Today, the company’s presence in the region hangs on a thread, caught between Washington’s strict restrictions on exporting its most advanced chips and Beijing’s retaliatory outright ban on Nvidia chips altogether (announced in September). For all of Nvidia’s successes elsewhere, the stinging loss of the Chinese market cannot be overstated. China currently accounts for roughly one-third of global semiconductor sales, making the country the largest single-country market in the world.
While the latest developments deal yet another blow to Nvidia, they certainly work to China’s gain. Semiconductors, by virtue of their irreplaceable role at the heart of the computing stack, are a cornerstone in China’s push for technological self-sufficiency. Viewed through this lens, restrictions on Nvidia’s GPUs only serve to accelerate the drive for domestic chip innovation. As it stands, China already appears to be mere “nanoseconds behind” the US in chip development, to quote Huang.
China’s push for semiconductor self-sufficiency has certainly shifted into high gear. In July, a consortium of Chinese AI companies launched the Model-Chips Ecosystem Innovation Alliance, aiming to boost the adoption of locally developed chips. Hot on the heels of this development, a nationwide mandate requiring public data centres to source at least half of their chips locally came into effect in August. And, in November, this mandate was revised to a total ban on foreign chips altogether.
China — the world’s largest semiconductor market — now faces the question of who will fill the void left by Nvidia. Fortunately, there is no shortage of contenders, established players and emerging companies alike, which are actively vying for the prized crown.
See also: The reasons for ‘unrational’ negative enterprise value stocks in exchanges
So, while Nvidia is currently relegated to — in Huang’s own cautiously optimistic words — “explain[ing] and inform[ing] and hold[ing] on to hope for a change in policy”, Chinese chip companies are already acting swiftly to seize a slice of this lucrative market.
Cambricon and the fabless four
Leading the charge among China’s fabless semiconductor specialists are Cambricon, Biren Technology, Enflame, Moore Threads and MetaX (which is neither Meta nor X). All are founding members of the Model-Chip Ecosystem Innovation Alliance and, like Nvidia, focus on designing rather than manufacturing chips (see Table 1).
See also: Strategic rationale behind the recent restructuring of our portfolios
Among them, Cambricon stands out as the only profitable and publicly listed firm in this list, making it the clearest bellwether of China’s domestic AI chip ambitions. Founded in 2016 by prodigy siblings Chen Yunji and Chen Tianshi, researchers with the prestigious Chinese Academy of Sciences (CAS), the company initially gained traction through intellectual property (IP) partnerships with Huawei. This, however, eventually led to a sharp setback when Huawei pivoted to in-house chip development for its Ascend line of chips via wholly-owned subsidiary Hisilicon.
For its second act, Cambricon shifted to developing chips for the server and cloud computing markets. In 2020, the company became the first domestic AI chip stock on the Shanghai Stock Exchange’s STAR Market. Its Siyuan 590 delivers approximately 80% of the performance of Nvidia’s (albeit older-gen) A100 chip, a widely used benchmark for AI training workloads. Looking ahead, the Siyuan 690 currently under development is expected to rival Nvidia’s newer and higher performance H100. Cambricon posted its first quarterly profit in 4Q2024 and, driven by the push for domestic substitution, the company registered a 4,000% surge in revenue to RMB2.9 billion in 1H2025.
The remainder of China’s promising AI-chip cohort — Biren, Enflame, Moore Threads and MetaX — remain privately held but have each expressed ambitions to tap the public markets. Currently, Moore Threads and MetaX appear the furthest along in the IPO process, having received the green light from the authorities for STAR Market listings. Both companies are helmed by industry veterans: Moore Threads’ CEO James Zhang had previously served as vice-president and general manager of Nvidia’s China operations, whereas MetaX’s founding team — Chen Weiliang (chairman), Peng Li and Yang Jian (co-CTOs) — are all AMD alumni. Separately, Biren and Enflame are reportedly eyeing Hong Kong listings in the near future. Investors should keep an eye out for these prospective IPOs.
Nvidia’s real moat
To be clear, Nvidia’s competitive edge extends far beyond GPUs alone. The company’s strategy relies less on selling individual chips and more on providing comprehensive and integrated solutions across a wide range of high-performance computing (HPC) and AI workloads.
Beyond its core GPU line-up, Nvidia’s product portfolio includes central processing units (CPUs; the Grace line is designed to complement Nvidia GPUs for heavy-duty workloads), data processing units (DPUs; the Bluefield line is specially for handling networking, storage and cybersecurity tasks) and networking interfaces (NVLink and NVSwitch, which connect GPUs with each other and/or other system components).
For more stories about where money flows, click here for Capital Section
On the software front, Nvidia’s parallel computing platform designed exclusively for its GPUs, Compute Unified Device Architecture or CUDA, underpins the company’s full-stack ecosystem. Launched in 2006, CUDA has become the de facto platform for AI development worldwide — offering a vast collection of tools, libraries and frameworks that simplify building and running AI systems across diverse use cases. Being Nvidia-specific, the platform effectively locks developers into the Nvidia ecosystem, creating a formidable moat for the company.
