Chinese Foundries Ride Domestic AI Chip Wave to Record Margins

The Numbers Tell a Story of Realignment

Semiconductor Manufacturing International Corporation and Hua Hong Grace Semiconductor delivered second-quarter net profits that climbed 261.7 percent and 385.9 percent year-on-year to $479.2 million and $38.6 million, respectively, according to the companies. Those figures, reported in early August, mark the sharpest quarterly earnings acceleration either foundry has logged in at least four years and underscore a fundamental reordering of capacity allocation across China's chip ecosystem.

At DailyTechWire, we've tracked dozens of foundry earnings cycles across Taiwan, South Korea, and mainland China over the past decade; sustained triple-digit profit growth in mature-node fabs is vanishingly rare outside of a technology inflection. The Q2 results suggest that inflection is now visible in production data, not just design-win announcements or government procurement pledges.

Capacity Reallocation Toward Inference and Edge

The profit surge reflects a confluence of volume and mix. Chinese AI chip designers - ranging from established players like Biren Technology and Moore Threads to a long tail of startups targeting inference accelerators, edge vision processors, and large-language-model training clusters - have shifted tape-out schedules toward domestic fabs in the eighteen months since Washington tightened export licensing for advanced logic and high-bandwidth memory. That regulatory environment has compressed the addressable supplier base for many Chinese system houses, particularly those building data-center infrastructure or autonomous-vehicle compute stacks that fall under dual-use scrutiny.

SMIC, which operates 300 mm lines spanning 28 nm down to its mature FinFET variants, has seen utilization rates climb back above 90 percent across several of its Shanghai and Beijing facilities, reversing the inventory correction that weighed on margins through much of 2025. Hua Hong, historically concentrated in power management, radio-frequency, and embedded flash on 90 nm and 55 nm nodes, has likewise reported tighter line loading as AI edge devices proliferate in consumer electronics, industrial automation, and smart-city deployments.

The mix shift matters as much as the volume. Inference chips - especially those optimized for INT8 or mixed-precision workloads - can often be fabricated on trailing-edge nodes with competitive power efficiency, provided designers are willing to trade absolute performance for lower cost and supply-chain sovereignty. That trade-off has become more palatable as export controls have severed access to cutting-edge packaging services and as Beijing has channeled subsidies toward domestic tool procurement and design-service ecosystems.

Margin Expansion Beyond Volume Leverage

Revenue growth alone does not explain the profit acceleration. Both foundries posted gross-margin expansion of several hundred basis points quarter-on-quarter, a signal that wafer pricing has firmed and that product mix has tilted toward higher-value process variants. In SMIC's case, the company has been able to command premium pricing for its 28 nm high-k metal-gate platform, which remains the workhorse node for many AI accelerators that do not require sub-10 nm transistor density. Hua Hong, meanwhile, has leveraged its specialty-process portfolio - embedded flash, silicon-germanium BiCMOS, and power discrete integration - to serve the analog and mixed-signal periphery of AI systems, where margins traditionally run higher than commodity logic.

Operating-expense discipline also contributed. Both companies held research-and-development spending roughly flat in dollar terms even as revenue climbed, resulting in R&D intensity that dropped by 200 to 300 basis points year-on-year. That restraint reflects a pragmatic calculus: with access to the most advanced lithography and deposition tools constrained by export policy, incremental R&D investment yields diminishing returns on leading-edge process development. Instead, both foundries have redirected engineering resources toward yield improvement, design-for-manufacturability collaboration with domestic fabless customers, and process-design co-optimization for AI workloads.

Regional Implications and Competitive Dynamics

The earnings upswing at SMIC and Hua Hong arrives at a moment of strategic uncertainty for Asia's broader foundry sector. Taiwan Semiconductor Manufacturing Company continues to dominate the leading edge - 3 nm and below - but has faced questions from investors and customers alike about the long-term viability of concentrating so much advanced capacity on an island 180 kilometers from the mainland. Samsung Foundry, after a difficult two years marked by yield challenges and customer defections, is in the midst of a turnaround effort centered on gate-all-around transistor technology and closer collaboration with hyperscale cloud providers.

Mainland Chinese foundries, by contrast, are carving out a defensible position in the middle of the cost-performance spectrum. They cannot yet compete on transistor density or power efficiency at the bleeding edge, but they offer something increasingly valuable to a large cohort of Chinese system companies: predictable capacity allocation, local technical support, and immunity from the licensing delays and end-use restrictions that have plagued cross-border semiconductor transactions since 2023.

That value proposition extends beyond AI. Automotive-grade chips, industrial controllers, and consumer IoT devices - categories where node leadership matters less than reliability, qualification pedigree, and supply continuity - are all seeing elevated demand in China's domestic market. SMIC and Hua Hong are well positioned to capture that demand, particularly as Chinese electric-vehicle manufacturers and Tier-1 suppliers accelerate efforts to localize their semiconductor supply chains in response to both policy incentives and risk-management imperatives.

The Capital-Expenditure Question

Sustaining the current growth trajectory will require capital. Both foundries have signaled plans to expand 12-inch wafer capacity over the next eighteen to twenty-four months, though neither has disclosed specific investment figures for 2027. The financing environment remains complex: US restrictions on the export of certain semiconductor manufacturing equipment mean that capacity additions must rely on a patchwork of domestic toolmakers, legacy tools purchased before licensing rules tightened, and selective imports from Japan and Europe that navigate a shifting regulatory landscape.

At the same time, Beijing's willingness to backstop foundry expansion through the National Integrated Circuit Industry Investment Fund - colloquially known as the "Big Fund" - has provided a measure of financial flexibility that few other jurisdictions can match. The third phase of the fund, launched in late 2024, earmarked several billion dollars for equipment procurement, advanced packaging, and design-service infrastructure. SMIC, in particular, has been a direct beneficiary, securing low-cost financing for capacity additions in Shenzhen and Tianjin that are scheduled to come online in 2027.

Forward View

The second-quarter results offer a snapshot of a foundry market in transition. Chinese AI chip demand - driven by a combination of genuine application growth, strategic stockpiling, and the imperative to build indigenous supply chains - has created a revenue and margin tailwind that few industry observers anticipated eighteen months ago. Whether that tailwind persists depends on variables that extend well beyond fab operations: the pace of AI model deployment in China, the trajectory of US export policy, the competitiveness of domestic AI accelerators relative to offshore alternatives, and the broader health of China's economy.

What is clear is that SMIC and Hua Hong have moved from defensive positioning to offensive growth. They are no longer merely surviving in the shadow of export controls; they are capturing share, expanding margins, and demonstrating that a viable foundry business can be built on mature nodes, specialty processes, and a captive domestic customer base. For the rest of Asia's semiconductor industry, that shift represents both a competitive challenge and a reminder that technology leadership is only one dimension of foundry success. Proximity, policy alignment, and supply-chain resilience increasingly matter just as much.