SMIC 3Q26 Gross Margin Reaches 29.9% at 97.3% Utilisation: Foundry Pricing Splits by Segment as 2027 Capacity Turns Very Tight

恒森科技
wafer foundrycapacitypricingprocurement
SMIC posted 3Q26 revenue of US$3.19 billion, up 6.1% sequentially, with gross margin at 29.9% and utilisation at 97.3%. Management said pricing will diverge by segment — no room for increases in handsets and low-end consumer, a steady upward trend in AI-support and automotive/industrial — and that 2027 wafer capacity will be very tight into mid-to-late 2028, with equipment lead times now at 16 to 24 months.

SMIC (SEHK: 00981; SSE STAR: 688981) reported third-quarter 2026 results on 12 November: revenue of US$3.19 billion, up 6.1% sequentially and above the upper end of guidance, with gross margin at 29.9%, up 4.6 percentage points quarter on quarter and also above guidance. At the earnings call the next day, co-CEO Zhao Haijun attributed the beat to AI-support chips and offered two judgments that matter more to buyers than the headline numbers: foundry pricing will diverge sharply by segment, and 2027 wafer capacity will be very tight.

Q3 growth came from AI-support parts, not consumer

The increase came from both volume and price. 8-inch-equivalent quarterly shipments passed three million wafers for the first time, up 5.7% sequentially, while average wafer price rose 1.8% as negotiated increases landed in batches and the product mix improved. Gross margin moved from 25.3% in 2Q26 and 22.0% in 3Q25 to 29.9%; gross profit was US$953.2 million against US$760.6 million and US$522.8 million (SMIC 3Q26 results announcement, 12 November 2026).

The application mix says more than the total:

  • Consumer electronics remained the largest segment at 41% of revenue, down 1% sequentially;
  • Smartphone revenue fell 8% sequentially, and the company expects further decline;
  • Computers and tablets, and industrial and automotive, each grew 26% sequentially;
  • Interconnect and wearables grew 18%;
  • China accounted for 90.8% of revenue and grew 7%, against 7.8% from the US and 1.4% from Eurasia;
  • 12-inch wafers were 79% of revenue and grew 9%, against 21% and 4% for 8-inch.

Zhao's explanation is that customers pulled in BCD, specialty memory and controller devices — AI-support parts — faster than expected, leaving those products in shortage. Growth in computers, tablets, industrial and automotive came more from supply-chain localisation and platform deployment than from end demand itself. That distinction matters for 2027: the first group tracks compute investment, the second tracks how fast domestic substitution clears qualification.

How utilisation reached 97.3%

Monthly capacity rose by 20,000 8-inch-equivalent standard logic wafers in the quarter, but the real change was in scheduling. By slowing engineering and R&D runs and prioritising customer shipments, SMIC lifted utilisation from 93.7% in 2Q26 to 97.3%, a 3.6-point gain, with output up a high single-digit percentage. Unit wafer cost dilution offset a 2.3% sequential rise in depreciation, and that is where most of the 4.6-point gross margin improvement came from.

In other words, the current tightness is not idle capacity coming back online; it is R&D and engineering headroom handed to volume orders. The other side of 97.3% is that little slack remains — at near-full utilisation, incremental demand can only be met through new-line qualification and cross-platform reallocation, both of which take time. The company also flagged visibly higher raw material prices this year and said it expects them to rise further, a clear pressure on cost and operations despite multi-source sourcing and centralised, competitive purchasing.

Pricing splits: no room in handsets or low-end consumer

The most concrete item from the call was the pricing hierarchy. Zhao's wording: foundry price increases are likely to continue, but the segments will clearly diverge.

  • Smartphones and mid-to-low-end consumer products face further demand decline next year as AI squeezes capacity — there is no possibility of price increases in these segments;
  • AI-support products and automotive/industrial, where demand is recovering, plus the company's proprietary technology platforms, have no possibility of price cuts; pricing should hold a steady upward trend next year;
  • Beyond demand, higher raw material prices and higher procurement prices for major process equipment are also drivers, and the company said it will share these costs with downstream customers.

