By comparing sector leaders — from EVs and cloud computing to semiconductors and defence AI — it becomes evident that the market has fundamentally mispriced China’s structural advantages. As its manufacturing prowess builds up, innovation accelerates, competition eases, and domestic substitution progresses, will these undervalued “juggernauts” be poised for a significant long-term re-rating?
However, the stock markets seem not to care much about it and have not fully priced in these successes, with the CSI300 declining over the same period to end-2025 and its price-to-earnings ratio remaining low at 12 times, even as a double-digit earnings recovery is widely expected this year.
While it is true that many major listed flag-bearers in its high-tech sectors have yet to deliver strong profitability, as hefty depreciation from recent massive annual capital expenditure during the ramp-up phase depresses earnings for stocks like Hong Kong-listed semiconductor foundry SMIC, the successful ones should make good profits in the medium term.
It is undeniable that intense competition across many sectors has depressed earnings. Still, as competition eases, as it ought to, earnings will likely gather strong headway, as Darwinian evolution asserts itself, as it should. In this piece, we will lay out the foundations for this scenario via some stock examples.
See also: Global 2025 power demand rose as EV, data centres grew, IEA says
Another reason for the depressed CSI300 valuation for most of the last five years is the thundering herd of Western media outlets insinuating that China is uninvestable, with what amounts to multi-year smear campaigns scaring off many a global investor, and even domestic investors. Their scary narratives include, but are not limited to:
Beijing’s adoption of common prosperity as its key ideological driver, President Xi Jinping and the Communist Party of China’s nostalgia for the Mao era,
The property sector meltdown is leading to a 20-year or even 30-year Japan-style deflationary spiral, and
See also: Hormuz chaos, Lebanon clashes undermine Trump peace deal hopes
The inevitable renminbi crash, etc.
These alarmist refrains have persuaded some investors to have zero allocation to China. Even the Chinese investors I have spoken to have been advocating Japanese, Indian and US stocks over Chinese stocks, especially before September 2024.
Across the Pacific Ocean, those same media outlets have effectively institutionalised the “American Exceptionalism” narrative, positioning US AI leaders as an asset class unto themselves — consistently framing the US as the “only game in town” for AI infrastructure. That group of anointed AI leaders has become a magnet for global fund flows, with mere announcements of yet more gargantuan capex rounds doubling share prices in a matter of weeks and months. However, sentiment seems to have shifted somewhat in recent months. Is the law of gravity eventually asserting itself?
By March, Larry Ellison’s Oracle had more than halved from its September 2025 peak of US$345 a share as investors soured on massive, debt-fuelled AI capital expenditure.
The unhinged printing of money by central banks has distorted the pricing of assets and securities, most noticeably in precious metals and cryptocurrencies. The mania created by this wall of excess liquidity even reached Welsh Wrexham Association Football Club. Private equity giant Apollo’s affiliate acquired a minority stake of reportedly just under 10% in the club, valuing the Championship side at approximately US$475 million, a staggering 18,900% increase over the US$2.5 million co-chairmen Ryan Reynolds and Rob McElhenney paid for the club in 2021. Investing history has taught us that easy money over an extended period tends to lead to investor hubris, irrationality and silliness.
The goal of this essay is to compare China’s key high-tech stocks with their American counterparts. While most of the stocks we will examine in this piece are high-tech, we will use a traditional telco stock with pedestrian growth rates as our canvas to paint the picture of how differences between players on opposite sides of the Pacific will show up in their valuations.
For more stories about where money flows, click here for Capital Section
Tesla vs BYD
Tesla and BYD are indisputably the global EV market’s “Twin Titans”, each with mastery of core technologies such as batteries, electric motors, and electronic controllers. They both possess vertically integrated manufacturing capabilities and have developed energy storage as secondary growth engines. Both have set up factories in China, the Americas, and Europe to compete for the global market.
