For years, three words trailed SMIC wherever it went: behind, restricted, catching up. It could not buy the best tools, could not take the most advanced orders, and the market kept asking the same question: when would it close the gap with TSMC? For a company carrying the hopes of China’s domestic chipmaking, simply surviving and building fabs was a kind of victory. Turning a real profit looked like a distant dream.
The company’s most triumphant moment has arrived precisely when it looked least able to make money. Its latest results show that in the second quarter of 2026, revenue crossed US$3 billion for the first time, up 36.1 per cent year on year, while net profit attributable to shareholders reached US$479 million, a surge of 261.7 per cent. Even after stripping out one-off gains, the core business rebounded sharply: wafer shipments in 8-inch equivalent reached about 2.9 million units, up 14 per cent quarter on quarter, average selling price rose 5.7 per cent, and capacity utilisation hit 93.7 per cent.
Why subsidies alone do not explain the numbers
The easy explanation is state support. When phone subsidies were folded into China’s trade-in programme in January 2025, SMIC’s customer restocking grew visible within a month. But in 2026 that logic cracked. Smartphones, which supplied 25.2 per cent of wafer revenue in the second quarter of 2025, fell to 18.9 per cent in early 2026 and then to 16.9 per cent, shedding nearly 8 percentage points in a year. Consumer electronics still anchors the business at 44.2 per cent, while industrial and automotive rose from 10.6 per cent to 16.5 per cent and PCs and tablets reached 15.6 per cent.
Phone demand is soft for a reason: memory and flash costs climbed so steeply that the subsidy’s saving was eaten by upstream inflation. China’s smartphone shipments fell 4.2 per cent to about 134 million units in the first half of 2026. Yet the broader digital category tells a different story. Subsidised purchases of digital and smart products reached 79.1 million units in the first half, up 13.4 per cent, with June alone jumping 32 per cent, and smart glasses shipments rose 30.6 per cent year on year.
So subsidies did not fail. The incremental benefit simply moved from phones to smart glasses, wearables and new device shapes. None of that explains the core paradox: why did wafer shipments, prices and utilisation rebound together just as phone revenue collapsed?
AI is the silent cash register
The answer sits one layer below the headline. The most valuable part of the AI boom is the top-end GPU, the 3-nm and 5-nm nodes and advanced packaging, none of which is SMIC’s strength. But a GPU is useless without the surrounding grid. High-voltage current entering an AI server must be converted, stepped down and distributed into the low-voltage, high-current supply a GPU needs, and the links between GPUs, HBM and switch chips depend on interface and control chips. More GPUs mean more supporting chips.
Infineon’s measurements put the power-semiconductor content of a standard server at just US$65 to US$80, but an AI server pushes that to US$850 to US$1,800. UBS forecasts the global AI data-centre power-semiconductor market will grow from about US$1.5 billion in 2025 to about US$2.5 billion in 2026 and roughly US$3.8 billion by 2028. Most of these chips need 55-nm, 65-nm or even 90-nm processes, exactly SMIC’s deepest home turf. The mature nodes once dismissed as “not advanced” are now the ones raising prices.
The second, quieter dividend is global capacity reshuffling. As TSMC and Samsung pour capital and engineers into advanced nodes, TrendForce expects global 8-inch wafer capacity to shrink 2.4 per cent in 2026. Supply is contracting while demand from cars, industrial control, consumer electronics and AI servers keeps rising. Since the second half of 2025, some high-voltage and image-sensor customers have begun shifting tape-outs to Chinese fabs to lock in certain supply. With the largest mature-node capacity and the broadest platform, SMIC is the natural recipient, and switching foundries is expensive enough that incoming orders stick.
Cashing the cheque, then betting it again
Depreciation rose from US$879 million in the second quarter of 2025 to US$1.088 billion and then US$1.212 billion a year later, up nearly 40 per cent. That is dangerous for a foundry, because depreciation never stops. But the AI dividend flipped the math: at 93.7 per cent utilisation, SMIC is raising prices on tight categories, moving from “build fabs and wait for orders” to “orders chasing capacity”. With little room to push more out of existing lines, it is expanding again, lifting capital expenditure to US$1.836 billion in the second quarter from US$1.563 billion, targeting about 40,000 additional 12-inch wafers a month by year-end versus the end of 2025.
It can afford the bet. Operating cash flow reached US$2.522 billion in the quarter and cash plus equivalents exceeded US$8.2 billion, while payables and contract liabilities rose to US$3.466 billion and US$768 million, cushioning the spend. The real shift is a virtuous circle: more orders lift utilisation, tightness lifts prices, prices lift profit and cash, and cash funds more capacity.
SMIC never forged the brightest jewel in the AI crown. It is turning the base that holds the crown into a quietly very profitable business. The most consequential moment for domestic substitution was never a loud slogan. It is fabs running full, prices rising, and the cash earned being enough to build the next fab.
Editor’s note: This is an adapted translation of the original Sohu IT report. It has been trimmed and restructured for readability for an international business audience.