which-school-is-right argument: more than one road pays, and the one available to you has a lot to do with where your hours already went. ## Further reading - Gooaye EP690 (2026-08-22), available on the usual podcast platforms - For the turnkey-versus-COT distinction, earnings call transcripts of chip design service firms are a good starting point — management gets asked about this shift directly - To quantify your own "buys well, sells badly" problem: export two years of trades, compute the return N days after entry versus the return you actually realised, and the gap is the price of your exit habits ## 帶得走的一件事 **One idea only: turn the decisions you make over and over into a rule you can write down in advance.** Buying well and selling badly, always wanting to wait a bit longer, always knowing better afterwards — none of these are failures of intelligence. They are one decision being forced to happen again at the worst possible moment: position on, emotions up, numbers moving. The oil seller's skill was never in that one accurate pour. It was that after several thousand of them, the pour no longer required a decision. **Something to try today, whether or not you invest:** Pick one thing you've hesitated over at least three times this month and re-argued each time — when to put the phone down, whether to accept a last-minute invitation, whether to buy the discounted thing, whether to reply to the message that unsettled you. Just one. Write a rule with an explicit trigger, in the form "when ⟨specific situation⟩ happens, I ⟨specific action⟩." Not "use my phone less" — rather, "when the clock hits eleven, the phone goes in the drawer." Then run it for seven days, unmodified, no exceptions, and on day seven answer one question: **how many of those days took willpower, and how many happened without my noticing?** That second number is what your rule is actually worth.