SMIC (Trans): AI ancillary chips in short supply through 2027

Compiled by Dolphin Research: $SMIC(00981.HK) Q3 2026 earnings call Trans

I. Key takeaways from the print

1) Guidance. Details below.

a) Q4: revenue up 4%–6% QoQ, with shipment volume and ASP each contributing roughly half. GPM at 26%–28%, in line with last quarter's guide.

b) Full year: based on unaudited 9M data and mid-point of Q4 guide, revenue to top $12.0bn (+~30% YoY). GPM around 26% (+~500bps YoY), with total depreciation up ~30% YoY.

2) Key financials this quarter. Details below.

a) Profitability: revenue $3.19bn (+6.1% QoQ). GPM 29.9% (+460bps QoQ), OP $710mn, EBITDA $2.04bn with 63.9% margin, and net profit to shareholders $428mn.

b) Volume/price: 8-inch wafer equivalents shipped topped 3.0mn for the first time (+5.7% QoQ). Avg. wafer ASP rose 1.8% QoQ.

3) Balance sheet and cash flow. Details below.

a) Total assets $59.1bn at quarter-end, including cash $13.4bn. Total liabilities $20.4bn, of which interest-bearing debt $14.9bn. Total equity $38.7bn.

b) D/E (interest-bearing) at 38.5%. Net D/E at 3.9%.

c) CFO net inflow $1.182bn, CFI net outflow $2.104bn, CFF net inflow $910mn.

4) GPM drivers and capex. Details below.

a) Q3 beat: rush orders from customers; the company slowed engineering/R&D wafers. Monthly capacity added by ~20k 8-inch equivalents, and utilization rose 360bps QoQ to 97.3%, lifting output high-single-digit QoQ, diluting unit cost and offsetting a 2.3% QoQ increase in depreciation.

b) Q4 moderation: depreciation and materials cost pressures continue, with more engineering/R&D wafers. Utilization edges down as added capacity keeps ramping.

c) Capex: tool lead times have lengthened, so FY capex will be slightly below plan but expansion milestones intact. Q4 depreciation remains consistent with prior full-year guidance; depreciation will increase further in 2027, with magnitude to be disclosed next earnings.

II. Details from the call

2.1 Management highlights

1) Demand and revenue drivers. Details below.

a) Versus three months ago, demand from customers for AI-adjacent chips such as BCD, specialty memory and controllers exceeded expectations, pushing Q3 revenue above the high end of guidance.

b) Volume: accelerated new line qualifications and flexed existing capacity. Price: staged price adjustments negotiated with customers took effect, while product mix kept improving.

2) Revenue mix. Details below.

a) By region: China, U.S., and Europe/Asia ex-China accounted for 90.8%, 7.8%, and 1.4%, respectively. China revenue rose 7% QoQ as local demand outpaced the company average, prompting capacity reallocation.

b) By size: 12-inch and 8-inch contributed 79% and 21%, respectively, a stable mix. Revenues grew 9% and 4% QoQ, with expansion focused on 12-inch.

3) End markets. Details below.

a) Mix: smartphones 14%, PCs/tablets 18%, consumer electronics 41%, connectivity/wearables 7%, industrial/auto 19%.

b) Softer segments: smartphone and consumer electronics declined 8% and 1% QoQ, with demand still weak.

c) Outperformers: PCs/tablets, industrial/auto, and connectivity/wearables saw limited market change overall, but benefited from onshoring and proactive platform rollouts. Company revenues in these areas rose 26%, 26%, and 18% QoQ.

d) Platform examples: BCD covers low/medium/high-voltage, supporting up to automotive Grade 0. Products span BMS, analog front-ends, system basis chips, digital isolation, isolated drivers and sampling front-ends, and smart power switches.

4) Q4 outlook. Details below.

a) Q4 is a seasonal trough, but the company expects a 'not-so-slow' season. Weaker smartphone and some consumer orders free up capacity that is being shifted to higher value-add products such as specialty memory and BCD.

b) Shipments should stay solid, while mix upgrades and market-based pricing drive ASP higher.

5) Medium to long term. Details below.

a) Two certainties: AI could extend the upcycle, reshaping growth mix and cycle pattern. Full value-chain localization continues.

b) Response: deepen strategic ties with customers and match Chinese customers' pace of innovation to strengthen specialty and customized processes. Expand capacity in line with customer-led demand, building an edge with higher value-add portfolios.

c) Operating goal: stabilize EBITDA, i.e., cash generation from operations, and improve cost control and efficiency.

2.2 Q&A

Q: GPM at 29.9% far beat guidance in Q3. Why guide Q4 still at 26%–28%? What drove the cost change?

A: Q4 is not a down quarter; guidance is the same as Q3, with swings driven by seasonal wafer start cadence. We plan utilization, including new capacity, at around 95%, reserving headroom for line debug, engineering runs, and R&D. In Q3, rush customer orders were high across several quarters, so we used all the headroom for production.

In Q4, customers conduct annual acceptance and planning, and wafers started from Oct–Dec ship in late Jan to early Feb, which coincides with end-demand seasonality and uncertain next-year forecasts, so wafer starts decrease.

Revenue holds up, but utilization and GPM carry more burden. Costs are not easing via depreciation: both full-year depreciation and revenue rise ~30%, keeping depreciation per revenue dollar roughly flat, with efficiency gains offsetting materials inflation. Depreciation per revenue dollar in Q4 will increase by about 1.4%–1.5%.

Q: Should EBITDA be the better operating metric?

