Mature-node shortage is confined to AI-adjacent segments (PMIC + sensors)
Mature-node shortage is confined to AI-adjacent segments (PMIC + sensors)
One-line summary: The "mature nodes are tight again" narrative is wrong as stated — TSMC's CEO says the shortage is confined to power-management ICs and image sensors, both pulled by AI data centers, while commodity consumer mature-node demand is explicitly not strong; the AI bottleneck reaches into trailing edge only where AI itself reaches.
The insight
Trailing-edge tightness is usually read as a broad cyclical recovery signal — consumer restocking, autos, industrial. Asked directly about a "very strong demand recovery" and "some supply issue" at mature nodes, cc-wei refused the general framing and drew a sharp line instead:
"the mature node cover a lot of different segment. Only the one which are related to AI is in shortage... number one, is power management IC, because of all the AI data center need a lot of power management. Those are the mature node technology like 0.18 micron, 19 nm... Also, the sensor portion... Other than that... the consumer product is not in a high demand, other segment is not so strong demand."
This is a segmentation finding of the same shape as the LTA/non-LTA seam in memory (see lta-contract-structure-as-price-insulation): the market divides by exposure to the AI forcing function, not by node. Two implications the general narrative gets backwards:
- AI-adjacent trailing edge is a real, distinct bottleneck. PMIC at 0.18µm/90nm-class nodes is tight because AI data centers need enormous power-delivery content. That is the trailing-edge shadow of the same forcing function driving ai-capex-to-power-and-materials-cascade — and it names different beneficiaries than the leading-edge trade.
- Commodity mature-node is a false positive. Anyone reading trailing-edge tightness as evidence of a consumer recovery is reading an AI signal and mislabelling it. Wei's consumer read is the opposite, and it is consistent with his own platform mix (smartphone −4% QoQ) and with the memory-driven consumer demand destruction in secondary-device-market-supply-constraint.
TSMC is adding mature-node capacity, but selectively and into exactly these segments — JASM Fab 1 (Japan) for CMOS image sensors and ESMC (Germany) for automotive/industrial.
Evidence
- cc-wei in 2026-07-16-earnings-tsm-q2-fy2026 — the load-bearing quote: "Actually, the mature node cover a lot of different segment. Only the one which are related to AI is in shortage, which is the most important one, is the number one, is power management IC, because of all the AI data center need a lot of power management. Those are the mature node technology like 0.18 micron, 19 nm, or something like that. Those are in shortage, definitely. Also, the sensor portion, because you need a lot of sensor to detect the environmental information and put into the AI data center to analyze it. Other than that, other area, just like you pointed out, the consumer product is not in a high demand, other segment is not so strong demand."
- The analyst premise Wei rejected — Jeff Su relaying Felix Pan (KGI) in 2026-07-16-earnings-tsm-q2-fy2026: "He notes there's lots of talk that mature nodes are seeing a strong demand recovery, and the supply is very tight, mature node pricing is very favorable or strong." Wei's answer confirms the tightness only for AI-adjacent segments and denies the general recovery.
- The capacity response is segment-selective, not broad — cc-wei in 2026-07-16-earnings-tsm-q2-fy2026: "we continue to increase, not decrease, our mature node capacity in the higher value added segment. For example, we are increasing our mature node capacity through JASM Fab 1 in Japan for CMOS image sensor application and ESMC in Germany for automotive and industrial applications. In today's market, outside of specific areas such as power management IC and CMOS image sensor, the mature node demand in other commodity areas is not as strong."
- Consistent with TSMC's own mix in the same quarter — wendell-huang in 2026-07-16-earnings-tsm-q2-fy2026: HPC +20% QoQ to 66% of revenue while smartphone −4% QoQ to 22%; IoT 5%, automotive 4%, DCE 1%.
Implications
- Beneficiaries (unverified as expressions — see below): power-management IC vendors with AI-data-center content — MPWR is the purest listed expression, with ADI / TXN / ON / STM as broader analog names; CMOS image sensors point at Sony (SONY) as the volume leader.
- The trap this closes: do not treat trailing-edge tightness as a consumer/industrial cyclical recovery signal. On TSMC's read it is an AI signal wearing a cyclical costume, and the consumer leg is actively weak.
- Reinforces ai-capex-to-power-and-materials-cascade — power delivery is a bottleneck at the silicon level, not just at the grid level.
Contradictions / tensions
- ⚠ The beneficiary names above are this wiki's inference, not Wei's. Wei names segments (PMIC, image sensors), not companies. No source in this ingest connects the PMIC shortage to a specific vendor's order book, pricing, or share. The forcing function is first-party and specific; the tradeable expression is unevidenced and should not be sized without it.
- Single-source. This rests entirely on one CEO's characterization on one call. TSMC also does not serve the whole mature-node market — its read may not generalize to UMC / GlobalFoundries / SMIC mix.
- An interested party: TSMC has a reason to talk down commodity mature-node demand it is de-emphasizing anyway in favor of "higher value-added" segments.
Open questions
- Which listed PMIC vendors actually have AI-data-center content, and is it visible in their guidance?
- Is the PMIC shortage a TSMC-allocation artifact (TSMC prioritizing leading edge) rather than genuine end-demand tightness?
- Does the ESMC (Germany, automotive/industrial) expansion imply Wei sees automotive recovering, contra "other segment is not so strong demand"?
Valuation snapshot
Last refreshed 2026-07-20 (pre-open; marks are the Friday 2026-07-17 close, markets closed over the weekend). Price fills tagged twelvedata.
