Bottlenecks Don't Disappear, They Relocate — Notes on Gooaye EP692

Prime years never come twice, nor does a morning return to the same day. Take heart while you can — the years wait for no one.

— Tao Yuanming, Miscellaneous Poems, No. 1 (Eastern Jin, c. 5th century; translated by the author)

What This Episode Is About

Gooaye EP692 (2026-08-29) carries a title that is nothing but a beer emoji. The first half has almost no market content at all. It is a father travelling through Hokkaido with two small children: hunting for restaurants with private rooms so the kids won’t disturb anyone, watching his son abandon a card game the moment he had mastered it, arguing with his wife over refusing to pay ten thousand for a bottle that should cost two or three.

The middle turns to markets — the Anthropic–Salesforce partnership, Marvell’s guidance raise, and where the supply chain is actually stuck after NVIDIA’s earnings call. The final stretch answers listener questions, including a semiconductor engineer weighing a job offer at double the pay, and an only child asking what Taiwan’s elder-care system is going to look like.

The interesting part is that the first half and the middle are making the same argument. The key line in the travel section comes from the driver: sometimes you pass through this village and there is no second inn. The key line in the market section is that if you route around the memory shortage, the shortage simply moves onto the substrate. Both say the same thing: a constraint is never eliminated. It only changes address.

The Main Points

1. The “software apocalypse” was denied by the people it was supposed to kill. Anthropic and Salesforce announced a partnership: dozens of Salesforce skills usable directly inside Anthropic’s interface, and Anthropic’s models available as the default option inside Salesforce. Both CEOs went on the same finance programme and said plainly that there is no software apocalypse — this looks more like a positive-sum game. The host adds the sharper observation: strictly speaking, the megacaps are software companies too. Are they all supposed to die? What he has been arguing against for a year is the clean split — software here, hardware there, one pile lives, the other dies. What is actually happening is a redistribution of where value accrues, not an extinction event.

2. Salesforce’s move matters more than its product reputation. The host notes the product itself gets mediocre reviews and plenty of complaints, but once the founder re-engaged, the company turned remarkably fast: it gave up the interface layer immediately and pushed everything into the application and action layer, because that is where its accumulated process knowledge and enterprise data sit. “Whoever has the data has the gold” has been said for a decade; the difference now is that there is finally a way to use it. Commercially, the partnership is two separate bills — you pay each side for what you consume through it. Cordial on the surface, and underneath a fight over who holds pricing power and who becomes the smaller half of the bundle.

3. Optical interconnect is the engine of this rally. Marvell raised guidance. The host reads the largest contributor as custom silicon for a hyperscaler starting to ship in volume, but a big part of the raise was optical — the CEO named Scale Up optical interconnect switching silicon as a major growth driver ahead. This group corrected for a long stretch before waking up, and it is running from raw materials all the way through packaging.

4. NVIDIA’s real moat may not be the chip. An analysis firm benchmarked a purpose-built inference chip and found it beat the current flagship on tokens per watt, and the doubts about NVIDIA started up again. The host’s response: the test is fair, but that chip was designed for inference and would still fall short on brute-force training. More importantly, in both models and silicon, someone announces the strongest thing every couple of months until you genuinely cannot remember who is ahead. What he actually cares about is a question he raised on the show months ago — why was NVIDIA buying optical and laser components in quantities far beyond what its own products required? His answer then was that it is squeezing competitors out of the supply. To ship a server you need optical components, switches, substrates and assembly capacity, and every one of those is short right now. A better benchmark does not ship if you cannot buy the capacity. Seen this way, all those dinners in Taiwan look less like a taste for the local food and more like a very deep position being built.

5. Shortages relocate; they do not resolve. The market is discussing dropping memory stacks from twelve layers to eight, possibly four, on cost and supply grounds. But the capacity the KV cache needs does not shrink because of that — making it up requires more substrate area, and substrate is already one of the tightest links. So the change moves the gap from A to B. The host contrasts two temperaments here: Musk’s position is that AI is growing at 200% a year while memory output grows at tens of percent, so of course prices rise, and he’ll pay. Most executives don’t have that latitude — they answer to shareholders, so they go looking for the route that makes the financials look better. Every one of those detours creates a new shortage, and a new opportunity, somewhere else.

