Nvidia to ship China AI chip by year-end, domestic share 41%

Nvidia plans to begin shipping small volumes of an AI chip designed for Chinese customers by year-end, The Information reported Thursday, as domestic rivals have captured 41 percent of China's AI accelerator server market.

The report, citing two Nvidia employees, follows Financial Times reporting that H200 chips have already reached China in small shipments after the US Bureau of Industry and Security moved the processor to case-by-case review in January 2026.

Chinese GPU and AI chip makers captured nearly 41 percent of China's AI accelerator server market in 2025, according to IDC data cited by Reuters, while Nvidia still led with about 55 percent — a sharp drop from roughly 95 percent before sanctions. TrendForce forecasts show domestic suppliers led by Huawei's Ascend line and Cambricon set to take a larger share of China's AI server market in 2026.

The move reflects Nvidia's effort to hold ground in a market where export controls have pushed Chinese cloud firms and chip designers to build roadmaps around domestic alternatives. SMIC's second-quarter net profit more than tripled to $479.2 million while Hua Hong's profit soared 385.9 percent, showing the scale of the shift toward local foundries.

Export Controls Reshaped China's Chip Market

The US Bureau of Industry and Security eased part of the AI chip regime in January 2026 by moving Nvidia's H200, AMD's MI325X and similar chips to case-by-case review for China and Macau, subject to strict conditions. A separate presidential proclamation put a 25 percent tariff structure around covered advanced computing chips, and policy analysts at CNAS noted that shipments to China are capped against US domestic sales.

On paper, that gives Nvidia and AMD a controlled path back into China. In practice, it does not unwind three years of planning around risk. Chinese cloud firms and chip designers have had every reason to avoid building roadmaps around a license that can change with the next rule. Huawei's Ascend line, Cambricon, SMIC, Hua Hong and other domestic suppliers do not need to win every benchmark to benefit from that uncertainty — they only need to be available when the purchase order is due.

Domestic Foundries Scale on AI Demand

The shift extends beyond accelerators. SMIC, China's largest foundry, reported second-quarter net profit of $479.2 million, more than triple year over year, with revenue up 36 percent to $3.01 billion and gross margin at 25.3 percent. Its utilization rate reached 93.7 percent. Hua Hong Semiconductor's net profit jumped 385.9 percent to $38.6 million on record revenue of $717.5 million, with wafer shipments up 17.9 percent year over year to 1.538 million 8-inch equivalent wafers.

AI systems do not run on accelerators alone. They need power-management chips, image sensors, radio-frequency components, controllers and networking parts — the mature-node layer that Chinese foundries specialize in. Hua Hong guided for third-quarter revenue of $770 million to $780 million with gross margin between 16 percent and 18 percent. SMIC forecast a further 2 percent to 4 percent revenue increase and gross margin of 26 percent to 28 percent.

Hua Hong is also consolidating capacity. In July, China's securities regulator approved Hua Hong Grace's plan to buy 97.4988 percent of Shanghai Huali Microelectronics through a share issue, a transaction priced at 8.268 billion yuan, or roughly $1.15 billion. Huali brings 12-inch foundry lines focused on 65/55-nanometer and 40-nanometer processes — exactly the kind of mature and specialty capacity now benefiting from AI-related demand.

Nvidia's China-designed chip, if it ships by year-end, would face a market where domestic alternatives have already established supply chains and customer relationships. The company's shares have priced in the export-control drag, but a successful China product could reopen a market that once accounted for roughly a quarter of Nvidia's data center revenue. Samsung's foundry business, meanwhile, has raised prices 10 to 15 percent on new orders for its 4-nanometer process as AI demand tightens capacity, showing the broader pricing power across the semiconductor supply chain.

This article is for informational purposes only and does not constitute investment advice.