Huawei’s Ascend 910B chips capture 15% of China’s AI accelerator market amid Nvidia’s licensing delays, as U.S. export controls accelerate Beijing’s push for semiconductor self-reliance.
Nvidia executives warned U.S. lawmakers in a June 18 closed-door briefing that Huawei’s chiplet integration technology could render current export controls ineffective by 2026, according to Reuters sources.
U.S. Sanctions Spark Parallel Innovation Ecosystems
Huawei shipped 400,000 Ascend 910B units in Q2 2024 according to TechNode’s June 19 report, doubling Q1 volumes despite SMIC’s 7nm production constraints. This follows Beijing’s $22 billion 2023 investment in chip R&D through the National Integrated Circuit Industry Investment Fund.
‘What we’re seeing is the law of unintended consequences in action,’ Dr. Li Zhang, semiconductor analyst at TechInsight Asia, told CNBC. ‘The 2022 export controls forced Chinese firms to develop domestic alternatives – now Huawei’s chiplet architecture allows them to bypass yield limitations through advanced packaging.’
Open-Source AI Fuels Hardware Localization
China’s DeepSeek R1 model, boasting 1.6 trillion parameters, has driven 300% year-on-year demand for domestic AI accelerators according to CSIA data. Singapore-based NeuReality announced on June 21 it’s replacing Nvidia A100 GPUs with Huawei’s Ascend platform in its inference servers, projecting 40% operational cost savings.
Meanwhile, Nvidia’s China-specific H20 GPU faces adoption hurdles with only 10% market penetration as of June 17 per Reuters. Delays in obtaining U.S. Commerce Department licenses for cloud service providers have complicated deployment timelines.
Regional Players Navigate Tech Decoupling
Taiwanese chip designers like MediaTek are discreetly collaborating with Huawei on 5nm AIoT processors, leveraging chiplet designs to circumvent U.S. restrictions according to Digitimes sources. This comes as Iron Mountain activates its $200 million hyperscale data center in Johor, Malaysia on June 20 – the first of six planned ASEAN facilities emphasizing vendor-neutral AI infrastructure.
The strategic shifts follow Nvidia CEO Jensen Huang’s May warning to shareholders: ‘We’re witnessing the emergence of separate technology stacks. The next generation of AI infrastructure will be defined by regional compliance requirements as much as technical capabilities.’
Historical context shows China’s semiconductor push mirrors its successful 2015-2020 5G infrastructure strategy, where U.S. exclusion from Huawei equipment markets propelled domestic alternatives. Similarly, the 2018 ZTE sanctions crisis prompted Beijing to accelerate chip investments, leading to SMIC’s 7nm breakthrough in 2022. Analysts note that current AI chip developments follow this pattern of external pressure catalyzing domestic innovation.
The ASEAN data center boom also recalls the 2020 wave of neutral cloud infrastructure investments during the U.S.-China trade war. Google and Amazon’s $2 billion Jakarta cloud regions, announced in 2021, similarly positioned Southeast Asia as a neutral ground for competing tech ecosystems – a strategy now extending to AI hardware deployment.
