Trump's Midterm, TSMC's Final Exam, and Who Got Flunked by Inflation | Notes on MacroMicro After Meeting EP.218

Contents

1. The case for: this episode makes “the fundamentals haven’t broken” as strong as it gets
2. The case against: turn the same evidence over
3. My read: the two sides differ on three conditions you can check
4. Worth a look
5. The one thing to take with you

In her right hand the gleanings, on her left arm a broken basket. I listened to them talking, and the words were hard to bear: the tax took the whole harvest, so she gathers this to fill her hunger.

—— Bai Juyi, “Watching the Wheat Harvest” (Tang dynasty, c. 806; translation mine)

In episode 218 of MacroMicro’s After Meeting, recorded on 16 October 2026, host Roger sat down with US analyst Ralice and Taiwan analyst Jat to put two report cards from the same week side by side. TSMC posted third-quarter revenue of NT$1.5 trillion, up 17% from the prior quarter and 50% from a year earlier, and raised its full-year revenue growth outlook to 44% — the third upgrade this year. That same week US headline CPI rebounded to 3.7% year over year and core CPI rose 0.3% month over month, beating expectations for the second consecutive time. Ralice’s read is that an October hike is not a done deal, with market-implied odds already back below 30% from above 50% — but her read rests on one condition: shelter has to cool off next month. Jat’s read is that the AI supply chain is still bottlenecked on shortage rather than glut, conditional on new capacity not arriving until the second half of 2027. Neither conclusion is “so we’re fine.” Both are “so watch the next print.”

The case for: this episode makes “the fundamentals haven’t broken” as strong as it gets

My first reaction was how hard TSMC’s numbers make the “AI demand is still here” argument. Jat flagged a detail that gets skipped: those margins were delivered while absorbing the early-ramp cost of 2nm and the dilution from overseas fabs, and both gross margin and operating margin still landed inside guidance. Leading-edge nodes and advanced packaging — stacking multiple dies into one package — are short enough that TSMC can price its way through its own cost pressure. When C.C. Wei said in English that there is no ceiling on TSMC’s profitability, the weight of the line comes from the quarter he chose to say it in.

Capex wasn’t revised up this time, and Jat isn’t bothered: it was already raised sharply last quarter, and the CFO said capex over the next three years will run meaningfully above the past three. For the equipment and component names downstream, the word “three years” matters more than any single quarter. TSMC also kept its long-term AI accelerator growth outlook at 55% to 59% — and that’s the narrow definition, excluding CPUs.

On inflation, Ralice’s optimism comes from opening up the components. More than half of this print’s increase came from gasoline. And the items that drove the two consecutive core upside surprises were different ones: last time a spike in wireless telecom services, this time shelter, used cars, recreation, medical care, airfares. Her inference is rotation rather than one line item compounding every month, and she adds that since the September meeting, officials including Waller, Bowman and Williams have publicly said there’s no urgency. The market agrees: odds for October fell from above 50% on the night of the release to under 30%. The bond market was the odd one out in an interesting way — a bad inflation print, yet the US 10-year yield pulled back from a 5.3% high to 5.1% or slightly below, and Japan went back under 3%.

Inventory sits on this side too. TSMC’s days of inventory climbed for two quarters to prepare for the 2nm ramp, then fell as shipments began in Q3. ASML upstream looks even cleaner: revenue growth jumped more than 50%, all three margin lines rose, full-year guidance went up, and days of inventory made a new low. Further downstream, rental price indices for H100 and B200 are still rising, which is Jat’s basis for saying the economics of compute still work — you can calculate the revenue each gigawatt of capacity produces. As for where demand comes from, his structure is agents: AI that finishes tasks on its own rather than just chatting. Long-running tasks store far more context, which pulls HBM and high-end memory; one person running several agents needs CPUs to schedule them and decide which tool gets called; passing work between them fast means high-speed networking chips.

The case against: turn the same evidence over

The episode offers its own test: you need three prints to call a trend. We’re at two. Listening to it, I thought “we haven’t hit the third one yet” was doing the work of reassurance, when it’s also just another way of writing “the alarm has gone off twice.” And core CPI at 0.3% month over month already sits above the 0.2% they themselves treat as acceptable. That isn’t one stray data point.

The rotation argument has a specific weakness this time: shelter is among the items that spiked, and shelter carries the largest weight. Ralice says so herself — whether it cools next month is the thing to watch. So the support for optimism is “what rose this month will fade next month,” and what rose happens to be the stickiest line in the index. Add oil: Brent is still above $100 a barrel and WTI is pushing back toward $100, while Trump says the war ends once the election is over. He also said, before taking office, that he’d end the war and end inflation. I’m not willing to price lower oil off that record.

On valuation, the tape already spoke for this side. With a report card like that, the Philadelphia Semiconductor Index fell seven sessions in a row and more than 10% from its high; Taiwan’s index, which had been closing in on 50,000, briefly traded under 47,000; Korea is down more than 20% from its early-October high. What the market priced this week was the discount rate, not TSMC’s sequential growth. Roger’s line about scoring 100 and being asked for 150 is about sentiment, but written as arithmetic it says: however good the numerator, move the denominator and the answer still looks bad.

The AI doubts aren’t hollow either. Anthropic’s disclosed revenue growth slowed from roughly 100% to 46% quarter over quarter, which Jat attributes to competition — OpenAI’s Codex has gained enough capability to take enterprise share, with Meta’s MUSE and Musk’s Grok bot both breaking out. The three of them are their own evidence: Jat switches between tools, and the thing in the corner of his screen that Roger mistook for a cartoon character turns out to be a Codex pad, while Ralice prefers ChatGPT. A larger total market with more players explains aggregate capex just fine. It doesn’t explain returns at any single company: competition pressures price and raises burn at the same time, which lengthens payback.