Both founders — Tesla’s Elon Musk and BYD’s Wang Chuanfu — have technical backgrounds. They both went all in during the EV industry’s nascent stage and disrupted it: Musk bet heavily on Tesla and SpaceX, while Wang entered the automotive industry as early as 2003 to challenge the traditional giants of fossil-fuelled vehicles. Both men can stomach risk in pursuing aggressive business expansions and extensions.
Their similarities end here.
Tesla differs in strategic positioning and product routes. It aims to become a software-centric “vehicle cum AI ecosystem” technology company. At the same time, BYD strives to be a manufacturing giant covering all price points, with a fully vertically integrated supply chain. Accordingly, their product positionings also differ. In terms of hardware, Tesla focuses on two mainstream mid-to-high-end pure-electric models: the Model 3 and Model Y. Its Model S and Model X, as well as its Cybertruck, account for only 3% of total sales. It relies on its Full Self-Driving (FSD) autonomous driving and intelligent cockpit to improve gross margins and long-term value, deploying its Robotaxi mobility network, and pursuing a closed loop comprising software, operations, and data. Tesla is also expanding into the humanoid robot sector, aiming to integrate AI and the Tesla ecosystem into its consumer products.
In contrast, BYD’s strategy is to offer a full suite of models with a price range, from US$10,000 to US$200,000, in pure electric and plug-in hybrid routes, with more than 40 models on sale and over 20 new models launched annually, of which a dozen are all-new or fully redesigned models. Relying on its overwhelming scale advantages and full control over its vertically integrated supply chain, BYD indisputably commands a significant cost advantage in manufacturing.
For vehicles of comparable size and functions, BYD offers consumers increasingly superior hardware specifications at significantly lower prices than Tesla.
Tesla achieved initial success with a blockbuster model, while BYD followed, learned from it, and eventually surpassed Tesla. Beyond BYD, Chinese EV makers have also flourished, and soon, more will emerge as global automakers, bringing fiercer competition not only to Tesla but also to the entire global automotive industry. Thanks to its superior supply chain, Chinese automakers hold an absolute cost advantage in manufacturing. In software, however, Tesla currently leads in autonomous driving, though the gap has narrowed as technological iteration has slowed. In the field of humanoid robots, many of China’s automakers have also been catching up with Tesla. While Tesla currently leads in the zero-to-one stage for both autonomous driving and humanoid robots and thus commands a valuation premium, Chinese automakers will eventually assert themselves in the one-to-10 scaling stage.
Tesla’s current valuation of US$1.5 trillion is 12 times higher than BYD’s US$131 billion, despite its sales volume being just a third of BYD’s. Its P/E and P/B ratios are also considerably higher, at 13 and three times, respectively. The colossal valuation gap suggests high expectations among investors for Tesla’s next technological revolution. Is this premium justified? In our view, investors are overly optimistic about Tesla’s next technological breakthrough and are underestimating BYD as an undisputed leader in the EV space.
Amazon and Alibaba are both formidable e-commerce and cloud players, but their valuations differ starkly. Amazon is trading at a market capitalisation of US$2.1 trillion, whereas Alibaba is trading at a market cap of RMB2 trillion ($374 billion). Given the USD-CNY exchange rate difference, one Amazon is equivalent to seven Alibabas.
The far higher valuation of Amazon is, to some extent, understandable. With its domestic 40% market share, larger than that of its next few competitors combined, Amazon faces less competition in its e-commerce business than Alibaba and has more room to raise its operating margins. Amazon Web Services (AWS) is also far more profitable than Alibaba Cloud.
In China, Alibaba’s e-commerce operating margins may be much higher than Amazon’s, as it does not have to handle fulfilment. But Alibaba’s former 80% market share more than 10 years ago has crumbled to just 35% today, under pressure from numerous formidable competitors like Douyin, Pinduoduo (PDD), and JD.com, each of which has captured 15%–20% of the market. No surprise, then, that Alibaba’s 60% China e-commerce-adjusted operating margins 10 years ago have declined to just 40% in 2023 and 2024. These juicy margins are now plunging to 20% in 2025 and 2026 as Alibaba invests heavily to subsidise users’ orders made through its AI chatbot and to drive instant retail.