Looking ahead, Nvidia’s emerging ventures into 3D simulations for industrial and robotics (via Omniverse), health-tech (Clara) and autonomous driving (AGX) extend its presence further across multiple AI verticals. The company has, in effect, built a synergistic ecosystem spanning virtually all AI end-markets, entrenching its already dominant position in the market.
China’s full stack contenders
In China, the most relevant comparison to Nvidia’s full-stack ecosystem comes from legacy tech giants. While the alternative AI stacks they have built are certainly less comprehensive or mature than Nvidia’s, these incumbent giants benefit greatly from leveraging their existing businesses for accelerated ecosystem adoption. Alibaba and Baidu, for instance, are heavily invested in semiconductor research and design through their respective semiconductor units, T-Head and Kunlunxin.
T-Head’s parallel processing unit (PPU), an alternative to GPUs for AI workloads, made headlines recently when China’s national broadcast channel CCTV aired a chart that showed it surpassing Nvidia’s China-exclusive A800 in key specifications. T-Head is currently the primary supplier for China Unicom’s 20,000 petaflop data centre under construction in Qinghai. Meanwhile, Baidu’s Kunlunxin has among its customers major networking and telecom suppliers, including H3C Technologies and ZTE, first-tier partners of the nation’s largest wireless carrier, China Mobile.
The clearest contender for Nvidia’s vast ecosystem, however, is Huawei. As China’s leading technology powerhouse, Huawei has emerged in recent years as the central orchestrator of the nation’s semiconductor efforts. Per analysts tracking China’s semiconductor industry, the company effectively operates with “a blank cheque from the NDRC” (National
Development and Reform Commission), China’s
central agency for economic planning and industrial policy. One such initiative is the Tashan Plan that aims to build a fully indigenised semiconductor manufacturing line, an effort that Huawei has led since 2020.
Huawei’s stack
Huawei’s Ascend AI chips have emerged as a top alternative to Nvidia’s GPUs across multiple AI workloads. Performance benchmarks indicate that the chip performs on a par with Nvidia’s A100. In recent years, Huawei’s Ascend chips have been the primary hardware used to train more than 70 Chinese large language models (LLMs). The company is also increasingly competitive on the software front, with its Compute Architecture for Neural Networks (CANN) platform, an open-source alternative to Nvdia’s CUDA (see Table 2 for an overview of Huawei and Nvidia’s AI stack).
Separately, Huawei also offers networking solutions through United-Bus Mesh (UB-Mesh), providing universal interconnect interface across the entire hardware stack, unlike Nvidia’s NVLink and NVSwitch that primarily handle GPU connections. UB-Mesh underpins Huawei’s “cluster computing” systems that consist of hundreds of individual chips linked to a unified computing system (marketed as the Atlas SuperPoD/SuperCluster or Cloudmatrix line). On brute power alone, Huawei’s Cloudmatrix 384, which packs 384 Ascend 910C chips into a single system, overshadows its Nvidia counterpart, the GB200 NVL72 system.
In September this year, Huawei’s rotating chairman Eric Xu unveiled the company’s three-year grand strategy to unseat Nvidia in AI chips, which includes the ambitious bid of launching four chips within the time frame. The Atlas 960 SuperPoD cluster expected to be rolled out in 2027 is designed to connect more than 15,000 Ascend chips into a single system, while the same-variant SuperCluster model (composed of multiple SuperPoDs) will connect over one million Ascend chips.
The next frontier
For all the ground we have covered on China’s existing semiconductor scene, we have thus far said little about the moonshot projects that Chinese academics and enterprises are undertaking in pursuit of breakthroughs in semiconductor technology. And there have certainly been many. One such effort that has come to fruition in recent weeks is the analogue AI chip that was developed by researchers at Peking University. Reportedly, the chip achieved processing rates that were 1,000 times faster than today’s leading GPUs.
Unlike digital chips, which reduce all data to 0s and 1s, analogue chips operate across a continuous spectrum — using physical phenomena such as voltage, resistance and current to perform computations. Accordingly, what this prototype analogue chip demonstrates is the potential of using an altogether different paradigm for computing.
For historical context, it’s worth noting that analogue computing actually predates the digital revolution, with roots traceable to the water clocks of Egyptian and Babylonian antiquity. And while the continuous nature of analogue computing affords it the advantage in simulating complex systems and is thus useful for aeronautics, scientific modelling and, naturally, AI workloads, it has long been plagued by limitations. Specifically, its reliance on physical processes makes it inherently prone to imprecision and instability, as minor inconsistencies across these physical components compound over time.
