Little wonder, then, that every chip design firm wants to be the next Nvidia. The AI chip behemoth’s gross profit margins are a mouthwatering 75%, far better than software giant Microsoft’s 68%. In comparison, the average listed software-as-a-service (SaaS) — or enterprise software — firm has gross margins of under 75%. Indeed, Nvidia’s profits are so huge that any competitor that chips away at some of its market share would do just fine.
After Nvidia’s stock first went on a tear in early 2023, much of the commentary centred around how the AI chip trade might pan out.
The consensus was that Nvidia’s run would not last because just about every chip company on earth was probably spending every penny it had to come up with chip designs that were as good as, if not better than, any state-of-the-art AI chips such as the Blackwell GPUs that Nvidia could ship.
For now, Nvidia faces competitors on several fronts. For starters, it has rivals such as Advanced Micro Devices (AMD), which, like Nvidia, designs its own graphic chips such as the Instinct MI450 GPUs, which it will start shipping in the second half of next year. OpenAI recently sealed a deal with AMD that gives it the right to own a 10% stake in AMD in return for buying its chips and has another chip deal with database software firm Oracle. There are also smaller Silicon Valley-funded AI chip start-ups such as Groq, which is building specialised language processing units (LPUs) for low-cost AI inference, as well as SambaNova, Zero ASIC, Cerebras Systems, Habana Labs and Enflame.
See also: Qualcomm debuts chip to rival Nvidia in AI accelerator market
Then, there are customers or AI data centre chip users like Amazon, Microsoft, Google, OpenAI, Meta Platform and Anthrophic that want chips specifically tailored for their own use. They use chip firms such as Broadcom and Marvell Technology to get application-specific integrated circuit (ASIC) chips made for them. Since none of them — from Nvidia and AMD to Broadcom and Amazon, and other cloud service providers — have plants or wafer foundries of their own, the actual manufacturing of all AI chips is done by Taiwan Semiconductor Manufacturing Co (TSMC).
The most formidable competitors that Nvidia is likely to face are emerging from China. Already, several Chinese chip design houses have been touted as Nvidia clones. Among them are Cambricon Technologies, MetaX Integrated Circuits, Moore Threads Technology Co and Biren Technology. There is also Huawei Technology Co, the Shenzhen-based chip, smartphone and telecom equipment giant that has modelled itself after South Korea’s Samsung Electronics. And there are the big Chinese users of AI chips like TikTok’s owner Bytedance; e-commerce and cloud giant Alibaba Group; superapp, video gaming and music firm Tencent Holdings; and search powerhouse Baidu, which are taking a leaf out of the books of giant American users of AI chips to design customised chips for their own in-house use.
The Great Wall of Chips
See also: Anwar says chip tariff not an issue for now after Trump meet
Chips, of course, are at the core of the tech rivalry between Washington and Beijing. Their key role in AI and other areas with national-security and economic significance has made them the
focus of tensions between China and America — the world’s two superpowers.
China is nurturing its home-grown AI chip makers in part because, until recently, it imported most of its cutting-edge semiconductors, including AI chips. As a manufacturing powerhouse and the world’s second-largest economy, the country probably does not yet produce higher-end semiconductors such as AI chips at scale — one of the last major gaps in its industrial capabilities.
Moreover, since his first term in office, US President Donald Trump has used chips as leverage to force China’s hand on rare earths, which are used in key electronics such as smartphones, electric vehicle (EV) batteries as well as sophisticated defence and medical equipment. China controls 65% of global rare earth mine production and 85% of rare earths global processing. That, in turn, has led the US and its Western allies to attempt to build independent supply chains for rare earth minerals and magnets. As geopolitical tensions rise, it is no surprise that China would also build its own domestic supply chain of chips, particularly AI chips.
The race to develop a strong domestic base of chip design and manufacturing comes in the wake of Beijing’s battle against neijuan, or “involution”. It literally means “inward swirling” and refers to overcapacity or unnecessary business expansion that leads to cut-throat competition, excessive price wars and profit destruction. Instead of building more capacity in sectors where China dominates, such as EVs or solar panels, Beijing now wants to focus on expanding in areas where the country needs to catch up with competitors in the developed world. Semiconductor chips, where it has long lagged the US, are a key long-term focus for Chinese leaders.
In the aftermath of Covid-19, China’s once-breakneck growth pace has slowed because of a combination of challenges, such as a struggling real estate sector, weak consumer spending and deteriorating demographics, as well as US tariffs and global trade tensions. To kick-start growth, it has identified AI as one of the next engines of growth.
Nurturing homegrown Nvidia clones, however, will need a comprehensive ecosystem including foundries that are at par with Taiwanese chip giant TSMC, chip tool makers that can match Dutch giant ASML as well as electronic design automation (EDA) tools that are as good as those made by Cadence Design System or Synopsys. It might take China five years to get there, but if it has cutting-edge design houses like Cambricon and Huawei, an ecosystem will rapidly evolve.
