🚨 China has just manufactured a chip that outperforms NVIDIA by up to 478 times on specific AI tasks.

Researchers at Peking University have developed a neuromorphic chip capable of reconstructing complex brain surfaces in less than half a second.

The key difference from standard GPUs lies in the architecture.

Traditional chips like Nvidia's A100 separate memory and processing into two independent units, creating a bottleneck every time data moves between them.

This chip eliminates that bottleneck by performing storage and computation simultaneously within the same memory array.

This is important because that bottleneck is one of the main reasons why AI training today is so expensive and power-hungry.

Nvidia's dominance is built on solving that problem with brute-force hardware and massive memory bandwidth.

This chip solves it in a different way, at the architectural level.

The 478 times figure applies to a specific task: real-time neural surface reconstruction.

This is not a claim about general performance. On most standard AI workloads, Nvidia's H100 and B200 chips are still far ahead.

But the comparison with the A100 is important because the A100 is still the chip that most enterprise customers are actually running. Nvidia's latest chips are out of reach for most buyers.

If a 40-nanometer chip manufactured under China's restricted export environment can outperform the A100 by such a margin on targeted workloads, it does raise a real question: how necessary is access to cutting-edge chips for specific AI applications.

This is precisely the argument DeepSeek made in January, that matching cutting-edge AI performance is possible with better architecture alone, without the need for frontier hardware.

That announcement wiped $600 billion off Nvidia's market value in a single day.