To address chip shortages, Chinese AI firms are diversifying their partnerships beyond traditional American suppliers, looking towards European and Asian chipmakers for support.

Transitioning from Nvidia's proprietary software to Huawei's solutions poses significant challenges for Chinese AI companies, requiring substantial time and resources.

The situation underscores China's vulnerability in securing advanced AI hardware, as the country imported $385 billion worth of semiconductor chips in 2024, reflecting a growing dependency.

In the first quarter of 2025, Chinese companies ordered between $12 billion and $16 billion worth of H20 processors, with substantial deliveries occurring prior to the sales ban.

Huawei Technologies is gearing up to mass-ship its advanced AI chips, the Ascend 910C and the upcoming Ascend 920, to Chinese customers, marking a pivotal moment in the AI hardware market amid ongoing U.S. export restrictions.

In response to these restrictions, Chinese companies are actively seeking alternatives to Nvidia, with Huawei's chips emerging as a viable option.

The Ascend 910C chip merges features from previous models to achieve performance levels that rival Nvidia's H100 chip.

The Ascend 920 chip, which follows the 910C, is produced using a 6nm process technology by Semiconductor Manufacturing International Corp (SMIC) and promises a significant performance boost.

With 900 TFLOPS of performance and 4000GB/s memory bandwidth, the Ascend 920 is expected to deliver a 30-40% increase in speed compared to its predecessors.

Huawei has improved its chip production yield from 20% to 40%, indicating greater efficiency and profitability in its operations.

Despite facing challenges due to sanctions, Huawei has managed to fabricate the Ascend 910C using local foundries and sourcing components through intermediaries, although this has attracted regulatory scrutiny.

While there is optimism about China's chip manufacturing capabilities, particularly with SMIC, uncertainties remain regarding advancements at smaller process nodes.