Meanwhile, Amazon’s international business is also profitable, unlike Alibaba’s, which has been outpaced by Shopee, Shein, Temu and TikTok Shop.
Cloud is where Alibaba’s biggest promise lies. Unlike US corporates, China is in the early stages of cloud adoption, especially AI cloud adoption. And while Amazon invests in Anthropic and OpenAI to gain exposure to the upside of large language models, Alibaba has a well-regarded self-developed model, Qwen.
Three of Amazon’s data centres in the Middle East have been bombed by Iran, and Amazon may in the future be deprived of the opportunity to build data centres which can tap into abundant and cheap natural gas.
Alibaba, however, can enjoy China’s significantly lower electricity costs. Combined with a growing domestic chip supply, it can continue to aggressively price its cloud offerings and attract customers in Asia Pacific and the emerging markets. In the long run, energy costs will play a defining role in deciding whether an AI agent or a large language model can compete effectively.
Alibaba’s cloud competitiveness raises the tantalising prospect that Alibaba Cloud, already the largest player in China with a 35% market share, may one day gain significant market share against Amazon, Google and Microsoft. It has already been reported that many US AI app companies and enterprises are using Qwen because of its compelling price competitiveness. If so, sell-side model valuations of AliCloud at about six times forward revenues may have upside, especially against the current AWS valuations of six to 10 times 2027 revenues.
Meanwhile, Alibaba’s eventual AI capabilities, combined with its AMap app — akin to the Google Maps of China — may, in the coming years, revitalise its e-commerce business relative to competitors, leading to a re-rating. With lower P/E and P/B multiples, yet with future e-commerce profit growth matching Amazon’s and cloud profit growth significantly faster, Alibaba’s market cap at less than 15% of Amazon’s seems unreasonable. This is even more so given that the Chinese e-commerce market is the largest in the world, and the AI market, especially its applications, will likely eventually be at least as large as the US market.
Apple vs Xiaomi
When Xiaomi listed on the Hong Kong Stock Exchange in July 2018, its market cap was around US$50 billion, 5% of Apple’s US$1 trillion market cap, the company it has spent its life trying to emulate. Eight years later, Xiaomi ranks third worldwide for smartphone shipments, behind Samsung and Apple. But its market cap has only increased 1.2 times to US$110 billion, 25 times 2026 estimated earnings, even while Apple’s market cap has quadrupled to almost US$4 trillion, or 28 times 2026 estimated earnings.
Apple is enjoying a tailwind from its successful iPhone 17 launch, which enticed users to replace their ageing iPhones. Apple is expected to sell 15% more iPhone 17s than iPhone 16s. Xiaomi, however, was forced by the surge in memory prices to raise prices for its mass-market customers. Its phone shipments are expected to fall 10% in 2026.
Financial markets may have downgraded Xiaomi due to a near-term slowdown in EVs and smartphones, but its longer-term growth story remains largely intact. Apple, on the other hand, has squeezed higher margins out of its core products and boosted its share price through share buybacks. Its innovative chutzpah seems to have faded, as evidenced by its lacklustre sales over the past few years.
Xiaomi has notably succeeded where its rival failed: To build an EV that acts as another extension of the smartphone. Ford’s CEO, Jim Farley, memorably praised Xiaomi’s SU7 as a car he does not want to give up. Revenue from Xiaomi’s EV business and other initiatives accounted for 10% of total revenue in 2024, 23% in 2025, and is expected to reach 40% by 2028, when it may sell more than a million cars. Compared with its EV rivals, Xiaomi seems more attuned to what its customers want and offers a compatible software ecosystem across numerous other products.