The sharpest objection is CoreWeave’s corporate bond yield spiking to 15%. Jat’s counter is that GPU rental prices are still rising, but those two facts answer different questions. Rental prices say “this batch of cards pencils out today.” A 15% yield says “the next refinancing may not be available, or may be far more expensive.” These neo-clouds — operators who borrow to buy GPUs and resell compute — scale through revolving financing, and the bond market is pricing whether that loop breaks, not this quarter’s revenue. Jat’s own pass condition is explicit: revenue growth has to outrun GPU depreciation plus debt interest. That line is driven directly by rates, and rates are exactly what got worse this week.

Politics lands on this side as well. The 3 November midterms put all 435 House seats and 35 Senate seats on the ballot, and Polymarket odds have Democrats favored in both chambers, with the Senate above 60%. History doesn’t help the incumbent party either: from 1934 to 2022 the president’s party lost an average of 27 House seats and 3 Senate seats. The only two modern exceptions in the House were 1998 and 2002 — one driven by backlash to impeachment, the other by post-9/11 solidarity — and neither backdrop exists now. If divided government arrives, Ralice names two pressure points: appointments, where the seat opening when Powell’s term ends in 2028 becomes much harder to fill with a loyalist, and the debt ceiling. US debt is past $40 trillion against a $41.1 trillion limit, with the binding date widely estimated between February and July next year — a stretch of time that generates its own volatility in a bond market where term premium is already elevated. TSMC’s earnings call reaches none of this.

The last piece of evidence for the bears came from the show itself. Answering a listener question on margin maintenance ratios, Jat noted Taiwan’s ratio still sits on the high side, meaning investors remain optimistic — which is part of why this pullback is shaking out the weaker hands. Optimistic positioning has never been support. It’s supply that hasn’t cleared yet.

My read: the two sides differ on three conditions you can check

I lean toward the bulls, but on narrow ground — narrow enough to write down as three things verifiable within a month.

First, a question that used to trap me: why does the stock fall on good news? My old approach was to hunt for the bad news I must have missed, and when I couldn’t find it, to doubt my ability to read a filing. Now I split it into two columns: earnings move the numerator (how much the company makes), rates move the denominator (what those earnings are worth today). This week the numerator hit a record and the denominator jumped too, so the direction depends on which moved faster. TSMC posting great numbers while the stock ignores them stops being a contradiction, and I save the hours I used to spend looking for a ghost story.

Second: the news sounds great, can I trust it? My filter is to trust claims that come with a failure condition. Ralice attached conditions — shelter cools next month, core CPI doesn’t make it three in a row. Jat attached conditions — no new capacity until the second half of 2027, neo-cloud revenue growth outrunning depreciation plus interest. A conditional view can be scored later; an unconditional one is only good for warmth. So I compress the episode into three watch items: the next shelter print, whether core CPI completes a third surprise, and where neo-cloud refinancing costs go. The first two set the denominator; the third sets the funding chain for the whole AI trade. If two of the three turn the wrong way, “TSMC’s earnings were great” stops being my reason to feel calm.

Third, the hardest one: I’m underwater, do I cut or hold? The episode doesn’t answer that, and I won’t answer it for anyone either. What I’ll share is the one tool I took from it. On the inventory question, Jat gave a clean decomposition: rising inventory isn’t scary by itself — look at whether raw materials and work-in-progress are driving it or finished goods are piling up, and compare inventory growth against revenue growth. The spirit of that is this: the same rising number means the opposite thing depending on which hand is pushing it. I moved that spirit over to checking my own holdings. When something drops, I ask which hand is pushing — the business got worse (numerator), the market’s discount rate changed (denominator), or margin positioning is being flushed. Those three call for three different responses, and blending them into one feeling leads to exactly one decision: selling because of how the day felt.

Worth a look

• MacroMicro After Meeting EP.218, recorded 16 October 2026, hosted by Roger with Ralice and Jat
• The latest US CPI and PPI prints (headline CPI 3.7% y/y, core CPI 0.3% m/m, PPI from 5.6% to 6.4%)
• TSMC’s Q3 2026 earnings call and guidance
• ASML’s latest quarterly results (revenue growth, margins, days of inventory)
• Polymarket odds on the US midterm elections
• US midterm seat-change statistics, 1934 to 2022
• Taiwan Stock Exchange’s market-wide collateral maintenance ratio versus the narrower measure of margin-purchase market value divided by margin loan balance (different sensitivity, different use)

The one thing to take with you

Your report card sets the numerator; the environment sets the denominator. What a person, a company or a piece of work delivers only fixes the top half of the equation. What it’s worth still gets divided by whatever the world is dealing with that week. TSMC handed in a near-perfect quarter and the market shrugged. Anyone who sees why won’t read “I did this well and nobody noticed” as “I didn’t do it well enough.”

Here’s something I’ve tried for exactly that kind of swallowed frustration, and it helped: pick one recent case — a proposal nobody replied to, a favour nobody thanked you for, a drawing you praised that your kid didn’t react to. Take a sheet of paper and make two columns. On the left, what you delivered, specific down to numbers and details. On the right, what the other person was dealing with that week — three items, no guesses like “they just don’t care,” only things that actually happened. Then read the right column once and write a single line: if these are still true next month, how do I deliver differently next time?

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.