China’s AI boom, self-sufficiency push fuel earnings surge for local chip firms

The sharp profit rise comes from two forces working together. Chinese companies are supplying chips, memory, optical parts and materials for expanding AI infrastructure. At the same time, China’s push for self-reliance is encouraging customers to replace foreign processors and equipment with domestic alternatives. This creates stronger demand for local suppliers.

Yuanjie Semiconductor Technology expected quarterly net profit to rise as much as 660 per cent. It linked shipments and high margins to data-centre demand. GPU designer MetaX expected to move from a loss to a profit as orders increased. Hygon Information also cited rising demand for high-end domestic chips.

The gains are spreading beyond chip designers. Accelink benefits from AI-related optical-module spending, while Longsys benefits from higher DRAM and NAND prices. Quartz-material and silicon-wafer suppliers are also seeing stronger orders, although not every company is profitable without one-off gains.

The reported gains are unusually large, but they differ by company and measure. Yuanjie Semiconductor Technology forecast quarterly profit growth of 579 per cent to 660 per cent. Longsys expected nine-month profit of 14 billion to 14.9 billion yuan, about 20 times the previous year. Other companies reported revenue or profit growth rather than comparable profit multiples.

The strongest examples span the supply chain. MetaX turned profitable, while Hygon projected revenue growth of 50 per cent to 64 per cent. Accelink forecast profit growth of 65 per cent to 82 per cent. Feilihua expected 35 per cent to 56 per cent growth, and Xian Eswin reported wafer sales up about 43 per cent.

The gains reflect several mechanisms. AI and data centres increase chip and optical demand. Memory prices raise profits. Import replacement supports domestic processors and materials. Yet Xian Eswin still expected an adjusted loss after excluding investment gains.

A GPU, or graphics processing unit, is a chip designed to handle many calculations in parallel. Unlike a general-purpose processor, it uses large numbers of smaller computing units for tasks that can be split into repeated operations. This makes GPUs useful far beyond graphics, especially for artificial intelligence.

AI systems process huge amounts of data and repeatedly perform mathematical operations. GPUs can carry out many of those operations simultaneously. In data centres, groups of GPUs can therefore support the training and operation of demanding AI models. The article identifies MetaX as a Chinese GPU designer whose increasing orders helped it turn profitable.

This matters because AI infrastructure requires more than software. It needs processors, memory, networking and optical connections. Chinese AI investment is increasing demand for these components. The article also reports that Chinese-designed chips could account for almost 90 per cent of China’s high-end AI processor market this year.

Chinese customers are increasingly choosing domestic processors because the country is pushing for greater self-reliance in technology. Replacing imported chips can reduce dependence on foreign suppliers and support a local ecosystem. The shift is especially important as demand for high-end processors grows through AI and data-centre construction.

Hygon Information, whose processors support general computing and AI workloads, projected first-nine-month revenue growth of 50 per cent to 64 per cent. It attributed that outlook to rising demand for high-end chips and customers increasingly selecting domestic alternatives over foreign suppliers. MetaX also linked its move into profit to growing orders.

The change is becoming large enough to reshape the market. TrendForce projected that Chinese-designed chips would represent almost 90 per cent of China’s high-end AI processor market this year. That forecast suggests domestic demand is supporting chip designers even as the wider industry remains exposed to supply, pricing and customer-spending cycles.

AI investment creates business across the semiconductor supply chain because an AI system needs many connected parts. More processors require memory to hold data, optical modules to move information, silicon wafers to make chips, and specialised materials for manufacturing equipment. Demand at the system level therefore spreads to suppliers that may not design the final processor.

The article gives several examples. Longsys benefited from higher DRAM and NAND flash prices. Accelink cited global AI spending and Chinese cloud providers’ data-centre investment. Xian Eswin ran production lines at full capacity, with wafer sales up about 43 per cent. Feilihua linked stronger semiconductor demand to quartz-component opportunities.

This expansion is not uniformly profitable. Longsys said price increases slowed in the third quarter, while Xian Eswin expected an adjusted loss despite investment gains. Still, domestic equipment and materials orders could keep spreading as China replaces imported products and builds more AI infrastructure.

A chip boom can slow without any collapse in AI demand. If manufacturers add enough capacity, supply may catch up with orders. Prices and profit margins can then weaken. Customers may also pause or reduce purchases after building inventories or completing major infrastructure projects. Either development could cool sales.

The article applies this warning especially to memory. Morgan Stanley said memory prices could remain firm through 2027, but warned that extra supply or slower customer buying could end the boom even without weaker AI demand. Longsys already reported that DRAM and NAND price increases slowed in the third quarter, giving profits a smaller boost.

This makes current forecasts vulnerable to market timing. Strong AI investment can lift several suppliers at once, but capacity expansion and purchasing cycles may change the balance. Companies with high margins, full factories or rapid growth could therefore still face pressure if supply rises faster than orders.

Semiconductor companies prosper when demand exceeds available supply, their factories can meet orders, and prices support healthy margins. Government-backed efforts to replace imports add another source of demand, especially for domestic processors, equipment and materials. But strong demand alone does not guarantee profit if production costs or capacity problems remain high.

Yuanjie benefited from data-centre shipments and maintained high margins. Xian Eswin ran its lines at full capacity and increased wafer sales, yet still expected an adjusted loss. Feilihua cited replacement opportunities for imported products, while Hygon benefited from customers choosing domestic high-end processors. These examples show how different parts of the chain respond to the same forces.

The balance can change quickly. Rising supply may weaken prices, and slower customer purchases may reduce orders. Longsys already saw memory-price increases moderate. Companies combining strong local demand, usable capacity and defensible margins appear best placed, but the article warns that the boom remains cyclical.