China's SMIC Sees Price Power Return as AI Demand Lifts Peripheral Chip Orders

Foundry Economics Flip as AI Peripherals Take Center Stage

Semiconductor Manufacturing International Corporation has drawn a line in the sand on pricing. Despite visible softness in smartphones and automotive semiconductors, the Shanghai-based foundry told investors mid-August it sees "no chance" of lowering wafer prices, crediting a wave of orders for chips that sit around AI accelerators rather than inside them.

Co-CEO Zhao Haijun pointed to what he called "spillover effects" from artificial intelligence infrastructure. The term captures a shift happening across China's chip ecosystem: as data centers and edge deployments scale up GPU clusters and custom inference silicon, they generate corresponding demand for power management ICs, high-speed serializers, voltage regulators, and interface chips. These peripheral components lack the headline appeal of training accelerators, yet they ship in multiples per system and require mature process nodes that SMIC operates at volume.

At DailyTechWire, we've tracked similar dynamics in Southeast Asia and South Korea, where foundries running 28nm and 40nm lines report utilization spikes tied not to flagship AI chips but to the analog and mixed-signal devices that keep them fed with clean power and fast I/O. SMIC's commentary suggests China's localization push is now deep enough to sustain pricing discipline even as two of the industry's traditional bellwethers, handsets and vehicles, cool.

Localized Manufacturing as Demand Anchor

Zhao singled out the China market as the source of the company's strongest growth, driven by "robust demand for AI peripheral chips and continued strengthening of localized manufacturing." That phrasing is deliberate. Since 2020, when export controls began restricting access to cutting-edge lithography and design tools, Chinese system builders have leaned harder on domestic fabs for anything that can be produced inside the country's borders.

Peripheral AI chips fit that profile neatly. They rarely need sub-7nm geometries, they involve established IP blocks, and they're consumed in high volumes by the same hyperscalers and OEMs now racing to build out inference capacity for large language models, computer vision pipelines, and recommendation engines. For SMIC, that translates into longer lead times, fuller order books, and leverage to resist the price erosion that typically accompanies demand dips in consumer electronics.

The foundry's confidence also reflects a structural change in customer mix. Where SMIC once depended heavily on handset application processors and automotive microcontrollers, AI peripheral orders now provide a countercyclical buffer. Even if a smartphone brand cuts a wafer order by 20 percent, a cloud provider or AI appliance vendor may increase its PMIC or SerDes order by a similar margin, smoothing revenue and preserving gross margin.

Why Peripheral Chips Command Attention

Peripheral AI chips occupy an unglamorous but essential niche. A single GPU server might house two or four high-end accelerators, but it will also carry a dozen power-management ICs to handle multi-rail voltage sequencing, several clock generators for synchronizing data paths, and a handful of PCIe retimers to maintain signal integrity across long board traces. Each of those functions requires silicon, and each chip represents a potential revenue stream for a foundry.

From a design perspective, peripherals are less risky than leading-edge logic. They use proven processes, achieve acceptable yields quickly, and can be qualified in months rather than years. For a foundry operating under technology constraints, they represent a path to growth that doesn't require solving the next lithography node. SMIC's ability to hold prices on these products signals that supply is still tight relative to the pace at which Chinese AI infrastructure is being deployed.

The economics also favor volume. Peripheral chips ship in quantities that dwarf flagship processors. A single data-center deployment might involve tens of thousands of power ICs, each priced in single-digit dollars but cumulatively worth millions in wafer revenue. When multiplied across the dozens of Chinese cloud operators, AI startups, and hardware vendors now racing to field competitive inference platforms, the aggregate demand becomes large enough to move the needle for a foundry of SMIC's scale.

Automotive and Smartphone Weakness in Context

Zhao's remarks implicitly acknowledge that not all end markets are firing. Smartphone unit sales across China have been flat to down for several quarters, and the automotive semiconductor segment has softened as electric-vehicle production growth decelerates and inventories normalize. Both sectors have historically been important to SMIC's revenue mix, and both are now exerting downward pressure on utilization in certain product lines.

Yet the foundry's decision to hold pricing suggests it sees AI peripheral demand as more than a short-term sugar rush. If management believed the AI wave would crest within a quarter or two, the rational move would be to cut prices, defend market share, and ride out the cycle. Instead, SMIC is signaling that it expects localized AI manufacturing to sustain order flow for an extended period, effectively re-rating the foundry's pricing power.

This stance carries risk. If AI infrastructure spending in China slows or if peripheral chip inventory builds faster than anticipated, SMIC could find itself with excess capacity and fewer levers to stimulate demand. But for now, the company appears confident that the gap between domestic AI ambitions and available foundry capacity remains wide enough to justify firm pricing.

What It Means for China's Chip Stack

SMIC's experience offers a window into how China's semiconductor ecosystem is adapting to external constraints. Unable to access the most advanced nodes or tools, Chinese fabs and design houses are optimizing around mature processes and high-volume products. Peripheral AI chips fit squarely into that strategy: they're complex enough to require specialist expertise, mature enough to be manufacturable at scale, and essential enough to command stable pricing.

The localization Zhao describes is not merely a matter of national preference. It reflects a calculated effort by Chinese system integrators to de-risk supply chains, capture more value domestically, and build technical depth in areas where export controls are less binding. Peripheral chips are a natural starting point because they're less likely to trigger licensing restrictions and because the design and manufacturing know-how can be developed incrementally.

For the broader Asian semiconductor industry, SMIC's pricing posture is a signal that foundry economics are fragmenting. While leading-edge fabs in Taiwan and South Korea compete on transistor density and power efficiency, mature-node fabs in China are carving out defensible positions in volume segments tied to AI infrastructure. The two worlds are increasingly decoupled, each with its own demand drivers, margin profiles, and competitive dynamics.

As China's AI deployment deepens and localization pressures intensify, peripheral chips may prove to be the quiet backbone of the country's semiconductor strategy, less visible than cutting-edge logic but no less critical to the stack. SMIC's willingness to hold the line on pricing suggests it believes that story has years, not quarters, left to run.