In May, Nvidia CEO Huang lobbied Trump to relax the US ban on AI chips to China, arguing that the company’s market share there had dropped from around 95% about four years ago to around 50%. That’s down to zero. He attributed the sharp decline to US export controls on AI chips, which in turn had limited the products that Nvidia was able to sell in China. The Trump administration had allowed Nvidia to export a less powerful version of its AI chip, H20, to China. Huang believes the export controls not only damaged US businesses but also spurred Chinese competitors to accelerate their own development, as they were forced to use local options. China is one of two large AI markets in the world and, if Nvidia is shut out of China, it will lose a huge chunk of potential revenues and a captive user base.
Sink your teeth into in-depth insights from our contributors, and dive into financial and economic trends
Just when it seemed the US might relax more of its AI chip controls, China abruptly banned the import of Nvidia’s truncated AI chips in August. That ban came in the wake of “insulting” comments from US Secretary of Commerce Howard Lutnick. “We don’t sell them our best stuff, not our second-best stuff, not even our third-best,” Lutnick crowed to a TV interviewer. “You want to sell the Chinese enough that their developers get addicted to the American technology stack, that’s the thinking,” he said explaining US strategy. A week after Lutnick’s TV revelation, Beijing began its regulatory crackdown on US AI chips such as Nvidia’s H20.
Chinese tech firms including Alibaba and Bytedance were forced, first, to scale back and, weeks later, cancel their orders for H20 chips in response to pressure from the Cyberspace Administration of China. Regulators in Beijing reportedly invited top executives from Alibaba, Tencent, Bytedance, Baidu and Cambricon to caution them against buying H20 chips and call off testing of Nvidia’s chips and pressed them to explain why they might be willing to pick it over local options. Regulators have also forced local firms to stay away from Nvidia’s newest made-exclusively-for-China chip, the RTX Pro 6000D. The Financial Times recently reported that China’s regulators concluded that Chinese AI chips now perform at a similar level and, in some cases, better than Nvidia’s made-for-China chips that are allowed under US export controls.
US and Chinese officials are currently hammering out a broad agenda of what the two countries’ leaders will talk about when they meet on the sidelines of the Asia Pacific Economic Cooperation (APEC) summit in Seoul later this month. Whether Beijing and Washington agree on a trade deal or we see the escalation of tariffs and countermeasures, one thing is clear: Instead of buying overpriced fourth-rate AI chips that Lutnick wants to sell to them, China would rather make its own second-rate chip and work at improving it to bring it on a par with the top-of-the-line chips that Nvidia is shipping.
China’s homegrown ‘Nvidias’
So, who might be the Nvidia of China? Until recently, the big bets were on Huawei, which produces its own Ascend line of accelerated computing or AI chips. The US blacklisted Huawei in 2019, forcing it to produce more of its own cutting-edge technologies. Just last month, it unveiled its new top-of-the-line chip Ascend 950, which can handle different AI workloads using high-bandwidth memory (HBM) and which will start shipping early next year.
Over the last six months, the focus has been on four AI chip players — Cambricon, Moore Threads, Biren and MetaX — that are relatively unknown outside of China. Founded five years ago by former Nvidia staffers in China — including Zhang Jianzhong, who was once head of Nvidia’s China operations — Moore Threads is following Nvidia’s lead by focusing on making GPUs that are now used as AI chips. Among Moore Threads’ investors are Tencent Holdings and Bytedance. MetaX was founded and is being run by former executives of Nvidia’s biggest domestic rival, AMD. Its founder Chen Weiliang once worked as AMD’s global head of GPUs. Chen believes MetaX could deliver better GPU chips than H20, Nvidia’s made-for-China-only chip.
The AI chip company that has attracted the most attention is Beijing-based, Shanghai-listed Cambricon, whose stock has been on a tear. In the six-week period between mid-July and late August this year, it surged 185%. Despite a recent pullback, the share price is still up 90% this year. Founded by two brothers from the Chinese Academy of Sciences, it draws its name from the “Cambrian explosion” — symbolising their mission to ignite a revolutionary leap in computing and AI hardware. Cambricon’s flagship Siyuan chips are now being adopted by leading giants like Alibaba, Tencent, and DeepSeek for large model training and deployment.
The AI chip trade is not just for meme stock enthusiasts on Robinhood. Retail investors in China have chased every AI story that exists there. More supply of AI chip stocks is on the way. Biren is expected to list on the Hong Kong Exchange early next year in an IPO that will raise more than US$300 million. The company raised US$208 million from China’s state-linked investors earlier this year. With more AI chip listings in Shanghai and Shenzhen to follow next year, it looks like global investors will soon be spoilt for choice.
Assif Shameen is a technology and business writer based in North America
