SHANGHAI, 15 AUG 2026 — SMIC passed US$3 billion in quarterly revenue for the first time, tripled its profit, and told investors it intends to charge more.
The quotable line came from co-chief executive Zhao Haijun on the earnings call: there is still a large gap between industry-leading wafer prices and SMIC's, and so, he said, the company needs to negotiate with customers for fairer pricing. It has already raised prices once after negotiations in the first quarter. It will charge more again for wafers processed in the third.
A foundry running at 93.7 per cent utilisation is in a position to say that.
The quarter
Gross margin reached 25.3 per cent, up 5.2 percentage points on the quarter. Wafer shipments came to 2.9 million eight-inch equivalents, up around 14 per cent. Both revenue and profit beat the average analyst estimate.
Where the growth came from
The composition of that growth matters more than the total. It separates a foundry that is merely busy from one with pricing power. SMIC was both.
Multiply the two and you get about 21 per cent against a reported 20, which is close enough to say the quarter was roughly two parts volume to one part price. That ratio is the interesting number. Volume growth alone would make SMIC a busy foundry in a strong market. A blended selling price rising 5.7 per cent in a single quarter, with management promising more, makes it a foundry that has stopped competing on price.
What is actually scarce
The demand is not where the headlines usually point. SMIC does not make leading-edge AI accelerators; export controls on lithography equipment see to that. What it is selling out of is mature-node capacity.
Management described shifting capacity towards the tighter markets: logic for AI computing outside the traditional processor and accelerator categories, BCD power management, image sensors, display drivers and specialty memory. Smartphone and consumer electronics demand remains weak.
That list is not an AI story in the way the revenue line suggests. Those are the unglamorous parts. Every AI rack needs power management silicon. Every camera module needs an image sensor. Every panel needs a driver. They are made on mature processes, at fabs nobody has built for a decade because the returns were poor. Demand for them scales with the physical build-out of data centres, not with model training.
Capacity is not answering quickly. SMIC added 8,000 twelve-inch wafers of monthly capacity in the quarter, taking the total to about 1.1 million eight-inch equivalents — a rise of 1.7 per cent while shipments rose 14 per cent. Expansion is constrained by the same export controls that keep the company off the leading edge; it cannot freely buy the tools required to add capacity.
Constrained supply, rising demand, 93.7 per cent utilisation. The price rise is not a strategy. It is what that arithmetic produces.
The cost side is climbing too
But rising capital costs complicate the picture. First-half capital spending ran at US$3.4 billion, and first-half amortisation at US$2.3 billion, with full-year amortisation expected around US$5 billion — roughly 30 per cent higher than last year.
That is the delayed bill for capacity added in previous years, and it lands on the income statement whether or not the wafers sell. This spending explains why gross margin at 25.3 per cent still looks thin against the operating performance, and why third-quarter guidance of 26 to 28 per cent is an improvement, not a step change.
Guidance for third-quarter revenue is growth of 2 to 4 per cent sequentially, against 20 in the second. The price rises continue; the volume surge is not expected to repeat.
Why this matters to the region
The results matter to the region for two reasons, one of them immediate.
The first is what it says about Chinese AI economics. We noted this week that DeepSeek is raising API prices while Western labs cut, and moving to peak and off-peak rates that fall inside ASEAN working hours. The usual explanation is demand outrunning capacity. SMIC's results describe the same squeeze one layer down. A domestic foundry at 93.7 per cent utilisation, negotiating prices upward, means the cost of Chinese compute is rising from the silicon up. The cheapest tier in the market now has a floor under it that is not a pricing decision.
The second is more direct. Mature-node parts — power management, display drivers, image sensors — are inputs to the electronics manufacturing that Malaysia, Thailand, Vietnam and the Philippines actually do. Penang assembles and tests. Thai and Vietnamese plants build the products these chips go into. A foundry with 90 per cent of its revenue inside China does not set the regional price by itself. But by raising its own prices it removes an alternative from the market just as regional manufacturers need those parts most.
This is the same pressure we saw in Foxconn's quarter, where the constraint on 2027 was not factory space but advanced packaging. It is the same shape at a different layer. The bottleneck in the AI build-out has moved away from the parts everyone talks about.
What we could not establish
The size of the price rises. Zhao described negotiating for fairer pricing and confirmed increases for third-quarter wafers, but no percentage, no product scope and no customer detail were given. The 5.7 per cent blended figure includes mix effects, so it is not a price rise in itself.
It also remains unclear which customers are paying more, whether long-term agreements are being reopened, and how much of the AI-adjacent demand is domestic accelerator programmes against ordinary industrial and automotive orders. The ceiling on capacity expansion under current controls is likewise unknown. Reporting on profit growth varies between roughly 2.6 and 3 times year on year depending on the measure used, and one secondary account carries a full-year revenue forecast that does not reconcile with the quarterly figures and guidance, so we have left it out.
What to watch
Whether the third-quarter increases hold. A foundry can announce a price rise; whether it survives contact with large customers is a different question, and the third-quarter gross margin will show the answer.
Whether mature-node capacity is added anywhere at scale. This shortage is fixable with fabs, and fabs take three years. Any credible announcement now is a 2029 answer.
Finally, whether that concentration becomes a vulnerability. Ninety per cent of revenue inside China is insulation from export controls and exposure to a single economy at the same time. It has been the former all year.
