The Chip War in 2026: How China Is Routing Around US Semiconductor Sanctions

When the Biden administration imposed sweeping export controls on advanced semiconductors in October 2022 — and expanded them through 2023 and 2024 — the stated goal was to slow China’s AI development by cutting off access to NVIDIA’s A100 and H100 GPUs.

The strategy has not worked as intended.

The Efficiency Route

China’s AI labs responded to chip scarcity not by stalling, but by innovating around the constraint. DeepSeek’s training of V3 on older H800 chips (a downgraded export variant) for under $6 million proved that algorithmic efficiency could compensate for hardware disadvantage.

This approach — training smarter, not bigger — has become a competitive advantage. Chinese models routinely match or beat American counterparts trained on vastly more expensive compute.

Domestic Semiconductor Progress

SMIC, China’s leading chipmaker, has made significant advances in 7nm and 5nm process technology despite being cut off from ASML’s extreme ultraviolet (EUV) lithography machines. While trailing TSMC’s cutting edge, SMIC’s progress has been faster than most Western analysts projected.

Huawei’s Ascend 910B AI accelerator chip, manufactured at SMIC, is now being deployed in Chinese data centers as a domestic substitute for NVIDIA GPUs. Performance benchmarks remain below NVIDIA’s H100, but the gap is narrowing.

The Geopolitical Paradox

US export controls were designed to maintain American AI leadership. The unintended consequence: Chinese labs focused on efficiency and openness — areas where they now lead — while US labs continued scaling compute-intensive proprietary models.

The chip war has accelerated Chinese AI development in ways that pure investment never would have. Necessity, it turns out, is the mother of innovation.

With information from Semiconductor Digest and Reuters.