Xiaomi’s “Mi Home” business began out of necessity in 2016, when it urgently needed to diversify from its online-only, smartphone-only business model. Today, a Mi Home store sells TVs, air-conditioners, robotic vacuum cleaners, rice cookers, smart glasses, and many other gadgets. As a result of these diversification efforts, Xiaomi’s smartphone sales are expected to account for just 35% of its 2026 revenue, compared with 52% for Apple.
By the 2030s, the smartphone may not even exist in its current form. Meanwhile, Mi Home and EVs may reach their full potential given the pace of AI agent and robotics developments. In our view, the continued growth of Xiaomi’s non-smartphone businesses is not priced into its 15–20 times 2027–2028 P/E. Xiaomi, which literally means a small grain of rice or the millet grain in Chinese, may one day epitomise one of founder Lei Jun’s favourite phrases: “A grain of rice, to the Buddha, is as big as Mount Sumeru.”
At US$3.7 trillion and 50 times P/B, Apple seems priced for perfection, while Xiaomi, at US$110 billion and 2.8 times P/B, seems priced for ex-growth, despite achieving much stronger growth rates than Apple over the last five years. In the future, its growth rates, driven by its new and existing products, will exceed Apple’s.
Palantir vs Geovis
While America’s Palantir is a comprehensive data integration and analytics platform designed for complex decision-making and pattern recognition, China’s Geovis focuses specifically on the high-fidelity imagery and spatial analysis of geographic data. However, both Palantir and Geovis function as “national champions” in the AI sector’s defence and intelligence niche, where their stock values are driven by sovereign contracts and their roles as strategic proxies for the future of autonomous warfare and logistics. As the saying goes, “Same, same, but different”.
Palantir was founded in 2003 to address US counterterrorism intelligence needs, with deep ties to the Republican Party and national security apparatus. It operates as a universal big data decision-making platform, with satellite data among 160+ heterogeneous data sources. Its broad, decision-layer focus delivers cross-industry applicability, with commercial operations across a broad spectrum of sectors now its core growth engine, accounting for around 50% of revenue and a 60% y-o-y growth rate in 2025.
Geovis, spun off from the Chinese Academy of Sciences (CAS) Aerospace Information Research Institute, is rooted in indigenous capabilities in space-air-ground topography. It is a vertically integrated platform focused exclusively on orbital and airspace information, with satellite data as its core source. It also uses AI and big data to improve its analytical precision. In addition, its emerging commercial aerospace and low-altitude economy segments are closely aligned with its core geospatial expertise, boasting deeper moats in niches centred on geospatial data processing, imagery, and the underlying infrastructure. Its revenue structure is 50% GIS (Geographic Information System) civil, 30% GIS military, 15% commercial space and 5% low-altitude economy.
• Technology moats
Palantir’s competitive edge lies in AI-powered multi-source data fusion and end-to-end decision-making, with extreme customer switching costs stemming from its deep integration into clients’ core business workflows. Geovis’ technical moat stems from its full-stack, end-to-end orbital-atmospheric-terrestrial information-processing capabilities, with industry-leading moats in national defence, the low-altitude economy, and other geospatial-temporal-intensive scenarios. In the seamless flow of data, when a satellite in orbit detects a change, an atmospheric platform, such as a Medium-Altitude Long-Endurance (MALE) drone, is cued to get a closer look, and a terrestrial self-driving truck or a factory takes action based on that insight.
While Palantir’s ROA (return on assets) performance is stronger, Geovis’ growth trajectory holds greater potential, as illustrated by the metrics in the table below. For Geovis, while its legacy government GIS business moderated last year, which we understand to be temporary, its commercial aerospace, low-altitude economy, and AI cloud segments are in the early stages of commercialisation, with vast underpenetrated market opportunities. A huge business opportunity lies in overseas orders, which the company is tight-lipped about due to geopolitical sensitivity.
