SMIC (Trans): Price hikes lift GPM, depreciation still a drag

The following is Dolphin Research’s transcript of SMIC’s FY26 Q2 earnings call

I. Key takeaways

1) Q3 guidance: revenue is expected to grow 2%–4% QoQ, with GPM guided at 26%–28%. This margin improvement comes despite higher depreciation and summer peak electricity costs in Q3, and assumes utilization, including new capacity, stays around 95%, helping dilute unit fixed costs.

2) Q2 headline financials Revenue was $3.006 bn (+20% QoQ), with GPM of 25.3% (+520 bps QoQ). OP was $534 mn; EBITDA was $2.109 bn, implying a 70.2% EBITDA margin; attributable net profit was $479 mn.

Balance sheet: total assets were $57.2 bn at quarter-end, including cash of $13.9 bn. Total liabilities were $19.3 bn, of which interest-bearing debt was $14.0 bn; total equity was $37.9 bn. Interest-bearing D/E was 37%, with net D/E at 0.4%.

Cash flows: operating cash flow was $2.522 bn. Investing cash outflow was $2.271 bn; financing cash inflow was $607 mn.

3) 1H unaudited results Revenue was $5.511 bn (+23.7% YoY), with GPM of 22.9% (+150 bps YoY). OP was $782 mn; EBITDA was $3.545 bn, with an EBITDA margin of 64.3%; attributable net profit was $677 mn.

Capex: $3.4 bn in 1H.

4) Depreciation cadence Q2 depreciation was $1.22 bn, roughly $2.3 bn in 1H. Full-year depreciation is expected to rise about 30% YoY to nearly $5.0 bn, increasing by roughly $120–$140 mn per quarter.

GPM impact: the quarterly step-up in depreciation drags GPM by about 400–500 bps. Ex-depreciation increases, Q3 GPM uplift would be roughly 5 pts above the current guidance.

Peak timing: based on previously announced expansion plans, depreciation should peak in 2027. However, as capacity plans are being adjusted to orders and market conditions, the exact peak timing remains uncertain.

5) One-offs: Q2 EBITDA hit a record high and included other gains recognized in Q2. Fair value changes from associates and financial assets in Q3 are uncertain and may affect EBITDA, but ex one-offs, EBITDA should remain elevated and improving.

II. Detailed call notes

2.1 Management highlights

1) Volume, pricing and capacity Q2 revenue topped $3 bn for the first time, with both volume and ASP up, driving strong YoY and QoQ growth across core operating metrics. Shipments rose 14.4% QoQ, while ASP increased 5.7% QoQ.

The shipment increase was driven by AI-related demand for supporting chips and customer pull-ins. SMIC added 8,000 wpm of 12-inch equivalent capacity in Q2, with utilization at 93.7% (+60 bps QoQ).

2) Revenue mix by region and wafer size By region, China/US/Eurasia accounted for 90%/8%/2%, respectively, with absolute revenue growth across all. China led with +22% QoQ, supported by robust AI-related chip demand, onshoring, and order reshoring.

By wafer size, 12-inch and 8-inch contributed 78% and 22% of revenue. 12-inch revenue rose 24% QoQ, while 8-inch increased 11% QoQ.

3) Revenue mix by application Smartphones, PCs/tablets, consumer electronics, connectivity/wearables, and industrial/auto accounted for 17%, 16%, 44%, 7%, and 17%, respectively, with absolute increases across the board. Capacity was tilted toward tight segments; PCs/tablets and industrial/auto were up around 40% QoQ in absolute terms.

Other applications softened overall, but customer pull-ins offset some weakness. Smartphones, consumer electronics, and connectivity/wearables grew 8%, 16%, and 13% in absolute terms.

4) Profit drivers Q2 GPM was 25.3%, up 520 bps QoQ, mainly on higher ASPs and better-than-expected utilization, which offset depreciation headwinds. EBITDA margin rose nearly 13 pts QoQ to 70%.

