China Unveils $70B AI and Chip Subsidy to Counter US Controls

China has unveiled a subsidy package for semiconductors and AI worth up to $70 billion, announced during the annual Two Sessions parliamentary meetings that began March 4. The program is one of the largest government semiconductor support initiatives in history, dwarfing Beijing's previous Big Fund efforts and directly targeting self-sufficiency in advanced chips as US export controls expand to cover global sales.

Scale and Structure

The $70 billion figure encompasses direct subsidies, tax incentives, low-interest loans, and government-backed venture capital for semiconductor and AI companies. The package builds on China's three prior rounds of the National Integrated Circuit Industry Investment Fund (the "Big Fund"):

The 2026 package goes beyond chip manufacturing to cover the full AI stack: chip design, manufacturing equipment, EDA tools, packaging, memory, and AI model development. It includes a 50% domestic equipment mandate - state-funded fabs and data centers must source at least half their equipment from Chinese suppliers - and a blanket ban on foreign AI accelerators in government-funded data centers. Companies that reduce dependence on US-origin technology in their supply chains receive additional tax incentives.

Who Benefits

The primary beneficiaries are China's national champion semiconductor companies:

• SMIC - China's largest foundry, currently producing at 7nm using DUV lithography workarounds. The subsidy package includes funding to push SMIC+Hua Hong consolidation to create a single national champion foundry with the scale to compete with TSMC. Additional funding targets 5nm-equivalent processes without EUV equipment.
• Huawei HiSilicon - Designs the Ascend AI accelerators used by DeepSeek and other Chinese AI companies. Subsidies help offset the cost disadvantage of producing on domestic fabs rather than TSMC.
• CXMT (ChangXin Memory Technologies) - China's leading DRAM manufacturer, currently pursuing a $4.2 billion IPO. CXMT is targeting HBM3 (high-bandwidth memory) production by end of 2026 - the type of memory critical for AI training clusters that China currently cannot produce domestically.
• DeepSeek, Baidu, Alibaba, ByteDance - Chinese AI labs that have been forced to train on Huawei Ascend and Cambricon hardware rather than Nvidia GPUs. Subsidies could lower the effective cost of domestic compute.

The Strategic Calculus

The timing is pointed. The US Commerce Department drafted rules this week that would require government permits for all AI chip exports worldwide - not just to China. China's subsidy announcement reads as a direct response: if the US is going to restrict chip access globally, China will build its own.

The existing US export controls have already forced Chinese AI companies onto domestic hardware. DeepSeek trained its V3 model on Huawei Ascend chips rather than Nvidia. But the performance gap remains significant - Huawei's Ascend 910C is roughly competitive with Nvidia's A100 but falls well short of the H200 and GB300 generations.

$70 billion is the kind of investment that could close that gap over a 3-5 year horizon. Xi Jinping pledged a "whole-nation" approach to semiconductor self-sufficiency during the Two Sessions, setting explicit targets: 70% AI chip integration into domestic infrastructure by 2027, domestic EUV lithography prototypes by 2028, and 70% overall chip self-sufficiency by 2030. For comparison, TSMC's total capital expenditure in 2025 was roughly $36 billion. China is committing nearly double that in subsidies alone.

Market Implications

Analysts have flagged two risks for Western semiconductor companies:

Price competition. Subsidized Chinese chips could undercut Western alternatives on price, especially in markets where cutting-edge performance is less critical than cost - consumer electronics, automotive, IoT, and mid-tier data center applications.

Market share erosion. If Chinese foundries achieve process parity at 5nm or below within the subsidy window, companies like TSMC, Samsung, and Intel could lose customers in price-sensitive segments. The subsidy effectively socializes the cost of China's learning curve, allowing domestic fabs to compete at prices that wouldn't be viable without government support.

The counterargument: China has spent over $100 billion on semiconductor subsidies since 2014 and has not yet produced a competitive EUV lithography system or matched TSMC's leading-edge yields. Money alone hasn't solved the problem. Whether $70 billion more changes that depends on whether the bottleneck is funding or fundamental technical barriers that subsidies can't shortcut.

The subsidy announcement reframes the AI chip competition from a trade dispute into an industrial policy race. The US is trying to control where chips go. China is trying to ensure it doesn't need them from the US at all. Both strategies have significant execution risks - the US approach could alienate allies and lose market share; China's approach has a long history of expensive failures. The question is which bet pays off first.

Sources:

• China Prepares as Much as $70 Billion in Chip Sector Incentives - Yahoo Finance
• China Announces Aggressive AI and Semiconductor Subsidy Program - YourNews
• China's Two Sessions: What It Means for AI and Chips - SCMP
• China's Semiconductor Ambitions and the Big Fund - CSIS