Huawei says it trails its US rivals in chips but is closing the gap

For instance, cluster computing strategy allows it to compete at the server level on performance, said Shrish Pant, director analyst at Gartner. While the approach may be less efficient in terms of power consumption, it is effective for many applications.

“Performance on a chip level for Huawei’s 910C is roughly equivalent to Nvidia’s H100, and it can be a good alternative to now-restricted Nvidia’s H20 chips in China, provided the rest of the pieces fall in place,” Pant said. “Since Huawei cannot access the latest and greatest tech in semiconductor manufacturing yet, they are innovating in directions like non-Moore’s law approach, and one of the examples is architectural changes like joining two reticle-size GPU dies to double performance.”

The approach also aligns with Huawei’s strengths. In cluster computing, the key challenge often lies not in building large systems but in optimizing network performance across nodes to approach peak efficiency.

“Huawei is well known for its networking capabilities and may be using proprietary software-defined networking capabilities that can accelerate the cluster,” said Hyoun Park, CEO and chief analyst of Amalgam Insights.  “And there are mathematical tricks, such as the well-publicized DeepSeek use of an 8-bit floating point for training rather than the 16-bit version often used by most AI vendors.”

By simplifying model training and applying mathematical techniques that trade some accuracy for efficiency, Huawei could offset limited processing capabilities by relying more on power availability and software optimization.

Caution over sanctions

Huawei also has a vested interest in lowering expectations for its hardware on a global basis, as it is trying to avoid as many US and ally-based restrictions as possible.