SMIC expects AI chip supply to stay tight through 2027 as equipment delays stretch

SMIC expects demand from data centers and private clouds to keep AI-related semiconductor supply tight through 2027, co-CEO Zhao Haijun said November 13. The squeeze reaches beyond computing resources: demand for supporting chips has exceeded management’s expectations, while some equipment orders take more than two years to arrive, complicating the company’s capital spending.

Zhao delivered the outlook while answering a question about whether AI was forming a bubble, Reuters reported. His response emphasized an industry still in its early stages: demand remains strong, and many customers seeking computing resources still face shortages. That was his assessment of current demand, not a guarantee of future growth.

The pressure includes chips that support AI computing rather than perform its main calculations. Zhao identified power-management parts, specialty memory and controllers as categories where demand had surpassed initial expectations, according to Startup Fortune’s account of SMIC’s third-quarter call, which cited an Investing.com transcript.

Some power-management products already have orders visible through the end of 2027, Zhao said. These are BCD chips, which combine analog, power and logic circuits on one piece of silicon. They help feed power to processors on AI server boards, making them supporting components rather than substitutes for the processors themselves.

SMIC’s factories were also running closer to full use. The company reported third-quarter utilization of 97.3%, compared with 93.7% in the second quarter. That measures how much manufacturing capacity was being used; it is a completed quarterly result, unlike Zhao’s forecast for the wider supply chain.

The supply constraint extends to the machinery SMIC buys. Some equipment orders take more than 24 months to deliver, Zhao said, affecting the company’s capital spending in 2026. That puts a separate timing constraint on spending, even while customers are asking for more chips.

The spending effect does not necessarily mean an equivalent delay in near-term output. Startup Fortune, citing Investinglive, said equipment delivery times could shift the timing of capital spending without changing SMIC’s planned 2026 capacity additions. The distinction is between when money is spent and when production capacity becomes available.

Zhao gave memory a later potential balancing point: supply and demand should reach balance around the middle to end of 2028. That forecast is specifically about memory chips. It should not be read as a promised date when every part of the AI semiconductor supply chain will have caught up.

SMIC’s financial results show the demand backdrop in company-level terms. Third-quarter revenue reached $3.19 billion, up 6.1% from the previous quarter. Gross margin was 29.9%, above its own 26%–28% guidance and higher than the second quarter’s 25.3%, according to figures reported by Startup Fortune.

Management described phased price increases and a better product mix, while stressing longer-term customer planning rather than single-quarter metrics. It did not provide one overall figure for quarterly price changes. Zhao also said SMIC needed to negotiate more reasonable pricing because its wafer prices remained well below industry leaders’ levels.

For the fourth quarter, SMIC expects revenue to rise another 4%–6% from the third quarter, with gross margin of 26%–28%. Costs are rising too: Zhao said depreciation expense would increase about 30% in 2026. The outlook pairs further sales growth with a margin range below the latest quarter’s result.