Every Number Right, the Price Still Wrong: When the Market Starts Pricing Three Years of Capex

Seen from the side of change, heaven and earth cannot hold still for even an instant; seen from the side of the unchanging, all things and I alike are without end. — Su Shi, “First Ode on the Red Cliff”

Look at it one way and nothing lasts a moment; look at it the other way and nothing ever ends. Same river — the difference is the timescale you’re standing on. This episode asks whether the fundamentals have peaked, and that question is exactly this one: a near-four-percent drop belongs to the side of change, while node progression and demand visibility running out to 2029–2030 belong to the side of the unchanging. Both can be true at once. The conflict only shows up in your holding period.

tldr: Every line of the results beat expectations and the stock fell anyway — because the market has stopped applauding this quarter’s earnings and started pricing three years of capital spending and a permanently diluted overseas margin structure.

What This Episode Covers

The recording took place the morning after the earnings call. The host opened by asking the researcher whether they’d slept well, and got back that they’d finished writing the client note and gone straight to bed, exhausted. That small aside sets the tone for the whole hour: this isn’t a disaster, it’s fatigue.

A major upstream equipment supplier had reported strong numbers the day before. Then came the main event, where every figure came in ahead of expectations — and the US-listed shares fell nearly four percent overnight, dragging Taiwan equities and the Philadelphia Semiconductor Index down with them the next day. That gap is the episode’s subject. If every number is right, what exactly is the market selling? Is it simply that the stock ran too far, or have the best days of the semiconductor cycle genuinely passed?

The first half unpacks the results and guidance: the profit structure, node mix, a number that looks like a typo, the story behind the capex increase, and inventory days. The middle section looks outward from one company’s results to the wider industry — memory, custom silicon, analogue chips, consumer electronics, and where each sits right now. The final section takes listener questions: leverage in Korean equities, whether Taiwan’s export growth still works as a leading indicator, and whether to move money from equities into property.

The strongest impression afterwards: nearly every number that “got worse” in this episode admits two completely opposite readings, and the episode spends real effort teaching you how to tell them apart — which is worth more than any of its conclusions.

Key Takeaways

1. The results beat across the board; the market sold the capex

Revenue grew 36% year on year, gross margin hit a record 67%, and earnings per share came in at NT$27. Guidance was more aggressive still: full-year revenue growth was raised from above 30% to above 40% — an unusual size of revision for a company already this large. The episode noted that management used the word “stronger” three times while describing demand.

What the market seized on was the other half of the same deck. Capex, already raised once to US$52–56bn, was raised again to US$60–64bn, plus another US$100bn earmarked for Arizona and four more fabs. The company’s own prior guidance on overseas dilution runs to two or three percentage points of gross margin early on and three to four later — water, power, land, and labour all cost more than at home. In other words, the market wasn’t rejecting the quarter. It was repricing the depreciation schedule and margin structure of the next three years.

2. A virtue that has become an accusation

The episode makes a point worth writing down: capex increases used to be read as management signalling confidence in the cycle — good news. The same action is now read as a warning of over-investment. The fundamentals didn’t change; the interpretation did. That is a classic symptom of capital crowding into a single theme, where anything at all gets amplified into a negative.

Which is why, when the episode revisits last March’s US$100bn announcement and the roughly 25% drawdown that followed, it adds a caveat: you can’t put all of that on capex, because the tariff shock arrived right behind it. That caveat matters more than the anecdote. Saying “the price fell because of this news” too fluently is the mistake investors make most often and detect in themselves least.

3. The number that looks like a typo is actually a strategy signal

Pre-tax margin came in above gross margin this quarter. Intuitively impossible, since gross margin sits above operating expenses. The answer is non-operating: the company cut its stake in a mature-node affiliate from 27% to 19%. Below twenty percent, accounting treatment says you no longer exert substantive influence.

The episode’s read: that affiliate is mature enough in legacy nodes and eight-inch operations to run itself, so direct governance involvement is no longer necessary. The more important message is where the resources went — consolidating attention on leading-edge process and advanced packaging. A figure that looks at first like a reporting error is really a statement about what to hold tight and what to let go.