A: Yes. With heavy capex and a rising share of new capacity, EBITDA is more indicative than GPM. Every 3%–5% mix increase of new capacity can distort GPM, so margins can swing even in healthy growth.

Q: With power shortage concerns, how do you view next year's global AI demand? Any impact on foundry supply-demand and your capacity plans?

A: Customer-indicated demand remains undersupplied; we are constructive for next year and see 2027 as very tight. AI is still early, and compute users broadly face shortages. Enterprises are starting to build small data centers as rented tokens are volatile in both supply and pricing.

Power constraints are not confined to one region or company: U.S. firms will build near gas fields or in power-abundant areas, and there is independent data center demand in China. We are not reliant on a single market.

Consensus points to severe tightness in 2027. By mid to late 2028, announced memory expansions exceed compute and companion chips, so AI-related memory may balance, while compute and companion chips could still be short. We will expand per customer roadmaps and keep investing.

Q: What drove the back-to-back acceleration over the last two quarters? How do you view demand by application and platform?

A: AI data center adjacencies continue to widen the supply gap, and auto/industrial are recovering, with only smartphones and lower-end consumer soft. Total orders are ample; utilization shifts reflect seasonal pull-in timing, not demand scarcity.

AI data center and enterprise workstation companion chips are in shortage, including specialty memory, BCD power, optical module TIAs, and storage controllers.

Auto and industrial are growing counter-seasonally, with industrial customers starting to restock. Mid/high-end networking and IoT continue to pull in; low-end networking/IoT softened. LED wall driver ICs and fingerprint ICs are recovering, while panels, tablets, and PCs remain weak, with rush orders at the high end but low/mid segments subdued.

Q: Capex execution was behind in 9M. Will it concentrate in Q4? How to think about Q4 and next year's depreciation and the peak?

A: Longer tool lead times will make this year's capex slightly below plan; depreciation will increase next year, with the peak timing not provided. Industry-wide early ordering stretched tool lead times far beyond a quarter, pushing some deliveries planned for this year into next. New orders generally take 16–24 months, with critical tools over 24 months.

Q4 depreciation is broadly consistent with the prior full-year outlook and will still rise QoQ, but depreciation as a share of revenue changes little, limiting GPM impact. Next year's depreciation needs further budgeting; given heavy investments in recent years, it will keep increasing, with the exact rate to be disclosed next quarter.

Q: With raw materials inflation, what advantages do you have to mitigate cost pressure?

A: We can largely offset materials inflation for now, but pressures are mounting and can't be countered by utilization gains anymore. We diversify materials sources and run centralized and competitive procurement, plus longer-term agreements to smooth price increases.

We expect further materials price hikes. Utilization was already high in Q3 and Q4, so we will lean on quarterly cost-reduction targets, process improvements, and efficiency gains to absorb cost pressure.

Q: R&D expense rose notably QoQ in Q3. Does that pull forward Q4 spend?

A: Full-year R&D will be $800mn–$900mn; Q4 should be broadly flat vs. Q3. Quarterly swings reflect R&D progress and settlement timing across materials, tools, maintenance, labor, and software. The Q3 increase mainly came from materials and maintenance.

Q: Specialty memory prices rose in 1H. How did they trend in Q3, and what about sustainability and any cap that could hurt downstream demand?

A: One round of price hikes has completed, and we price to market; prices should stabilize for a period and not suppress downstream demand. Specialty memory is mainly NOR Flash and NAND Flash, where we are the largest domestic supplier. We reference Taiwan peers and industry trends, moving up in upcycles and down in downcycles.

Supply-demand for small-capacity NAND at 64Mb and below is largely balanced, so further increases are unlikely. High-density, high-quality products remain short, leaving room if market ASPs continue to rise.

From Q1 2025 to Q3 2026, specialty memory price gains are far smaller than DRAM and remain acceptable; for example, high-density NOR in earbuds has not lifted device prices.

DRAM, however, is creating pressure: DRAM in smartphones now exceeds the prior full device price of entry-level models, pushing entry models from ~RMB1,000 to ~RMB2,000. In IPC, DRAM cost exceeds previous system prices, and AI demand is crowding out non-AI segments, which are down double digits.

Q: How is the cycle for auto/industrial chips? Any capacity bottlenecks next year?

A: Auto/industrial is recovering, with orders improving QoQ since 2H; we did not directly comment on bottlenecks. After overstocking in 2022, the segment fell from 2023 and lingered at the bottom. It has not yet recovered to 2022's peak, but each quarter is better than the last, and our revenue mix in auto/industrial is rising.

Our growth may outpace the industry: legacy smartphone and consumer customers face AI crowding and macro headwinds, and may shift further into auto/industrial next year to seek growth. Even if the overall market does not grow much, their mix in these fields should rise, and we have pre-built the needed technology and platforms.

Q: For Q4's 4%–6% revenue growth, how do volume and price split? How much pricing room remains in Q4 2026 and 2027?

A: Volume and price contribute roughly half each, with shipments and price up 2%–3% apiece; next year's price hikes are TBD. The previously announced adjustments are gradually flowing into shipments.

We do not lead with price and plan with customers long term, not for single-quarter targets. With AI crowding, smartphones and lower-end consumer should decline further next year and have no pricing power; whether to cut depends on the industry.

AI adjacencies, auto/industrial, and specialty processes have no room for price cuts; pricing should be solidly higher next year, with magnitude depending on the market and customer tolerance. Materials and power cost inflation is currently offset by efficiency, but core process tool prices are rising; as costs move up, we will discuss sharing with customers.

Risk disclosure and statement:Dolphin Research Disclaimer and General Disclosure