SMIC Crossed $3 Billion Without a Leading-Edge Node
China's largest foundry posted $3.01 billion in Q2 on ~95% utilization and 25.3% gross margin, driven by domestic AI orders — none of it requiring the process technology it is barred from buying.
SMIC reported second-quarter 2026 revenue of $3.01 billion, up 20% sequentially from $2.51 billion and 36.1% year over year from $2.21 billion. It is the first time China's largest foundry has cleared $3 billion in a quarter.
Utilization ran near 95%, including newly added capacity. Gross margin rose to 25.3% from 20.1% in Q1 and 20.4% a year earlier. The company shipped roughly 2.9 million 8-inch-equivalent wafers, up 14% sequentially, and China revenue climbed 22% to become the primary growth engine.
The company plans to begin reporting AI chip revenue as a separate category.
Full utilization is the number that matters
Revenue growth can come from price or volume. Ninety-five percent utilization tells you which constraint SMIC is operating against: it is selling everything it can make.
That, combined with a five-point sequential gross margin expansion, is the signature of a supplier with pricing power. Foundries do not expand margin five points in a quarter through cost discipline. They do it because customers are competing for slots, and because the mix is shifting toward wafers that command better prices.
Management attributed growth to AI-related orders from Chinese customers, returning overseas orders, and accelerated localization — plus some orders pulled in earlier than expected. That last item is the one to discount: pull-ins borrow from the next quarter, and a business running at 95% cannot absorb much more of them.
The export controls didn't stop this, and the reason is instructive
SMIC is cut off from EUV lithography. It cannot buy the tools required to compete with TSMC at the leading edge, and no amount of domestic capital changes that in the near term.
It just posted record revenue anyway, because the assumption embedded in the leading-edge framing is wrong: that AI demand is exclusively demand for leading-edge silicon.
An AI data center is not only accelerators. It is power delivery, voltage regulation, network switching, optical transceiver drivers, sensors, microcontrollers, and memory interface logic — a large volume of which is manufactured on mature nodes that SMIC serves competently and at scale. Every accelerator shipped pulls a basket of mature-node components along with it.
The revenue split reinforces the point: consumer electronics 44%, smartphones 17%, industrial and automotive 17%, computers and tablets 16%, connected and wearables 7%. This is a broad mature-node business benefiting from an AI buildout it is not directly participating in.
Export controls were designed to deny the frontier. They did that. They did not deny the adjacency, and the adjacency is where the volume lives.
Separating out AI revenue is a signal, not an accounting change
Companies create a new reporting line when the number is about to be worth showing.
Breaking out AI chip revenue does three things at once. It gives investors a growth metric to underwrite, distinct from the cyclical consumer business. It gives Chinese customers a visible commitment that domestic AI silicon has a committed manufacturing partner. And it establishes a baseline against which policy — on both sides — will be measured for years.
It is also a statement about where capacity will be allocated when the fabs are full, which they currently are. A foundry that reports AI revenue separately is a foundry that intends to grow it faster than everything else.
What this does to the localization thesis
The strategic argument for Chinese semiconductor self-sufficiency has always run into an uncomfortable fact: the frontier gap is real and not closing quickly.
SMIC's quarter reframes the argument. Self-sufficiency does not require matching TSMC at 2nm. It requires that a domestic AI buildout can be supplied domestically for the majority of its component count, with the accelerator remaining the hard, contested piece.
At 95% utilization with margins expanding and localization cited as a named growth driver, that condition is being met for everything except the accelerator itself. The bottleneck narrows to one component — which is precisely why the domestic accelerator effort is where the state and the capital are concentrated.
Two risks sit against that. The first is cyclical: mature-node capacity is being built globally, and mature nodes have historically been where price wars start. The second is that a business running full is a business that cannot grow without capex, and capex on restricted equipment is slower and dearer than it is for competitors.
The read
SMIC's record quarter is not evidence that export controls failed. The frontier gap is intact.
It is evidence that the controls were aimed at a narrower target than the market. AI infrastructure spending is enormous, diffuse, and only partly leading-edge — and the diffuse portion flowed to whoever had capacity, which increasingly is the domestic supplier for domestic customers.
The next number worth watching is the one SMIC just promised to publish. If AI chip revenue is disclosed separately next quarter and grows faster than the total, the localization thesis stops being a policy aspiration and becomes a line item.