SMIC Warns AI Chip Supply Could Stay Tight Through 2027

China’s largest contract chipmaker says the squeeze on AI-related semiconductor supply could last through 2027, even as demand remains strong. SMIC co-chief executive Zhao Haijun made the assessment during the company’s Friday, 13 November earnings call, arguing that the AI industry is still in an early phase. For technology companies planning new data centres, the warning is that access to chips and manufacturing capacity may remain a constraint long after the current rush to order equipment.

Zhao said customers seeking computing resources still face shortages, particularly as data-centre and private-cloud projects expand. Reuters reported his comments after SMIC published its third-quarter figures on Thursday. He also said memory supply and demand could move closer to balance around the middle or end of 2028. Those are management forecasts, not a guarantee that every chip category or region will face the same conditions.

The latest SMIC results show third-quarter revenue of $3.19 billion, up from $2.38 billion a year earlier, with gross margin rising to 29.9%. The company expects revenue to grow another 4% to 6% sequentially in the fourth quarter, although it projects a lower gross-margin range of 26% to 28%. The figures show healthy demand for its foundry services, but they do not mean SMIC alone can satisfy the wider AI industry’s needs.

Manufacturing capacity cannot be expanded simply by placing more orders. Zhao said some equipment now carries lead times exceeding 24 months. New tools must be delivered, installed and brought up to reliable production yields before customers receive usable chips. SMIC also expects depreciation expense to rise about 30% year on year in 2026, a reminder that capacity growth brings substantial fixed costs.

The bottleneck reaches beyond the most advanced AI processors. Servers need memory, networking components, power management and mature-node chips as well as accelerators. A shortage in one part can delay a completed system even when other components are available. That helps explain why an apparently strong earnings quarter for a foundry can coexist with persistent customer shortages.

The comments arrive as investors question whether AI infrastructure spending is sustainable. Zhao’s answer was that demand remains real, but high demand is not the same thing as an unlimited supply of deployable computing systems. Foundries, memory makers and equipment suppliers face different production cycles, and their constraints do not clear at the same speed.

For China, SMIC is particularly important because restrictions on advanced chipmaking tools and US-origin AI processors have increased the value of domestic manufacturing. Yet the company’s results should not be read as proof that it has caught up with the world’s leading foundries in every process technology. The confirmed story is a strong quarter and management’s expectation of a prolonged supply squeeze. TechBooky has previously examined Huawei’s effort to build a more competitive Chinese chip ecosystem, which makes manufacturing capacity and component availability central to the industry’s next phase.

Customers deciding whether to build or lease AI capacity should watch the next production updates, not just the headline demand forecast. Equipment delivery, utilisation, yields and memory availability will determine how quickly new spending becomes working compute. SMIC expects the market to stay tight through next year. The practical implication is that the AI boom may be limited as much by the speed of the physical supply chain as by the amount of money companies are willing to spend.