China’s Advanced Chip Supply Gap Could Shrink to 34% by 2035

Goldman Sachs’ forecast points to a major reduction in China’s dependence on imported advanced semiconductors, not complete self-sufficiency. Domestic production of wafers made at 7-nanometer and smaller process nodes could grow at a 46% compound annual rate through 2035, well above projected demand growth of 17% annually. Even so, domestic supply would cover about 66% of demand by the end of the period, leaving a 34% shortfall. 2913

The headline numbers for 2035

Goldman’s model estimates that China’s monthly demand for 7-nanometer-and-below wafers will reach approximately 619,000 wafers in 2035. Domestic supply could rise to about 410,000 wafers per month, leaving a gap of roughly 209,000 wafers each month. 41014

That would represent a sharp improvement from 2025, when local production supplied about 8% of domestic demand and the shortfall stood at 92%. The projected shortfall falls to 34% by 2035 as supply growth outpaces demand growth. 21214

Measure	2025	2035 Goldman projection
Domestic share of advanced-wafer demand	About 8%	About 66%
Supply shortfall	92%	34%
Monthly domestic supply	—	410,000 wafers
Monthly domestic demand	—	619,000 wafers

The figures are projections rather than a guarantee of production. They describe what could happen if capacity expansion, manufacturing yields and customer qualification progress broadly as assumed in the model.

Why AI spending is driving the forecast

The investment case is closely tied to AI infrastructure. Goldman Sachs has raised its forecast for China’s semiconductor-industry capital expenditure to $82 billion by 2030, a 79% increase from its previous estimate. The bank expects double-digit annual growth in chip-industry investment through 2030.

Goldman also estimates that Alibaba, Tencent, ByteDance and Baidu will together spend about $102 billion on AI capital expenditure in 2026. That spending could support demand for domestic computing capacity and encourage foundries to add wafer capacity.

The important distinction is between demand and self-sufficiency. AI investment can create a powerful market for more chips, but it does not automatically solve the equipment, yield and process-control problems involved in manufacturing the most advanced wafers.

SMIC is the critical execution variable

The forecast assumes that Semiconductor Manufacturing International Corporation, or SMIC, will lead much of China’s advanced-node capacity buildout. Goldman’s assumptions include adding roughly 30,000 to 50,000 wafers of monthly advanced-process capacity each year from 2026 through 2031, followed by additions of about 20,000 wafers per month annually through 2035. 414

Capacity alone is not enough. A fab must produce a high proportion of usable chips at commercially viable cost. One report cited in the forecast analysis says SMIC would need to lift advanced-node yields from about 23% to 75%—roughly a tripling—to close the projected gap. 3

That makes yield improvement one of the biggest conditions behind Goldman’s estimate. The projection should therefore not be read as a claim that SMIC will match Taiwan Semiconductor Manufacturing Co. technologically. Rather, it is a scenario in which China expands capacity and improves manufacturing efficiency despite continuing equipment constraints. 8

Strong current utilization does not remove the bottleneck

SMIC’s recent operating momentum helps explain why analysts see room for a faster ramp. Reuters reported that the company’s monthly production capacity had reached 1.1 million 8-inch-equivalent wafers, with utilization at 93.7%; SMIC also added 8,000 wafers of monthly 12-inch capacity during the second quarter.

Those figures indicate strong overall demand and active capacity expansion. They do not, by themselves, establish that China can produce enough 7-nanometer-and-below wafers at competitive yields. Advanced-node output remains the narrower question, and the Goldman scenario still depends on process-specific improvements.

Lithography and export controls remain the main risks

Lithography is the central constraint in the forecast. China’s restricted access to leading-edge manufacturing equipment makes it harder to scale advanced processes and improve yields through the same equipment path available to leading producers elsewhere. 28

The practical risks are therefore interconnected:

• Equipment access: Export controls limit access to some leading-edge production tools.
• Yield improvement: More installed capacity will not translate into equivalent usable output unless yields rise substantially.
• Domestic substitution: Chinese equipment suppliers must increasingly replace restricted foreign tools and support reliable high-volume production.
• Demand execution: AI-related demand must remain strong enough to justify the projected investment and capacity additions.

Goldman’s conclusion is consequently conditional. China may narrow the advanced-chip gap dramatically, but the remaining 34% shortfall reflects unresolved manufacturing and supply-chain limits rather than a rounding error.

CXMT adds a separate memory challenge

Goldman’s advanced-wafer projection concerns logic-oriented 7-nanometer-and-below production. A related but distinct development is ChangXin Memory Technologies’ expansion in memory chips.

Goldman projects that CXMT could supply about 50% of China’s DRAM demand by 2028, subject to capital expenditure, yield improvement and customer qualification. Its monthly wafer capacity is projected to rise from about 270,000 in 2026 to 447,000 in 2028 and 665,000 in 2030. 510

CXMT’s HBM ambitions are more uncertain. One Goldman-linked projection puts HBM at 2% of CXMT revenue in 2026 and 27% in 2030, while another cited forecast says CXMT could supply about 40% of China’s HBM demand by 2028. These are future scenarios, not evidence that CXMT already matches Samsung Electronics or SK hynix in high-end HBM production. 10

Available industry data also shows the present gap: Counterpoint reported Samsung with 38% of global DRAM share in the first quarter of 2026, SK hynix with 29% and CXMT with 8%. CXMT’s projected growth could pressure Korean suppliers in China’s conventional-DRAM market, but it does not establish parity in advanced HBM technology or global supply.

What the projection means

The clearest takeaway is that China’s semiconductor strategy could shift from severe dependence toward substantial domestic coverage over the next decade. A 46% annual supply-growth rate would outpace the 17% annual growth Goldman expects for advanced-wafer demand, allowing domestic supply to reach roughly two-thirds of the market by 2035. 213

But “66% self-sufficiency” is not the same as technological independence. The forecast requires sustained AI investment, repeated SMIC capacity additions, much higher yields and progress in working around lithography restrictions. If any of those assumptions fall short, the remaining supply gap could be larger than Goldman projects.

For the semiconductor market, that makes the forecast significant in two ways: it signals a potentially much larger Chinese source of advanced-chip capacity, while also showing why the hardest part of the race remains the equipment-and-yield bottleneck rather than demand alone.