Investor sentiment is currently dampened by the delayed share placement, weak 2025 results, and a potential transfer of ownership from CAS to China’s State-owned Assets Supervision and Administration Commission (Sasac). Geovis’ strategic transformation has significantly crimped its net profit, which will likely remain modest in 2026, with a volatile P/E ratio.
Both stocks are considered super-growth stocks by investors, hence their rich valuations, as evidenced in the table below. Which stock outperforms the other depends on which company can grow faster over the next five to 10 years.
SMIC vs TSMC
Although founded and listed in Taiwan, TSMC is, for all intents and purposes, a US company because most of its big investors and clients are Americans. It is by far the most successful and impressive foundry in the world. Its competitive advantages lie in its advanced nodes (7 nanometres and below account for 77% of revenue) and in monopolising major customers like Nvidia and Apple, wielding its extremely strong pricing power to achieve gross margins of 62.3% and deliver an ROE of 35.4% to its shareholders. However, life is not a bed of roses, as TSMC’s business is constrained by American geopolitics, which prevents it from doing business freely in China.
Ironically, China’s leading foundry, SMIC, is a huge beneficiary of the US chip war against China. To nurture local champions, Beijing is providing SMIC with unconditional, comprehensive policy support, including tax incentives, funding, cheap land and grants. To achieve full semiconductor independence one day, Beijing is strongly pushing SMIC’s key customers, such as H (Huawei/HiSilicon), Cambricon, and Hygon, to replace foreign suppliers with local ones for advanced nodes. This is amid the persistent supply shortage for SMIC’s advanced nodes, which allows SMIC to price its N+2 (equivalent to 7nm) wafers higher than TSMC’s — the sovereignty premium.
However, SMIC is hobbled by constraints on equipment imports, a lack of access to high-end extreme ultraviolet (EUV) machines, forcing expensive and complex deep ultraviolet (DUV) multi-patterning workarounds, slower progress in advanced nodes relative to TSMC and difficulty in securing orders from global design leaders due to the American-led geopolitical campaign to stymie China’s technological advancement.
TSMC currently holds 70.4% of the global pure foundry market share, against SMIC’s 5.1%, and its annual capital expenditure of US$52 billion to US$56 billion is 6.8 times SMIC’s US$8 billion. Based on China’s commitment to rapid, extensive localisation and industry trends, as well as its strong pipeline of talent that continues to deepen its already substantial pool of STEM talent, we believe the gap between the two companies’ market shares will narrow in the not-too-distant future. This is driven by:
First, localisation policy: It is expected that over the next five years, Beijing will invest approximately RMB500 billion to RMB600 billion to expand domestic advanced node manufacturing capacity by 500,000 wafers per month. Semicon China, the organiser of the largest semiconductor exhibition in the world, forecast on March 26 that 47 of the 108 new wafer fabs in 2028 will be built in China. Currently, constrained by US restrictions, China’s advanced process capacity accounts for approximately 6% of global capacity, significantly below the potential global demand for China to take 30%. Beijing is pushing hard for all domestic demand for advanced process wafers to be 100% met by domestic foundries over the medium term. The government’s end-state goal for advanced fabs is 100% domestic equipment, moving past the current 50% floor.
Next, strong talent reserves: China graduates approximately five million STEM students annually, considerably higher than Taiwan’s 70,000-80,000 and the US’s approximately 500,000. The continuous fire hose of fresh talent will power China’s catch-up game in the foundry space. Meanwhile, many core technical personnel from leading global AI software and hardware companies, as well as semiconductor companies, are of Chinese descent and key talent in these areas has been returning to China. In the long run, talent will decide the winner of this contest.
Currently, SMIC’s rapid, massive capex has depressed its comprehensive gross margin to just 21% and an ROA of only 1.3%, which is primarily dragged down by a huge US$4 billion depreciation expense, equivalent to 44% of revenue and 3.7 times of pretax profits last year, compared to 17% and 31% for TSMC. This means that SMIC’s profits are currently depressed, and future profits must improve, also helped by higher yields. The other huge catalyst would be the successful development of domestic EUV lithographic equipment. Huawei and many research institutes have been working on it for more than five years. Based on our conversations with industry experts, 2027–2028 is the likely timeline, though the details remain confidential.
