DeepSeek V4 Will Run on Huawei Chips. The AI Chip War Just Got More Complicated.
The US government has spent two years restricting Nvidia chip exports to China, betting that cutting off access to advanced semiconductors would slow Chinese AI development. DeepSeek’s announcement that its next model — DeepSeek V4 — will run on Huawei chips suggests that bet may not be paying off the way Washington hoped.
What This Signals
Huawei’s Ascend chips were considered significantly behind Nvidia’s H100s and A100s in AI training performance. If DeepSeek can build a frontier-competitive model on Huawei hardware, one of two things is true: either Huawei’s chips have improved more than the public benchmarks suggest, or DeepSeek has developed training techniques efficient enough to work around the hardware disadvantage.
Both possibilities are concerning for the US strategy of maintaining AI supremacy through chip export controls.
The Efficiency Story
DeepSeek has consistently punched above its weight in efficiency. Their previous models achieved competitive performance at a fraction of the compute cost of Western equivalents. If they’ve applied the same engineering discipline to working around hardware limitations, the export control strategy has a significant leak.
The Buccaneer Take
The AI chip war was always going to be a temporary advantage at best. Hardware restrictions slow development — they don’t stop it. DeepSeek running on Huawei chips is a data point that the gap is closing faster than the export control strategy assumed. Watch the V4 benchmarks when they drop. That’s the real test. 🏴☠️