15 AI Chip Firms Compete for SMIC's 7nm Capacity, Shortage Could Reach 1.6 Million Wafers
Claire Weston
Fifteen Chinese AI chip companies are competing for slots on SMIC's only 7nm-class production line; the projected 2026 shortfall is roughly 1.6 million chips — nearly one in every two demanded — and allocation has shifted from market logic to political priority.
What is the real bottleneck for China's AI chips?
Not design capability — manufacturing capacity. SMIC's N+2 line (7nm-class process) is the only advanced node in China that can mass-produce AI chips at scale.
IDC data: China shipped about 4 million AI accelerators in 2025; Nvidia supplied 2.2 million. But Nvidia CEO Jensen Huang has said repeatedly that the company's China business has fallen to effectively zero.
This means → replacing Nvidia depends entirely on domestic fabs, and the production line that can do the job is a single facility — SMIC's southern plant in Shanghai.
How large is the supply gap?
Projected 2026 demand for Chinese AI chips: roughly 4.2 million units. SMIC's N+2 estimated annual output: about 2.6 million. The gap: approximately 1.6 million chips.
In plain terms = for every two chips needed, nearly one cannot be made.
SMIC's target is to raise N+2 monthly capacity to 70,000 wafers by end of 2026, but 15 firms — including Huawei, Cambricon, Hygon, Alibaba's T-Head, Biren, and Moore Threads — are queuing for the same line.
Why is capacity so slow to expand?
SMIC cannot obtain EUV lithography tools — the highest-precision machines that etch circuits with extreme-ultraviolet light. It relies instead on DUV (deep-ultraviolet) multi-patterning, exposing the same layer multiple times to achieve finer resolution.
This means → each wafer takes several times longer to produce than at TSMC or Samsung, and lower yields drive up costs further.
Industry sources say even with unlimited funding, building an equivalent line — from equipment procurement through process tuning to yield ramp — would take at least three to five years. The capacity ceiling is effectively locked in the near term.
Who gets the allocation?
Huawei takes nearly half of N+2 capacity. It is both the strategic centerpiece of China's push to break through tech restrictions and a deep participant in developing SMIC's advanced process.
Cambricon receives about 10%, backed by its role in the national AI-chip narrative and its Beijing ties. Hygon's CPU business is tightly bound to the "Xinchuang" strategy — China's push to replace foreign IT infrastructure.
In plain terms = the front-of-line slots are decided by political priority and state-capital backing, not by whose chip design is superior.
What happens to the rest?
T-Head, Biren, Moore Threads, and the remaining chip firms must split less than half of the leftover capacity.
According to Digitimes, the three factors driving allocation are, in order: national strategic priority, state-capital shareholding, and local-government informal quota mechanisms.
This reflects a deeper shift: the real contest in China's AI chip race is no longer about technology — it is about who secures manufacturing access. Most seats were assigned before market competition even began.
Content is for reference only, not financial advice.