✎ All six of this page's suggested tickers are now marked — today's Twelve Data pull is broader than the 07-17 run's 77-symbol set, which is why the rows below carry prices this run where they were blank before. Note these names have never been emitted as signals: the pull now covers them, but (per Contradictions above) the beneficiary link remains this wiki's inference rather than sourced evidence, so the marks are context, not a validated thesis.
| Ticker | Price | 52w range | Mkt cap | Fwd P/E | Day / vs 52w hi | What's priced in (one line) |
|---|---|---|---|---|---|---|
| MPWR | $1,312.00 | $702.32–$1,714.09 | — | — | +0.49% day; −23.5% from hi | Named here as the purest listed PMIC expression. ⚠ This is the wiki's inference, not Wei's — no source connects the PMIC shortage to MPWR's order book, pricing, or share |
| ADI | $375.36 | $218.37–$445.91 | — | — | −1.36% day; −15.8% from hi | Broader analog; same evidential gap |
| TXN | $284.02 | $152.73–$334.03 | — | — | −2.47% day; −15.0% from hi | Broader analog; same evidential gap |
| ON | $87.37 | $44.56–$134.92 | — | — | −0.85% day; −35.2% from hi | Broader analog; same evidential gap |
| STM | $62.06 | $21.11–$81.42 | — | — | −1.13% day; −23.8% from hi | Broader analog; same evidential gap |
| SONY | $21.12 | $19.32–$30.34 | — | — | −1.26% day; −30.4% from hi | CMOS image-sensor volume leader; TSMC is expanding JASM Fab 1 (Japan) for image sensors and Sony SSS co-locates there (tsmc-kumamoto-physical-ai-hub) — the closest thing to a named link on this page, and it is still a co-location fact, not a demand figure |
Read-across from names that were marked: TSM closed $398.37 (−2.77%), and its own platform mix is the strongest corroboration of the segmentation this page describes — HPC +20% QoQ to 66% of revenue against smartphone −4% QoQ to 22% (wendell-huang in 2026-07-16-earnings-tsm-q2-fy2026).
Forward-looking outcomes (12-month)
Bull case — a PMIC vendor confirms AI-data-center content in its guidance and the segment re-rates away from the analog cycle: the forcing function is first-party and unusually specific. cc-wei in 2026-07-16-earnings-tsm-q2-fy2026 refused the general "mature nodes are recovering" framing and drew the line himself: "Only the one which are related to AI is in shortage... number one, is power management IC, because of all the AI data center need a lot of power management. Those are the mature node technology like 0.18 micron, 19 nm... Also, the sensor portion." And the capacity response is segment-selective, which is corroborating rather than merely consistent — TSMC is increasing mature-node capacity only "in the higher value added segment," naming JASM Fab 1 for CMOS image sensors and ESMC for automotive/industrial. If a listed PMIC vendor puts AI-data-center content in its guide, the segment separates from the analog cycle it currently trades with. Implied price: unsized — see below. Cited: 2026-07-16-earnings-tsm-q2-fy2026.
Base case — the segmentation is real, correct, and not investable through these names: the shortage stays confined to PMIC and sensors, but no vendor discloses AI-data-center content as a separate line, so the names keep trading as broad analog cyclicals whose other end markets Wei explicitly describes as weak — "the consumer product is not in a high demand, other segment is not so strong demand." The forcing function helps one slice of revenue at companies whose bulk revenue faces the headwind. Implied price: unsized. Cited: 2026-07-16-earnings-tsm-q2-fy2026.
Bear case — the shortage is a TSMC allocation artifact, not end-demand: an equally consistent explanation is that PMIC is "short" because TSMC is prioritising leading-edge wafers and starving its own trailing-edge lines — in which case the tightness reflects supply triage rather than AI-driven demand, and it eases the moment TSMC's capacity catches up. This is an open question on this page, not a resolved one. Wei is also an interested party: TSMC has a reason to talk down commodity mature-node demand it is de-emphasising anyway in favour of "higher value-added" segments. Implied price: unsized. Cited: mature-node-ai-adjacent-shortage.md (contradictions section).
Currently undervalued vs base case? No — and the honest answer is that this page cannot support a valuation view at all, on two independent grounds.
- No prices. None of the six tickers could be marked this run. There is no snapshot to reason from.
- More importantly, no evidenced beneficiary — and that would block the answer even with prices. This page's own Contradictions section is explicit: "⚠ The beneficiary names above are this wiki's inference, not Wei's. Wei names segments (PMIC, image sensors), not companies. No source in this ingest connects the PMIC shortage to a specific vendor's order book, pricing, or share. The forcing function is first-party and specific; the tradeable expression is unevidenced and should not be sized without it."
The forcing function is strong and the chain is missing its last link. That is a research task, not a trade. The value this page delivers today is a trap it closes rather than a position it opens: do not read trailing-edge tightness as a consumer/industrial cyclical recovery — on TSMC's read it is an AI signal wearing a cyclical costume, and the consumer leg is actively weak. Anyone long analog on a "mature-node recovery" thesis is long the wrong reason.
Catalyst path:
- TXN Q2 (late July), ADI Q3 (August), MPWR Q2 (early August), ON Q2 (early August) earnings — the gate that decides whether this page ever becomes tradeable: does any of them name AI-data-center power-delivery content, with a number? Until one does, the beneficiary link stays unevidenced.
- TSMC Q3 2026 (October) — whether Wei repeats the PMIC/sensor carve-out, and whether the ESMC (automotive/industrial) expansion contradicts his own "other segment is not so strong demand" read. That tension is an open question on this page.
- Any UMC / GlobalFoundries commentary — TSMC does not serve the whole mature-node market, and this concept currently rests entirely on one CEO's characterisation on one call. A second foundry either corroborates the carve-out or reveals it as TSMC-mix-specific.