4. Two-nanometre contributed revenue in two quarters, and the lead act changed long ago

Three-nanometre reached 30% of revenue, another record. Two-nanometre entered volume production only in the fourth quarter of last year and delivered visible revenue contribution within two quarters — a quarter faster than the previous generation. The episode ties this to a new-generation accelerator platform entering production in the second half, consistent with earlier commentary that two-nanometre tape-outs in the first two years would exceed the equivalent period for three-nanometre.

The structural shift is blunter: high-performance computing is now 66% of revenue, a record high, while the once-dominant handset business has fallen to 22%, a record low. An analyst asked about customer concentration risk; the answer was that this is in fact safer. That answer is worth keeping on file to grade later.

5. Seventy-six days of inventory: stockpiling or unsold?

Inventory days climbed again to 76 — the point analysts pressed on and the market worries about. If demand rolls over while you’re actively restocking, aren’t you setting up the next destocking cycle?

The episode says not yet, and the reason isn’t “management said so” — it’s a checkable property. This is strategic material build ahead of a fast two-nanometre ramp, not finished goods sitting unsold; and there’s a queue of next-generation accelerators and in-house silicon still coming, with no sign of production being slowed. The same number, two opposite stories. What settles it isn’t the quantity of inventory but its nature and which link of the chain it sits in.

6. The adjective quietly dropped from the consumer electronics outlook

This is the most delicate observation in the episode. Last quarter, management said high-end demand still showed resilience. This quarter the word “high-end” didn’t appear at all — only that component price increases are pressuring price-sensitive consumer electronics, and that they are staying cautious.

The data backs the caution: handset shipments have fallen for two straight quarters, PCs began declining in the second. Yet premium brands are still growing, so “the high end is resilient” hasn’t actually been falsified — what changed was the conviction in the delivery. The host mentioned in passing that the researcher’s newly bought laptop had gone up in price too. Memory inflation working its way down to the shelf is easier to feel in that one line than in any long-run chart. Tone tends to move before numbers do, and tone is the part you only get by listening to the call.

7. The world outside: demand is still spilling over, but competition and sentiment are both hotter

Looking down the chain, tightness hasn’t materially eased. The big memory maker blew past estimates, the bellwether in custom accelerators beat on both results and guidance, and even the analogue supplier previously dragged down by automotive and industrial demand came in ahead, on strong data-centre demand for power management chips. Demand is spilling into every component category.

On competition, management said plainly that there are two competitors, one in Korea and one in the US. On the American one, the episode’s read is that leading-edge production has started but yield still lacks a stable answer — and, more decisively, that trust is the real barrier: will a design house hand its most critical designs to a foundry that also sells competing products? The extra US$100bn is itself the counter-evidence on order loss: customers asked for more capacity, which is why the money is being spent. As for whether the “silicon shield” is being hollowed out, the episode does a capacity count — roughly ten leading-edge and advanced-packaging fabs already in production at home with thirteen more planned, against four plus four in the US. The gap speaks for itself, and the core technology stays put.

The listener questions carry the same shape. Korean equities have no fundamental problem — the driver is the memory cycle, and a new fab takes eighteen months to two years from groundbreaking to clean output, so the gap can’t be closed quickly. The volatility comes from concentrated leverage; the local regulator suspended new single-stock leveraged product listings in mid-July. Fundamentals and price volatility are two separate things to process.

Going Further

1. A price target is two assumptions multiplied — which makes it far less stable than it looks

The episode closes with a very common calculation: mark the multiple down a little, multiply by next year’s earnings estimate, get a number. I’m not repeating that number, because the number isn’t the point. The structure is.

Target = multiple × earnings estimate. Both are estimates, and in bad news they move the same way: when the market compresses the multiple it will pay, analysts are usually cutting earnings at the same time. So the phrase “I already used a conservative multiple and there’s still upside” carries less margin of safety than it feels like — you were conservative on one variable while the other stayed in the optimistic case.

The more honest move is to invert it: what earnings path and what multiple does today’s price already assume, and do I disagree, and why? That ordering translates “it looks cheap” into a sentence that can be argued with — “the market is assuming A, and I think A is too optimistic because of B.” The first only persuades you. The second can be graded later.

