Silicon Curtain: How US Sanctions Are Forging China's Tech Sovereignty
Bappa Sinha
IN a bold challenge to US tech supremacy, Huawei has unveiled its most ambitious AI computing platform yet: the CloudMatrix 384, a high-performance AI training cluster based on its homegrown Ascend 910C chips. The platform is built to train large-scale AI models like those behind ChatGPT or Deepseek, which require enormous computing capacity. What makes the announcement significant is not just the technological advance but the context: it represents China’s most concerted and defiant push yet to build a parallel AI stack, independent of US technology.
Significantly, the announcement came as the US doubled down on export restrictions on China’s access to AI chips like the Nvidia H20 – a product already engineered to be less potent to comply with prior US sanctions on Nvidia’s advanced H100 chips. At stake is more than just silicon – it’s the future of AI power, economic sovereignty, and global digital infrastructure.
Huawei's new CloudMatrix 384 doesn’t just rival Nvidia’s top-tier GB200 NVL72 cluster – it’s designed to replace it. It integrates 384 of Huawei’s Ascend 910C chips and is reportedly designed to deliver superior overall system performance in key metrics like total compute capacity.
The CloudMatrix 384 surpasses Nvidia’s current comparable system solution by more than 67 per cent. This is achieved through system-level innovation and reflects a different calculation of priorities: for example, the CloudMatrix has more than five times the number of individual AI chips than Nvidia’s NVL72 cluster (384 Ascend 910C chips compared to just 72 Nvidia GB200 chips). Each Ascend 910C chip has 1/3rd the computing power of the Nvidia GB200 chip. So, the fact that Huawei is a generation behind Nvidia chips is more than compensated for at the system level using innovations in the networking, optics, and software layers. This pattern of bypassing limitations through clever engineering workarounds is visible across the board.
This technological leap has not occurred in a vacuum. Over the past three years, the US has implemented successive waves of export controls to block Chinese firms from accessing advanced chips, especially for high-end smartphones and AI use. These controls have prevented Chinese firms from buying many advanced products of Nvidia - the global leader in AI chips. In response, Nvidia created the H20, a downgraded chip designed to comply with US rules. But now, even that has been blacklisted. The siege is not just restricted to advanced chips. Since 2022, the US has weaponised the entire semiconductor supply chain, using its chokehold on key nodes – design software, lithography and the entire range of wafer and chip manufacturing tools, and high-end CPUs and GPUs (graphics processing unit) chips – to slow China's technological progress.
Washington thought it could cripple Beijing’s chip ambitions through export bans, blacklists, and licensing hurdles. What actually happened was the opposite. China launched the most intense industrial mobilisation in recent history. Far from kneecapping China, the sanctions lit a fire under China’s self-reliance drive. US restrictions have accelerated China’s push for full-spectrum self-reliance. What was once fragmented is now a nationally backed, vertically integrated ecosystem. In 2023, Huawei introduced its Mate 60 smartphone which shocked the US tech industry. It was powered by a 7nm chip from SMIC, a blacklisted Chinese company without access to the most advanced EUV Lithography machines from the Dutch company ASML. The production of such advanced chips was considered impossible at that time without the EUV machines. Yet, SMIC had figured out technical workarounds to produce these chips using ASML’s less advanced DUV machines. This story is repeating across the whole range of products comprising the semiconductor value chain as China is pouring over $500 billion into high-tech infrastructure. AI accelerators, 3D NAND, high-bandwidth memory, satellite networking, and photonic chips are all in full-throttle development.
Paradoxically, Nvidia’s CEO, Jensen Huang, recently warned against pushing China too hard. In a moment of candour, risking Trump’s wrath, he labelled the US export control policy a "failure." His reasoning was simple: denying US firms access to the vast Chinese market (a $50 billion opportunity for AI chips alone by 2026) doesn't halt China's AI development. Instead, Chinese firms will "just spend that money elsewhere, and rivals like Huawei will fill the gap." Haung is worried that the lucrative China market could be forever lost to US companies as companies like Huawei increasingly start becoming competitive. Nvidia has already written off a $5.5 billion financial loss from stranded inventory in H20 AI chips for the China market after the Trump administration extended restrictions to include those semiconductors.
Haung’s predictions have already come true, as illustrated by the release of Deepseek – an AI model at par with the most advanced AI models from the US, like ChatGPT and Claude. Deepseek could develop its model at a fraction of what it cost companies like OpenAI and Anthropic, while not having access to the most advanced AI chips from Nvidia. They were able to do this using the now familiar pattern of ingenious engineering workarounds that used lower-powered chips more efficiently than the brute force way in which US companies were utilising their more advanced chips. Apparently, the US engineers were spoiled by their abundance, while the Chinese, hamstrung by restrictions on the most advanced chips, were forced to innovate in software.
Huawei is dancing to its own beat as well. The company is preparing to mass-produce its Ascend 910C chips, which are already being integrated into Chinese data centers. While these chips don’t yet match Nvidia’s H100 or GB200 in raw power, they are already working on their next generation offering, Ascend 910D, aiming to outperform Nvidia's advanced H100.
Huawei is not stopping at just going head to head with Nvidia. Along with its subsidiaries, it is aiming for self-reliance across the entire semiconductor supply chain, including in chip design software (EDA tools), chip manufacturing equipment, lithography machines, and, of course, in the final chips and the products they go into like high-end smartphones. This scope and ambition on the part of a single company is unprecedented in the highly complex semiconductor space. And Huawei is just one company, albeit an extremely crucial one, China has a whole host of such companies trying to completely indigenise the entire semiconductor supply chain.
What we’re witnessing is not just technological competition – it's infrastructural bifurcation and at stake is the question of who gets to shape the global digital future. The US is trying to slow down China’s tech progress, fearing the loss of whatever technological edge it retains. China is doubling down on innovation and indigenisation. If China can build and export its own tech infrastructure, the vast Chinese market which accounts for 40 per cent of all chips produced globally will be forever closed to Western companies. Chinese products will also become competitive enough to surpass those made by US companies in the global market, as has been the case with electric vehicles and solar technology.
From a broader perspective, these developments echo past tech struggles – from the crackdown on Huawei’s 5G services and access to all advanced chips to the restrictions on TikTok – all in the name of “national security.” But behind this narrative lies a deeper contest: the US wants to maintain its monopoly over the digital tools that will define the 21st century. Meanwhile, China is determined to carve out a sovereign technological path, even if it means enduring short-term setbacks.
The lessons for the Global South are clear. As AI becomes a critical foundational technology for modern industry and economy, the power to build, control, and distribute its infrastructure is shaping up to be the key geopolitical faultline of our time. Countries like India, Brazil, South Africa, and Indonesia must decide: will they remain consumers of someone else’s digital empires, or will they invest in local capacity, regional alliances, and open technologies?
or reload the browser