# Gooaye EP678 Notes: Meta Just Put Compute Pricing in the Sunlight — the Fog Behind Dream Multiples Is Lifting

> Personal notes on Gooaye EP678: Meta repurposing old GPUs for inference and renting them out — refuting the 'AI server bad-debt' short thesis while making compute pricing transparent, which pressures dream-multiple valuations; rotation inside optical communications; the broad component price-hike wave. Educational notes, not investment advice; disclaimer at the end.

Published: 2026-07-13
Locale: en
Tags: podcast-notes, market-view, ai-infra, valuation, education

TL;DR: This episode: Meta is repurposing older Hopper-class GPUs for inference and renting compute out — proving AI servers aren't write-offs (contra Michael Burry-style depreciation shorts), while simultaneously making compute pricing transparent, which threatens the dream multiples of compute-rental names; money is rotating inside optical communications from 'narrow and fast' to 'wide and slow'; memory keeps rising at a slowing pace while power and passive components enter a broad price-hike cycle. My takeaway: transparency is fundamentals' friend and the dream multiple's enemy — when the fog lifts, you find out whether you bought cash flow or a dream.


![Realist-magical oil painting cover: on a mountain trail at dawn, silver mist parts like a curtain while a lone traveler stands on a stone outcrop as the mountain's true face emerges in morning light](/covers/gooaye-ep678-cover.png)

> *Seen sideways it's a ridge, face-on a peak; near, far, high, low — never twice the same. Why can't I see Mount Lu's true face? Because I'm standing inside the mountain.*<br/>
> — Su Shi, "Written on the Wall of Xilin Temple" (Northern Song, 1084; my translation)

> These are my **personal notes** on Gooaye **EP678** (aired 2026-07-11) — not a transcript, not official content. For the full episode, please support the original show. What follows is what the episode sparked for me, plus my own synthesis.

## The episode in one sentence

Meta answered two questions with one move: will AI servers become bad debt (no — old GPUs are earning money doing inference and being rented out), and can the compute business stay conveniently opaque (no longer — when a player this size starts renting out capacity, pricing becomes transparent). The same headline is bullish for fundamentals and bearish for dream multiples — and the episode's real value is laying both sides on the table.

## My highlights

- **Meta monetizing old compute refutes the "AI server bad-debt" thesis**: repurposing older Hopper-class GPUs for inference or external rental shows these machines are income-producing assets, not depreciation landmines — a direct answer to Michael Burry-style accounting shorts. The market responded with a gap up.
- **But pricing transparency is the reality test for dream multiples**: with a player of Meta's scale entering compute rental, prices stabilize and returns become calculable. AI-infrastructure and compute-rental names that enjoyed rich valuations precisely because things were "hard to see clearly" now face convergence.
- **Rotation inside optical communications**: money is moving from the "narrow and fast" favorites (silicon photonics, InP lasers) toward "wide and slow" stories (VCSEL, even further-out Micro LED).
- **Memory: still rising, at a slowing pace**: next quarter's contract prices are expected to keep climbing with narrower increments — while the hyperscalers' buying resolve shows no sign of pulling back.
- **A broad price-hike cycle in power and passive components is confirmed**: Taiwanese MOSFET and back-end packaging players, running at full utilization, are seeing sharper profit and margin jumps than their US peers; leadframes are in severe shortage with stretched lead times.
- **The copper-interconnect names once punished by the "optics replaces copper" narrative** have, per the episode, largely recovered. I checked the prices: they've recovered much of the drawdown but still sit meaningfully below prior highs — "recovered" is accurate, "new highs" would not be. Which is exactly why you verify numbers yourself.

## Where my thinking goes from here

What struck me most is the **two-sidedness of a single headline**.

Meta renting out compute is bullish on the asset side: five-year-old machines still earn. But it's bearish on the valuation side, because **a dream multiple is fundamentally a premium on opacity**. While returns can't be computed, the market prices with imagination; once Meta publishes the rate card, the market swaps imagination for a calculator. Holders of compute-concept names aren't about to face a deteriorating industry — they're about to face a story that can be audited. On a chart those two look identical (multiple compression), but the causes and the correct responses are entirely different.

Second, "narrow and fast" rotating into "wide and slow." My standing rule for intra-sector rotation: **rotation is an outcome, not a reason.** Chasing "the money is coming" means buying other people's expectations rather than the company. I wrote the same sentence in my last episode notes; this episode merely re-validates it with a different sector.

Third — the most tangible one — **the price-hike wave**. Memory, passives, power components, leadframes: price increases, shortages, stretched lead times. These flow straight into income statements and need no narrative assistance. The episode notes that near-term money prefers names where the upside shows up immediately in revenue and earnings. Agreed — and it matches my own habit: **between a story and a shortage, I look at the shortage first.** A production line running full with customers queuing beats a hundred-page vision deck.

## Where my head is these days

Honestly, I've enjoyed "hard to see clearly" markets too — some of my positions rose exactly when the market was at its most imaginative. Nobody rushed to compute returns back then, myself included.

Over the past half year I've been learning to treat clarity as a friend. I recently ran a rather brutal audit of my own records — a hundred-plus days of every command I ever typed — and the most painful finding was that I had spent sixty times more effort building verification systems than actually using them on myself. In other words, I too once preferred not to look at my own positions too clearly. That's why this Meta headline lands hard for me: whether the lifting fog is a threat or a gift depends on whether what you hold survives arithmetic. **Those who bought cash flow welcome transparency; those who bought dreams fear it.**

I want to be permanently in the first group. And on days I'm not, I at least want to know — honestly — which side I'm standing on.

**Disclosure**: I hold positions related to some sectors mentioned here (optical communications, AI infrastructure). That's precisely why I make myself write down both sides.

## References

- **Gooaye Podcast (the original show)**: EP678 — please listen on official podcast platforms and support the creator.
- One exercise: next time you see a "giant enters the market" headline, write two separate lines — "what this says about asset quality" and "what this says about valuation logic." They often point in opposite directions, and the market usually prices one before the other.

---

*These are personal, educational reflections after listening to a podcast — not official Gooaye content, not a recommendation to buy or sell any security, no price targets, not directed at any current holding of yours. Companies appear only to explain the episode's content or concepts; I hold positions related to some sectors mentioned. Investing involves risk; do your own research or consult a qualified professional.*