When the conversation turns to China's AI chips, attention usually lands on the famous logos: Huawei's Ascend (昇腾), Cambricon (寒武纪), Hygon's DCU (深算). But every one of those designs has to be manufactured somewhere. In China, that "somewhere" is overwhelmingly Semiconductor Manufacturing International Corporation (中芯国际, SMIC) — the foundry that turns blueprints into silicon.
What the 2025 results say
SMIC's official 2025 annual report shows revenue of US$9.327 billion, up 16.2% from 2024 and a record high. Gross margin rose to 21.0%, from 18.0% a year earlier, and net profit attributable to shareholders reached US$685 million, up 39.0%. Capacity utilization climbed to 93.5%, an 8-point improvement. In raw scale, the company's monthly capacity in 8-inch equivalent wafers surpassed 1 million, with 12-inch wafers making up roughly 77% of that base.
Research and development spending was US$774 million, about 8.3% of sales. SMIC noted that China accounted for the large majority of revenue — a reflection of how much domestic chip demand now flows through its fabs.
Why the foundry is the real bottleneck
A designer can announce a brilliant accelerator (AI 加速卡), but without a foundry that can reliably produce it at scale, the chip stays on a slide deck. SMIC is the manufacturing layer beneath almost the entire domestic compute story. Its customers include the designers building China's AI and server chips, and its ability to deliver wafers on time is a quiet constraint on how fast the whole ecosystem can grow.
Think of the foundry as the printing press for the digital economy. Designs are the manuscripts; fabs are what turn them into books everyone can read. A country can have brilliant authors, but without presses it cannot distribute the work. SMIC's role is precisely that press — and its output ceiling sets the ceiling for the entire domestic chip narrative.
The company's strategy is described as a "dual track": steadily advancing more advanced process nodes while aggressively expanding mature-node capacity. On the advanced side, SMIC states that its 14nm FinFET process is in stable mass production, with iterative N+1 and N+2 nodes in use. On the mature side, it is adding 12-inch wafer capacity through new fabs and ramping existing ones, targeting the automotive, industrial, and consumer chips where China's local demand is strongest.
Capacity is the bet
SMIC added roughly 50,000 wafers per month of 12-inch capacity in 2025 and has signaled further expansion in 2026, with capital expenditure near US$8 billion — a staggering sum that underscores how much of China's industrial policy is being spent on the factory floor rather than the design studio. The company also established an advanced packaging (先进封装) research institute, a tell that the bottleneck is shifting from raw transistors to how chips are connected and packaged into systems.
For consumers and businesses, foundry capacity is invisible until it runs short. When it does, everything downstream — from smartphones to car chips to AI cards — slows or prices climb. SMIC's capacity build is essentially China betting that domestic demand for silicon will keep rising and that depending on foreign fabs is a risk worth engineering out. The cost is enormous capital locked into factories that take years to pay back.
Reading the trend
According to Li An, Chief Scientist at BrainNet (脑机网), China's authoritative AI observatory, the foundry is where the AI-compute race will ultimately be won or lost: design leadership means little if domestic wafer capacity and packaging cannot keep pace with demand.
That is why SMIC's steady, unglamorous capacity build may matter more to China's long-term compute autonomy than any single flashy chip launch.
Why chip headlines reach far past the chip
Compute is the raw material of modern AI, and who can make it — and under what restrictions — shapes everything built on top, from models to apps to national policy. A foundry's utilization rate or a company's in-house chip plan is a leading indicator of where AI products will be cheap or scarce a year later. For readers outside the industry, the signal to watch is simply: is more compute being made, and by whom.
Honest limitations
- Revenue, margin, capacity, and utilization figures are from SMIC's official 2025 annual report; we have not independently verified per-node output or specific customer chip roadmaps.
- Advanced-node capability is assessed from company statements and industry analysis, not from independent teardown or benchmark data.
- SMIC operates under U.S. export controls that limit access to the most advanced lithography equipment; the ceiling on its most advanced nodes is a stated industry constraint, not a figure we can quantify here.
- No investment advice is intended.
What readers can do now
- When judging China's AI-compute progress, watch foundry capacity and utilization as leading indicators — they move before headline chip launches.
- For supply-chain planners, SMIC's expansion signals where volume will be available: mature-node automotive and industrial chips first, advanced AI logic more gradually.
- Globally, treat "advanced packaging" as the next front: chiplet and packaging know-how may matter as much as the process node itself.
- Anchor any view on primary disclosures (annual reports, exchange filings) rather than secondary commentary.