6. The toy paradox: his son mastered the card game and then quit. A three-year-old who could identify every creature by its sound and had the type matchups down cold lost interest completely the moment nothing was left to learn — and moved to a new franchise where the pleasure had shifted to pulling rare cards and throwing the common ones away. The father’s alarm: what the child loves is drifting from the toy to the act of buying. The previous generation of parents didn’t have the money, so kids turned a twig and a battery into wrestlers for an afternoon. This generation has the opposite problem and may be raising the dopamine threshold early. His conclusion is that “I’m doing this for your own good” is itself the wrong thought — every generation of parents has something to overcome, and abundance is this one’s.

7. You will always want more money — so keep a record you can check yourself against. He recommends the phone’s built-in journal app: a photo, a few lines, a map pin. What he writes in it includes a morning where he traded badly, decided he was an idiot, and considered quitting stocks for the year — followed by the next day, when the position gapped up and he suddenly felt he’d done nothing wrong. People forget, and they sugar-coat what they remember, and the drift shows up in your thinking. This is the trading journal in a form you will actually keep.

Going Deeper

”The news says my industry is about to be replaced by AI. Should I sell?”

The problem is the frame, not the position. “X is about to be replaced” is popular because it sorts the world into two piles and all you have to remember is which pile you’re in. But that is exactly the split the principals themselves rejected in this episode. What moves is where value concentrates, not which category disappears wholesale.

So replace the question with: which layer does this company occupy, and is that layer replaceable? Salesforce gives an unusually operational test — it handed over the interface layer voluntarily. The interface is the easiest layer for a newcomer to take with a better experience, so it took that loss and pushed resources into what it genuinely cannot be displaced from: years of accumulated process knowledge, and the customer’s decision about who gets to hold the data. Which yields a second test: watch which way data ownership tips. Plenty of enterprises hesitate to hand data to a model vendor; some insist it stays in their own jurisdiction; and the models are currently trained mostly on public data. Where the closed enterprise data ends up will decide where the value lands. The host’s own guess is that the model side captures more — and he says openly that he may be wrong.

The honest version of this exercise: instead of “will it die,” answer “if I’m wrong, which layer am I wrong about.” If you can answer that, you have a view.

”It won the benchmark. Should I switch?”

This episode draws a distinction that rarely gets stated: product capability and the ability to deliver are two different things, and the binding constraint right now is the second one.

In a normal year, the better product wins, because capacity can be bought. The environment described here is one where every link is short — the host won’t even name the tightest one, because naming it moves the stock. In that environment, control of resources decides who ships, more than any technical metric. Buying enormous quantities, buying regardless of price, locking the chain end to end: that is an advantage of scale, not of engineering, and it is hard to replicate quickly, because it takes years of supply-chain experience. Slip once and your schedule stretches; by the time you ship, the incumbent is on the next generation.

From this grows a habit for reading good news: whenever you hear that some bottleneck has been routed around, ask where the gap went. Dropping memory layers looks like solving memory; in practice it hands the pressure to substrate. Announcements usually describe only the half that was removed, never the half it was moved onto — and the half it moved onto is often where the real scarcity and the real pricing power now live.

The same frame works in reverse. When a group wakes up after a long correction, the question isn’t “has it run,” it’s “who moved the constraint onto it.” Optical holds up here because every additional server shipped requires more of it. That is structural, not sentiment.

”Should I take the job that pays double?”

A listener with seven years at a foreign semiconductor firm asks: the company is highly profitable but shares none of it, an old colleague is offering more than double, and the cost is that he’d barely see his kids on weekdays. He adds that finances are fine, but he’d like to keep growing.