That lines up with the fourth-quarter guidance: revenue up 4% to 6% sequentially, but gross margin guided at 26% to 28%, below the 29.9% just reported, because of depreciation and material costs plus higher engineering and R&D runs and slightly lower utilisation. Volume growth continues, but the "no off-season" quarter depends on shifting released capacity into specialty memory, BCD and other high-value products rather than on broad price increases. For the full year, SMIC expects revenue above US$12 billion, up nearly 30%, with gross margin around 26%, roughly 5 points higher year on year.

The binding constraint is equipment: 16 to 24 months

Expansion is limited by equipment supply, the key variable for 2027. Zhao said demand across the industry has been so heavy, and pull-in driven so much by fear, that equipment lead times keep stretching: "every equipment order we place comes back with 16 to 24 months to delivery." That does not stop this year's expansion plan from landing, but it can disturb capex execution for the year.

The company's timeline for the industry: compute remains in shortage overall, and the consensus is that 2027 capacity will be very tight, extending into mid-to-late 2028, after which AI-related memory supply and demand balance out while compute and supporting chips stay scarce. Demand-side data points the same way. TrendForce estimates in October 2026 that the nine largest cloud service providers will spend US$957 billion in capex in 2026, up 106% year on year, and more than US$1.5 trillion in 2027, another 60% increase, with AI server shipments up nearly 36% in 2027. The same report notes that some component suppliers now require prepayment to guarantee supply, prompting cloud providers to bring cash outlays forward (TrendForce, 14 October 2026).

Downstream: lock capacity or wait, depending on the part

Translated into procurement terms, the conclusions are not uniform. Industrial and automotive customers face tightening on both price and allocation. SMIC's BCD portfolio spans low, medium and high voltage and supports Grade 0 automotive standards, covering BMS analog front ends, system basis chips, digital isolation, isolated drivers, sampling front ends and smart power switches — exactly where automotive and industrial incremental demand is concentrated, and exactly the side explicitly placed under "no price cuts." Handset and low-end consumer customers face the opposite: no room for foundry increases, with cost pressure coming from their own weakening demand rather than from wafer pricing, which argues against building inventory on the expectation of a price rise.

There is also a timing mismatch. For an automotive platform already in volume, a part swap means re-running bench and vehicle validation, a quarterly exercise. For a platform still in sampling, devices are not frozen, so the window to re-evaluate opens exactly when cost pressure peaks. And a 16-to-24-month equipment lead time means 2027's incremental capacity is already committed by orders placed today: the closer 2027 gets, the less there is to negotiate.

HSY Perspective

The 2027 foundry squeeze is not another shortage cycle; it is AI rewriting the order in which capacity gets scheduled. SMIC ran at 97.3% utilisation with average wafer price up 1.8% sequentially — volume and price rising together points to real shipments rather than pull-in ahead of price letters. Equipment ordered today arrives in 16 to 24 months, so this tightness cannot be capex'd away within the year. TrendForce's October 2026 forecast points the same way from the demand side: the nine largest cloud providers will spend more than US$1.5 trillion on capex in 2027, up 60% year on year, and some are already prepaying suppliers to secure allocation.

On a bill of materials the pressure is uneven. Smartphones and mid-to-low-end consumer have no room to absorb increases, and that revenue is already falling; AI-support BCD, specialty memory, controllers and automotive/industrial parts sit explicitly on the no-price-cut side. The same price letter lands as two different situations — one where cost is pushed straight to the end customer, one where price and allocation tighten together. The automotive cost is not the percentage increase but the requalification: swapping a part on a Grade 0 platform means re-running bench and vehicle validation, a quarterly exercise that outlasts any effective date.

The boundary deserves stating. This rests on SMIC's third-quarter disclosure and call commentary plus TrendForce's demand-side estimate. If consumer demand recovers in the first half of 2027 and SMIC backfills that released capacity into handsets, the divergence stops holding. If equipment lead times stretch further and prepayment spreads beyond memory into other categories, tightness spills over to second- and third-tier foundries and the split between price and allocation becomes steeper than it is today.

Sources: SMIC, "SMIC Reports 2026 Third Quarter Results," 12 November 2026 (smics.com); Jiwei/Ijiwei, "SMIC's Zhao Haijun: AI-support chips in shortage, 2027 wafer supply will be very tight," 13 November 2026 (ijiwei.com); TrendForce, "CSP CapEx Projected to Exceed US$1.5 Trillion in 2027, with AI Server Revenue Growing Over 90% YoY," 14 October 2026 (trendforce.com).