5) 2H outlook AI and spillover effects should continue to generate broad-based demand for foundry services. SMIC will flexibly allocate existing capacity and fast-track new capacity ramps to ease bottlenecks.

Supply chain cost inflation has reached manufacturing, and the company will actively mitigate the impact. Management remains optimistic on the industry and company trajectory.

2.2 Q&A

Q: Shipments rose 14.4% QoQ in Q2, well ahead of capacity growth (~1.7% QoQ) and the slight uptick in utilization. Was the delta mainly from new fab ramps or WIP/finished goods release?

A: Revenue growth was primarily shipment-driven, with pricing also contributing. Customers urgently needed wafers, so we accelerated back-end stages near shipment while slowing some new tape-ins. With the same total capacity, we shifted more process capability to later stages, enabling more deliveries in Q2 in response to customer pull-ins.

There was also net new capacity. Capacity built over recent years has been undergoing qualification, with a typical cycle of around 16 months from line build and product insertion to customer PCN release. Those prior investments are now ramping with demand, contributing to net capacity additions.

Q: Industrial/auto grew 42% and PCs/tablets 39%. How much was AI-related supporting chips, and which products specifically?

A: AI supporting chips plus PCs/tablets together delivered roughly 40% growth. We cannot precisely split AI versus industrial/auto yet due to strict reporting taxonomy, as AI supporting chips were historically classified under industrial or edge computing. We may break AI supporting chips out in coming quarters, but demand is large and we pulled in significant volumes for customers, with reported growth around 40%.

Q: Q3 guidance implies revenue +2%–4% QoQ and GPM of 26%–28%. Does this mean margins keep moving up? Was volume relatively flat due to Q2 pull-ins, and how should we read the slower growth vs. Q2?

A: Revenue depends on both shipped wafers and whether downline wafers are booked at old or new pricing. Many wafers shipped in Q2 were downlined in Q1 or Q4 last year at old prices, as our pricing is set at downline. From Q3, wafers downlined or shipped will increasingly reflect new prices, lifting margins.

Depreciation rises every quarter and drags GPM by 4%–5%. Q3 margins can still move higher, and ex the depreciation step-up, GPM would rise about 5 pts more than current guidance. Thus, Q3 is primarily price-led, with limited volume increase.

On capacity, our ramps are steady and linear; we cannot unleash a sudden surge. We added about 8,000 wpm of 12-inch on Avg. in Q2, and Q3 is similar. We already met more end-stage demand in Q2, leaving less room for further pull-ins in Q3, as over-pulling back-end could disrupt steady front-end flow.

We aim to smooth revenue in Q3, balancing customer pull-ins with healthy line operations. This is the equilibrium we target.

Q: How many pricing rounds have occurred YTD and by how much? Was it broad-based or selective, and is there further upside into H2 and 2025? Q2 Avg. ASP was ~$990–$991 (8-inch equivalent), +5.7% QoQ. How much was pure pricing vs. mix?

A: Our approach differs from peers; we do not issue blanket notices. We negotiate with customers segment by segment, prioritizing tight nodes and markets where we have built advantages. Given smartphones, auto, and industrial have been weak and under pressure, we did not push increases there.

We signaled price hikes around Feb and discussed with customers, raising prices on the tightest capacity. Q1 already showed a 5.7% effect; Q2 GPM rose further despite a 4%+ depreciation headwind, implying a 9%–10% sequential uplift ex depreciation, reflecting prices agreed in Q1 flowing through in Q2.

We will continue to negotiate on fully loaded, undersupplied capacity and align with industry pricing. We are neither the first nor the most aggressive on pricing, but seek reasonable terms recognized by customers.

Q: Which sub-segments are seeing tighter demand at the margin, and where is value-add higher?

A: Pricing follows supply-demand. Where demand exceeds supply and we possess competitive tech and quality, pricing moves up. The tightest areas are compute-related and board-level supporting devices for compute, which we call supporting chips, with demand set to grow further.