The answer takes a detour before landing: if you already have spare capital to invest, what you want is no longer money — it’s more, and more has no terminal value. No matter how much you have, you will want more; very few people know when to stop. What actually sits behind financial freedom is a falling appetite for consumption, and it has little to do with income, because a modern economy will always find a way to absorb whatever you earn. He quotes a venture investor on the irony: none of the genuinely wealthy people he knows ever held the “once I hit the number I’ll retire” attitude — the people who talk that way tend not to reach the number. Which makes the whole thing something of a curse.

The useful part comes next, and it works outside investing entirely: as your income shifts from labour to assets, what you should be buying is freedom; if you aren’t getting it, something is wrong. Plenty of people watch their assets grow while their discretionary time shrinks. That is not the market’s doing — it is an objective function written wrong, and the error hides well because it looks exactly like diligence.

The sharpest moment in the episode isn’t his, though. It belongs to the driver in Hokkaido, who has a twelve-year-old daughter who no longer talks to him — she grew up local and her world is her friends now. So when the two boys got loud in the car and the host started apologising, the driver said: let him talk; sometimes you pass through this village and there is no second inn. The host’s follow-up is honest — everyone tells you this, and in the moment you’re still irritated, and only at night, looking at their sleeping faces, do you think you should have let them talk more. His advice to the listener follows from that: if you can already clearly identify that what you want is the children, take the children first. Once they’re older and no longer talking to you, there will still be time to go make money.

One more thing worth keeping. He mentions that several commentators he used to recommend have since dropped off his list — some he now finds attention-seeking, one he simply thinks isn’t sharp. And then he turns it on himself: the world is dynamic; you like this show now and one day you may not, and that is normal — as you grow you’ll find what actually suits you, and you’ll get better at spotting who is genuinely strong. That is a healthy thing to hear from anyone you follow.

Worth Looking At

• Gooaye EP692 (2026-08-29), the source for everything above — the listener-question segment in particular rewards listening in full
• The latest earnings call transcripts from Marvell and NVIDIA (free on their investor relations sites), where the guidance raise and the shortage commentary can be read in the original
• A journal app on your phone — the show’s argument for the built-in one is simply that the barrier is low enough that you’ll actually use it
• Tao Yuanming’s Miscellaneous Poems, public domain and freely available in full

One Thing to Take With You

One idea: bottlenecks don’t disappear, they relocate.

The market segment makes this unusually clean. Memory is expensive and scarce, so use less memory — but the capacity requirement doesn’t vanish, so the pressure lands on substrate, which was already the tightest link. You never solved the problem. You changed its address.

The rule holds well outside markets, and there it gets ignored more often. You decide money is the constraint, so you take the job that pays double, and the constraint becomes time. You decide time is the constraint, so you outsource every errand, and the constraint becomes attention. You decide a noisy child is the constraint, so you hand over the iPad, and the constraint becomes that he stops needing you there. Every time you relieve a constraint you are placing a bet that you can live with the next one — and almost nobody writes that bet down.

A practice you can do today: write down the one constraint you most want gone this week, then add one line underneath — “once it’s gone, what will be stopping me next?” — and date it.

One line. Not a plan. The second line is the whole point: it forces you to make the world-after-the-fix concrete, once. Come back in a week and you’ll get one of three outcomes. You were right, which means you’re clearer-eyed than you assumed. You were wrong, which usually means the thing actually blocking you sits closer to you than you wanted to admit. Or you find the constraint no longer matters — in which case it never deserved the attention.

This beats “think about it more” for one reason: it leaves behind something you can check yourself against. Which is exactly why the episode recommends keeping a journal. People forget, and what they remember comes sugar-coated. What you need isn’t better judgement — it’s a note from the person you were a week ago, in their own handwriting, that you can no longer edit.

Raw Markdown

This article is an educational discussion of investment method. It is not advice to buy or sell any individual security, offers no target prices, and does not analyze any current holding. Investing carries risk; make your own decisions or consult a qualified professional.