As a result of the headwinds the company faces, SMIC’s market cap is only 4% of TSMC’s. By 2030, SMIC’s gross margin should reach 32.2% and its ROA will likely improve to 7.2%, from the current 21% and 1.3%, respectively.
In conclusion, TSMC will continue to generate strong profits, while SMIC will continue to improve across all financial metrics. By 2030, will SMIC’s market cap still be just 4% of TSMC’s? We do not believe so.
Naura vs Applied Materials
Global semiconductor equipment leader Applied Materials has two core competitive advantages. First, there’s deep integration with cutting-edge process-technology R&D at leading wafer fabs such as TSMC and SK Hynix. Next, a broad range of equipment categories (thin-film, etching, ion implantation, etc.) provides pricing advantages across its product portfolio.
However, its growth is hamstrung by America’s geopolitical considerations and higher operating costs, such as labour and utilities, than those of mainland China manufacturers.
Naura is China’s leading domestic semiconductor equipment company, which can count on Beijing’s end-state goal for advanced domestic fabs to move beyond the current 50% floor for domestic equipment use toward full substitution of foreign suppliers via subsidies and policy mandates. Domestic equipment companies also have significantly lower labour costs and facilities costs. However, Naura is handicapped by its inability to participate in new-technology R&D projects with leading global manufacturers and to adopt state-of-the-art technology standards, risking becoming a perpetual follower with weak global market competitiveness. Because Naura is restricted from working with TSMC or Intel on their sub-2 nm roadmaps, it cannot participate in the “Standard Definition” phase — where the physical dimensions, electrical tolerances, and chemical requirements for the next generation of chips are set.
While Applied Materials and Naura have highly similar equipment category coverage, Applied Materials’ near 20% global market share is five times Naura’s 4%. Accordingly, Applied Materials’ annual RMB24 billion R&D investment is four times Naura’s RMB6 billion. That said, these two chasms will continue to narrow over the medium term due to China’s strong localisation push and industry trends, as well as its strong talent pool and product pipeline.
Consider these two key factors:
Firstly, in addition to the support for expanding domestic advanced-node manufacturing mentioned earlier, China’s localisation rate for semiconductor equipment has reached 25%, with the non-lithography equipment localisation rate at 35%. It is expected that by 2030, the non-lithography equipment localisation rate will increase to 50%–60%, with semiconductor equipment localisation rates for advanced nodes and DRAM achieving significant improvements, reaching 20%–60%.
Next, leading US semiconductor equipment companies such as Applied Materials and Lam Research have a significant number of Chinese professionals in their core R&D teams. Besides the high-profile Gerald Yin of AMEC (Advanced Micro-Fabrication Equipment Inc China), an increasing number of Chinese talents have recently “bit the bullet” and uprooted their families and returned to China to start their own companies. Renouncing his US citizenship in 2025, 82-year-old Yin is the face of China’s “reverse brain drain.”
By neutralising the “citizenship liability” created by US export controls, the Silicon Valley veteran — who held senior roles at Intel, Applied Materials, and Lam Research — now leads AMEC’s critical 7nm and 5nm programmes. Widely viewed as the “Godfather” of Chinese etching technology, Yin famously founded AMEC in 2004 at the age of 60, when most of his peers were retiring. After spending over 20 years at the heart of America’s chip industry, his high-profile return and recent renunciation of US citizenship have inspired a wave of elite Chinese talent to follow suit, returning home to drive the nation’s domestic “Hard Tech” sovereignty. China graduates approximately 100,000–150,000 electrical engineering students annually, while the US graduates only 20,000–30,000 yearly.
