In a clear move to replace NVIDIA as the dominant chip maker, Huawei is preparing to test a new artificial intelligence (AI) semiconductor that it touts as potentially rivaling NVIDIA’s top processors.

Huawei’s new chip, called the Ascend 910D, is fabricated by Chinese state-owned chip foundry Semiconductor Manufacturing International Corp (SMIC). Earlier versions of the 910 chip were fabricated by Taiwan Semiconductor Manufacturing Company (TSMC) until U.S. export controls blocked the partnership in 2020.

Huawei expects delivery of the 910D in late May, after which it will distribute the chip to Chinese tech firms to test its functionality in real world uses cases. These tests will be used to prepare the 910D for wider sales to business customers.

When the process is complete, preliminary reports suggest that that 910D may be a more advanced processor than NVIDIA’s H100, which debuted in 2022.

Yet some industry experts are skeptical of the Chinese firm’s efforts to replace NVIDIA. “Huawei’s efforts to catch up with—let alone unseat—NVIDIA as the world’s leading provider of high-end GPUs is in its very early stages,” said Nick Patience, VP & practice lead for AI, The Futurum Group. “NVIDIA remains many generations ahead of Huawei in this regard and is obviously not standing still. So although Huawei has a large, protected market for domestic AI chip production, there’s a lot of work needed to go from design to mass production of any chip. That’s a process that no chip company understands better than NVIDIA.”

Indeed, reviews of earlier Huawei Ascend chips were lackluster. The company promoted the earlier version, the 910C, as roughly equivalent to NVIDIA’s H100, but reports from tech experts in the field said the chip failed to impress.

As one of China’s leading tech giants, Huawei’s efforts reflect the growing Chinese focus on developing domestic sources of AI infrastructure. The Chinese tech sector has significantly lagged U.S. tech vendors in AI development in general and chip production specifically, and Huawei’s newest chip is an attempt to rectify this situation. The effort  is even more urgent now, given the current trade war between the U.S. and China, during which the Trump administration has placed greater export controls aimed at lessening China’s ability to develop AI.

“Crippling Huawei permanently might have worked had it not been for export restrictions that starved China of AI-enabling hardware and platforms,” said Oliver Blanchard, Research Director, the Futurum Group. “It isn’t surprising that the Chinese government and Huawei would have leveraged their previously successful relationship to address this problem. Export restrictions on ‘AI chips’ fueled Huawei’s resurgence, and is the reason why Huawei is now posing a threat to NVIDIA in China. Had we flooded China with NVIDIA chips, there would have been less of a need to prop up Huawei to become competitive with NVIDIA again.”

Huawei is enjoying commercial success with its Ascend chips. The company expects to sell more than 800,000 of the 910B and 910C semiconductors to clients ranging from ByteDance, which owns TikTok, to Chinese state-owned companies.

The 910C in particular is expected to find more buyers after the Trump administration blocked export of NVIDIA’s H20 chips. All of these sales are encouraged by the Chinese government, which is actively promoting sales of Chinese-made chips to domestic buyers.

Huawei is also producing advances in the so-called yield of its chips, meaning the percentage of functional chips produced by its facilities. (While SMIC fabricates the chips, Huawei packages them). Reports by the Financial Times indicate that Huawei has boosted its yield to 40%, double the percentage a year ago. The higher yield means that Huawei Ascend production line is now profitable, sources said.

Huawei is also working to produce more powerful systems by combining less advanced chips into networked packages, a growing trend in chip production. The company recently debuted the CloudMatrix 384, which connects 384 Ascend 910C semiconductors into a single device. According to research firm SemiAnalysis, “This solution competes directly with [NVIDIA’s] GB200 NVL72, and in some metrics is more advanced than Nvidia’s rack scale solution. The engineering advantage is at the system level, not just the chip level, with innovation at the networking, optics and software layers.”