China’s Huawei develops new AI chip, seeking to match Nvidia

The steady advance by one of China’s flagship technology companies points to the resilience of the country’s semiconductor industry despite efforts by Washington to stymie it, including by cutting off access to some Western chip-making equipment.

Huawei has approached some Chinese tech companies about testing the technical feasibility of the new chip, called the Ascend 910D, people familiar with the matter said. The company is slated to receive the first batch of samples of the processor as early as late May, some of the people said.

The development is still at an early stage, and a series of tests will be needed to assess the chip’s performance and get it ready for customers, the people said.

Huawei hopes that the latest iteration of its Ascend AI processors will be more powerful than Nvidia’s H100, a popular chip used for AI training that was released in 2022, said one of the people. Previous versions are called 910B and 910C.

Huawei has emerged as China’s champion in a technology field where the U.S. remains ahead. The Shenzhen-based company has developed some of the country’s most promising substitutes for Nvidia’s AI chips. It is part of Beijing’s effort to groom a self-sufficient semiconductor industry.

Huawei, which has been on a U.S. trade blacklist for nearly six years, showed its ability to shrug off American restrictions by releasing a high-end smartphone in 2023. The model, the Mate 60, was powered by a locally produced processor and raised eyebrows within the U.S. government when it was introduced during a visit to Beijing by then-Commerce Secretary Gina Raimondo.

Earlier this month, Washington added Nvidia’s H20 chip—the most advanced processor the company could sell in China without a license—to a growing list of semiconductors whose sales are restricted there. Nvidia said it would take a $5.5 billion charge as a result.

The restrictions offer an opportunity to Nvidia’s Chinese rivals such as Huawei and Beijing-based Cambricon Technologies, which have developed similar chips.

This year, Huawei is poised to ship more than 800,000 Ascend 910B and 910C chips to customers including state-owned telecommunications carriers and private AI developers such as TikTok parent ByteDance, people familiar with the matter said. Some buyers have already been in talks with Huawei to increase orders of the 910C after the Trump administration restricted the exports of Nvidia’s H20s, the people said.

Despite manufacturing bottlenecks, Huawei and several Chinese chip firms have already been able to deliver some products comparable to Nvidia chips, albeit with a lag of a few years. Chip makers have been turning to technologies that can pack several chips together to create more powerful processors, as it gets harder and more expensive to make the circuitry inside chips smaller.

Beijing has also encouraged Chinese AI developers to increase purchases of domestic chips. State data centers have said most chips they used were from Chinese suppliers.

Still, previous Huawei chips have struggled to live up to their hype. The 910C was marketed to clients as comparable to Nvidia’s H100, but engineers who have used the two chips said Huawei’s performance fell short of its rival.

Huawei faces challenges in producing such chips at a significant scale. It has been cut off from the world’s largest chip foundry, Taiwan Semiconductor Manufacturing. China’s closest alternative, Semiconductor Manufacturing International, is blocked from purchasing the most advanced chip-making equipment.

Washington has also blocked China from directly accessing some key components for AI chips, such as the latest high-bandwidth memory units.

Given such constraints, Huawei executives have talked about focusing on building more efficient and faster systems to leverage their chips, instead of making individual chips more powerful.

In April, Huawei introduced the CloudMatrix 384, a computing system connecting 384 Ascend 910C chips. Some analysts said the system was more powerful than Nvidia’s flagship rack system, which contains 72 of Nvidia’s Blackwell chips, under some circumstances, even though the Chinese system consumes more power.

Connecting more chips in a system isn’t a trivial task. It requires stable networks as well as software and engineering to prevent network failures, industry practitioners said.

“Having five times as many Ascends more than offsets each GPU being only one-third the performance of an Nvidia Blackwell," research firm SemiAnalysis wrote in a report. “The deficiencies in power are relevant but not a limiting factor in China."

Write to Liza Lin at liza.lin@wsj.com and Raffaele Huang at raffaele.huang@wsj.com