Industry capex is being revised up. We heard US commentary lifting AI investment from $600 bn to $880 bn, and Chinese internet companies are also raising hardware budgets. Customer plans for follow-on cooperation and downline volumes are well above prior expectations.

Three lines stand out. First, support chips directly tied to data-center boards and compute boards, such as logic, BCD, and optical module Tx/Rx, are structurally undersupplied. Second, AI-related capacity is crowding out traditional products overseas, e.g., memory vendors exiting specialty memory and NOR, leaving demand unmet and prices rising.

Third, reshoring/return flows: smartphone-related chips and DDIC for large and mid-size displays now face capacity shortages, prompting negotiations on future capacity and pricing. Overall, capacity is insufficient: AI and supporting chips lead the tightness, then crowding effects, and then shortages lift pricing for smartphone and panel driver chips as capacity becomes scarce.

Q: Can high utilization be sustained? Could you bucket PMICs, DDICs, and analog into shortage, balanced, and destocking? Industrial/auto mix rose — is it restocking or end-demand-driven?

A: Heat varies across segments. Undersupplied areas include AI, compute, data centers, and edge computing, especially logic and all power management on boards. Power needs are significant across GPUs/CPUs, HBM/DRAM, SSD drivers and SSDs, cooling motor drivers, and optics.

A single rack with 72 GPUs plus peripherals can require over 16,000 power supply/PM components, and supply is short. Destocking is mainly in smartphones, where prior pull-ins are being digested. DDIC and other consumer segments see low prices and limited orders, with subdued demand.

As capacity tightens, customers fear shortages and higher prices next year for smartphones and consumer devices. Suppliers are building some inventory now to secure capacity and mitigate future price risk, driving visible return flows and discussions on year-end and next-year capacity allocations.

The higher industrial/auto mix reflects customers whose products can ship into auto/industrial as well as smartphones/panels. Recently, they kept ordering, mainly for auto/industrial, while smartphones/consumer saw less uptake, so our shipments there declined but mix rose.

Q: How do you allocate capacity among consumer, industrial, auto, and AI clients? Ex mix, how did pricing move QoQ across processes? Are all Q3 price steps locked in, or is there more to come?

A: On pricing, we follow industry trends. We discussed partial increases with customers in Feb–Mar, not across all products. We have not raised smartphone or DDIC pricing given the sector’s challenges, and we did not cut prices during the downturn either.

As capacity tightens and with anticipated recovery and stocking in consumer/smartphones, we will reassess pricing and capacity allocation. We are not the first mover in price cycles.

On capacity allocation, first we honor strategic, long-term commitments. We will not switch to higher-paying newcomers at the expense of committed partners. Second, in undersupplied areas where we are first-tier in the industry, we expect appropriate recognition in pricing, reflecting our quality and competitiveness.

For consumer and smartphones, we aim for best-in-class execution but let pricing follow industry conditions. We segment our portfolio into 8–10 tracks and 5 major application buckets, which may expand to 6–7, including a more granular AI supporting chip category.

Different segments have different cycles. Some customers are price-insensitive and prioritize capacity and R&D support, while others need price support to get through the cycle. This year in consumer and smartphones, we are supporting customers without price hikes or large capacity cuts, hence steady revenue growth and prioritized shipments when customers pull in.

Q: Q2 EBITDA margin exceeded 70%, a top-tier level historically and vs. peers. With GPM guided up again in Q3 and depreciation still high, how do you see EBITDA margin ahead?

A: Near term versus long term: it is too early to comment on next year. For this year, we have provided Q3 guidance and already received Q4 and year-end orders, and we do not see price cuts in 2024. Current pricing, margins, and utilization are sustainable.

Depreciation is up about 30% this year; Q2 was $1.22 bn and full-year should be $4.9–$5.0 bn. Each quarter adds $120–$140 mn, so EBITDA guidance must net out that increase if prices are flat.