Analogue computing reimagined
The Peking University breakthrough leverages multiple innovations to overcome this long-standing limitation in analogue computing. The chip uses “memristive” elements whose electric resistance can be precisely tuned to both store and process data through analogue means. By arranging these elements in grid-like structures, the chip performs calculations directly using electric signals — that is, in an analogue fashion. Further, by breaking each calculation down into smaller operations and subsequently recombining them to produce the final result, this allows for near-digital precision. Using these groundbreaking methods — low-precision matrix inversion circuits, crossbar arrays and bit slicing — the Peking University analogue chip achieved improvements from earlier analogue systems by five orders of magnitude.
Because memristive elements act as both storage and computing, the chip supports in-memory computing, performing operations directly where data resides. This has significant implications, as it bypasses a long-standing limitation in digital computing known as the von Neumann bottleneck. Traditionally, because data and computing instructions share the same pathway, they compete for bandwidth, creating a fundamental limitation on computing speed. In-memory computing sidesteps this constraint altogether by removing the need to shuttle data back and forth, performing computations directly where the data resides. This, in turn, potentially enables far faster and more efficient processing.
As such, the significance of this breakthrough in analogue computing lies not just in its technical achievement, but in what it represents: China’s willingness to probe entirely different computing paradigms and consequently, potentially, leapfrog past Western innovation.
Still in the foundry
Therein lies both the headache and the thrill of trying to peg China’s next tech titan. There is every likelihood that whatever comes next from there may very well chart an entirely different technological path than anything the West has produced.
When it comes to expressing China’s resolve in charting its own path, Huawei CEO Ren Zhengfei captured this spirit perfectly. Asked why Huawei insisted on keeping its name (which, incidentally, translates roughly as “Chinese achievement”) as it went global, he explained: “We […] will teach foreigners how to pronounce it. We have to make sure they do not pronounce it like Hawaii.”
This blend of pride, audacity and sheer determination is emblematic of China’s approach to innovation.
Likewise, it may very well be the case that China’s next champion will not be an Nvidia-alike or anything remotely like it — and yet prove to be every bit as formidable. And if it happens to be one of the companies already making waves, it will step into a landscape no less dynamic and unpredictable, demanding innovation at every turn.
The bottom line
And where exactly does this leave investors looking to take a nibble of China’s AI chip market? Of the above-mentioned that are currently listed, there are Cambricon, Alibaba and Baidu.
Cambricon’s meteoric rise this year has hardly gone unnoticed. At its August high, following blockbuster 1H2025 results, the stock was up 500% year on year (see Chart 1 above). And on a more practical note, it belongs to the narrow cohort of STAR Market companies accessible only to qualified foreign institutions. For retail investors who remain optimistic that Cambricon’s second act still has room to run, the only viable option is investing through an exchange-traded fund or ETF. Table 3 highlights the relevant ETF options for indirect exposure to the company. With Cambricon’s price-to-earnings ratio hovering around an eye-watering 300, however, we are content to stay on the sidelines in search of more compelling, risk-adjusted opportunities.
Meanwhile, our positive stance on Alibaba remains unchanged. It is a core holding in both our Absolute Returns and AI portfolios, and we continue to monitor Baidu closely as it navigates the ongoing disruption to its core advertising segment from AI.
The investable pickings in the AI chip space remain limited for now but, as we have highlighted above, the pace of China’s technological progress is certainly impressive. It is simply a matter of time before further opportunities emerge. Follow along as we continue monitoring this space.
The Malaysian Portfolio gained 1.4% for the week ended Nov 19, outperforming the benchmark FBM KLCI, which fell 0.5%. United
Plantations (+6.7%), Kim Loong Resources (+3%) and Hong Leong Industries (+3%) were the three gaining stocks while Malayan
Banking (-0.1%) and LPI Capital (-0.1%) ended marginally lower. Last week’s gains lifted total portfolio returns to 194.4% since inception. This portfolio is outperforming the benchmark index, which is down 11.2% over the same period, by a long, long way.
The Absolute Returns Portfolio, on the other hand, fell 2% last week amid profit-taking in the US and global markets. All three US bellwether indices — the Dow Jones Industrial Average, S&P 500 and Nasdaq Composite —fell sharply. Nevertheless, this portfolio is still performing well with total portfolio returns at 41.8% since inception in March 2024. The only gaining stock last week was Ping An Insurance - A (+0.5%). Goldman Sachs (-6.3%), Tencent (-5.5%) and JP Morgan (-5.3%) were the three biggest losers.
The AI Portfolio also ended sharply lower, down 5%. The loss pared total portfolio returns to 2.1% since inception. Naura Technology (+10.8%) was the only gaining stock for the week while the biggest losing stocks were Horizon Robotics (-10.3%), Marvell Technology (-9%) and Amazon.com (-8.8%).
Disclaimer: This is a personal portfolio for information purposes only and does not constitute a recommendation or solicitation or expression of views to influence readers to buy/sell stocks, including the particular stocks mentioned herein. It does not take into account an individual investor’s particular financial situation, investment objectives, investment horizon, risk profile and/or risk preference. Our shareholders, directors and employees may have positions in or may be materially interested in any of the stocks. We may also have or have had dealings with or may provide or have provided content services to the companies mentioned in the reports.