The companion step is writing the falsifying condition down first: what evidence would make me change the multiple. Write it first so that price doesn’t do it for you. Cutting your multiple after the stock falls isn’t valuation, it’s ratification.

2. When a company trades margin for location, the scarce thing has already changed

Put two pieces of the episode side by side and they converge. First, the long-run margin dilution from overseas fabs, plus the earlier commitment that roughly thirty percent of sub-two-nanometre capacity would sit in Arizona. Second, memory: eighteen months to two years from groundbreaking to output, with bottlenecks in the expansion itself, so no new capacity before the second half of next year at the earliest.

The first says something specific: a company with genuine pricing power at the leading edge — one that explicitly chooses not to price to the maximum — is willing to accept a permanently lower slice of gross margin in exchange for where its capacity physically sits. Paying that price means the scarce input is no longer “can it be built,” which is precisely its strongest suit, but “where can it be built, and will the power, the land, and the politics allow it.” The second gives the same shape from a different industry: everyone knows to build, everyone is building, and time itself is the wall.

For stock selection this changes the question from “whose product is best” to “which layer breaks first if demand doubles again.” The first compares competitiveness; the second is where pricing power comes from.

The counterweight matters, or the inference over-extends fast. “This layer will break” and “companies in this layer will all profit” are different claims, with customer concentration, contract structure, and second-source availability sitting in between. And bottleneck rents come with an expiry date — they compress as new capacity arrives. “Second half of next year at the earliest” is that expiry date being said out loud.

3. Split “growth is slowing” into base effect, demand, and your holding period

A listener asked a genuinely valuable question: does the old rule that Taiwan’s export growth peaks about six months ahead of equities still hold? The answer is the most portable tool in the episode — when growth decelerates, first separate base effect from demand. Against this year’s exceptionally high base, next year’s numbers are almost guaranteed to look poor, but that’s arithmetic in the denominator, not a turn in the cycle. The leading indicator only means something when the deceleration comes from demand.

That test maps straight back onto every deteriorating number here: inventory days rising — stockpiling or unsold? Consumer electronics declining — structural downgrade or price-sensitive deferral? The share price falling — multiple compression or earnings cuts? None can be answered by feel, and all three have public, scheduled, gradable things to watch: the direction and nature of inventory days, whether end brands can pass price increases through, and when new memory capacity actually arrives. A risk without something to watch becomes emotion; a risk with something to watch becomes discipline.

The last layer is holding period. The episode warns anyone running large leveraged positions that volatility is rising, and offers the line “make investments you can sleep through.” Its force isn’t moral — it names the real cost of leverage: the problem was never whether your direction is right, it’s that leverage takes away the time you need to wait for the answer. When capital is concentrated in one theme, even good news gets sold, and at that point your holding period is the only thing you still hold.

Back to Su Shi: from the side of change nothing lasts an instant; from the side of the unchanging, nothing ends. Both readings are correct. What decides which one you see is how long you intend to hold — and that is one of the few variables entirely within your control.

Sources Worth Checking

• The episode itself: MacroMicro After Meeting Podcast EP.207 (2026-07-19)
• For company-level financials, guidance, and capex plans, use the investor relations site’s filings and earnings decks directly — don’t make decisions on second-hand numbers
• Taiwan export and trade statistics: Ministry of Finance customs and trade statistics
• Taiwan property lending share and credit control measures: Central Bank of the Republic of China (Taiwan) board meeting statements and monthly financial statistics
• Korean regulatory action on leveraged products: official announcements from the Financial Services Commission
• US inflation data: BLS CPI report

Disclaimer

This is a personal set of notes and reflections on a podcast episode, written as educational content. It is not investment advice and does not constitute an offer, solicitation, or recommendation to buy or sell any financial instrument. Companies, industries, and figures mentioned are used only to illustrate a framework; they do not represent a view or rating on any security, and no price targets are given. Verify all data against original sources and official company disclosures; market conditions change constantly and these views may cease to apply shortly after publication. Investing carries risk, and leveraged products carry considerably more. Make your own independent judgement based on your financial situation, risk tolerance, and objectives, and consult a qualified professional where appropriate. The author accepts no responsibility for decisions made on the basis of this article.

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.