Utilization around 95% will be maintained to preserve about 5% capacity for R&D. Further upside relies on price release as capacity tightness persists, but consumer and smartphone/DDIC pricing will not surge this year.

In short, absent the depreciation step-up, current levels could be maintained. Depreciation will trim some metrics, partially offset by modest consumer price gains, though we cannot quantify now. Overall, we remain optimistic as technology and products are well received, and we aim to sustain orders, utilization, EBITDA, and margins.

(CFO) Q2 EBITDA and GPM reached recent highs, helped by other gains and one-offs recognized in Q2. Q3 fair value changes from associates and financial assets are uncertain and could impact EBITDA, but with higher ASPs and growing GPM, ex one-offs, EBITDA should stay high and trend better.

Q: With 2026 depreciation up 30%+, GPM would be pressured, yet margins improved in Q2 and are guided up in Q3. Which contributed more, pricing or mix?

A: Pricing was the larger driver than mix. We do have some flexibility to optimize mix, reallocating capacity from less-needed products to tighter, higher-priced ones, such as shifting some logic to MPU or virtual memory, or moving from drivers to NOR when appropriate.

New capacity is largely pre-ordered by customers, and ramps are quickly filled with higher-quality, better-priced products. This creates a structural advantage.

Q: You previously guided ~30% YoY increase in depreciation this year. How is it tracking, what is H2 depreciation, and when is the peak or inflection?

A: (CFO) We previously guided roughly a 30% YoY increase. H1 depreciation was about $2.3 bn, and full-year should be close to $5.0 bn, consistent with prior comments. On the peak, prior project plans implied a 2027 high, as successive investments enter depreciation.

However, we are adjusting expansion plans to orders and market demand. The exact peak will depend on those updates. With Phase 2 assets entering depreciation, charges will keep rising, so peak timing remains uncertain.

(Co-CEO) Based on disclosed capex, one could compute the precise peak by quarter under the prior plan. But since Feb, customer AI demand has risen materially, and we are adjusting sites and adding equipment. We will communicate changes via meetings or announcements.

Thus, the earlier view of a single peak followed by a steady decline and rising margins may no longer hold. We will expand while safeguarding margins and competitiveness, avoiding undue negative impact, and will keep investors updated.

Q: How is BCD platform supply/demand and pricing vs. other areas, and how much is AI-driven? Many analog vendors emphasize deep process IP and virtual IDM — how much room is there for product migration and process differentiation?

A: The AI-driven wave is boosting both ends of the spectrum: advanced compute/standard logic on one end and mature power circuits on the other, especially 8-inch, where growth far exceeded expectations. Mid-nodes like 40 nm and 28 nm benefit more gradually via spillovers, as they do not directly feed data-center compute.

BCD and analog are broadly undersupplied, and vendors report strong growth. Two points matter: product variety is vast and built over long cycles, so no single vendor can cover all demand, only the highest-spec and best-priced portions. Some high-volume SKUs can migrate to 12-inch, but many niche parts cannot be moved quickly.

Some products must stay on 8-inch due to maturity and equipment fit, and 12-inch lines often use more advanced tools that require time to adapt for older designs. The market is segmented: large-volume, high-voltage/high-current needs are mostly on 8-inch, while some single large runs migrate to 12-inch as 8-inch capacity is constrained.

Demand visibility extends at least through end-2025. Data centers, GPUs, and leading-edge fabs coming online will not be idle; every GPU or compute ASIC requires matched analog, power, optics, data management, HDDs and SSDs. Hence, BCD demand and pricing should remain firm and undersupplied.

Two differentiators stand out across companies. First, voltage capability: higher-voltage designs, e.g., 48V rails requiring 80–100V tolerance, need long reliability proofs and have fewer qualified suppliers. Second, deep, long-standing customer relationships enable rapid scale-ups when demand surges; shifting suppliers mid-cycle is slower.

<End of text>

Risk disclosure and statement:Dolphin Research disclaimer and general disclosure