Besides its superior talent pipeline, China possesses deeper reserves of semiconductor equipment-related scientific research talent. Already, about 80% of the peer-reviewed semiconductor research papers are produced by Chinese research institutes and academia.
Whilst Naura continues to benefit from China’s rapid domestic substitution drive, continuing apace, we expect the company’s global market share to grow to 6% by 2030, from the current 4%. Naura’s global market share can even rise to 10%–15% if the company invests heavily in R&D, and, of course, provided that global market barriers are removed.
Applied Materials currently leads via R&D and its edge in “Standards Definition” integration. However, Naura is narrowing this gap, aided by Beijing’s nearly 100% substitution goal, lower costs, and the “reverse brain drain”.
Supported by China’s superior pipeline of STEM graduates, its massive domestic market, and potentially better access to the global market, we believe Naura is positioned for stronger growth.
Hygon vs Nvidia
Nvidia stands as the undisputed titan of the AI era, protected by a US$4.4 trillion valuation and a CUDA ecosystem that remains the “AI GPU gold standard”. Yet while Nvidia navigates the challenge of sustaining “10 to 20” growth at the summit of the global market, a new challenger is emerging from a far lower base. Hygon, with a market cap of US$76 billion, though currently outmatched in raw chip-to-chip brawn, is fast catching up through elite system-level engineering and a software strategy that effectively dismantles the “CUDA moat,” offering a seamless migration path for a domestic market that Nvidia can no longer fully serve.
The investment thesis for Hygon rests not on matching Nvidia’s prowess, but on capturing a structurally protected, US$200 billion domestic opportunity where sovereignty and availability now outweigh absolute speed. By leveraging China’s “power of power” — a stable, low-cost energy grid that allows for AI output at one-tenth the price of Western competitors — Hygon is transitioning from a local substitute to a full-stack exporter. Positioned at the explosive “1 to 10” growth phase, Hygon offers superior earnings elasticity and a domestic market narrative in the AI era.
Nvidia’s technological prowess lies in its near-impassable CUDA ecosystem moat and a hardware roadmap — from Blackwell to the upcoming Rubin architecture. Acknowledging this supremacy, investors have crowned Nvidia as the world’s most valuable company.
Hygon is one of only two Chinese players with a credible full-stack domestic compute offering, spanning both general-purpose x86 CPUs and General-Purpose Graphics Processing Units (GPGPUs). Indeed, Hygon’s single-chip performance is considerably behind Nvidia’s. However, it is compensating for engineering innovation at the system level, enabling the chips to work together efficiently as a cluster.
Its ScaleX640 super node packs 640 accelerators into a single, hyper-efficient, high-density unit that treats hundreds of chips as a single giant brain, linked by an ultra-fast 400G Native RDMA data connection that lets chips communicate intelligently and quickly. Its remarkable Power Usage Effectiveness (PUE) of 1.04 enables industry-leading cluster-level efficiency and significantly improves throughput in large-model training and inference workloads. Efficiency in data centres is measured by PUE, where 1.0 represents a theoretical “perfect” system with zero energy wasted on overhead. Hygon’s 1.04 score places it in an elite tier with a mere 4% energy overhead — surpassing the 1.2 “very efficient” standard used by Google or Microsoft’s hyperscale centres and far outperforming the 1.6+ inefficiency typical of older, air-cooled server rooms.
For software, Hygon does not need to replicate CUDA. It bypasses the need to build a proprietary software ecosystem by leveraging its ROCm-compatible software, an “open-source” alternative to CUDA, that allows different brands of chips to work together. With 99% compatibility for the world’s most popular AI tools and support for major frameworks like PyTorch and TensorFlow, it offers a seamless migration path for developers that materially lowers customer switching costs, effectively eliminating the technical barriers that usually lock customers into Nvidia’s ecosystem.
Hygon’s growth will likely compound at a much faster rate than Nvidia’s, albeit from a lower base, in a regulatory-protected, supply-constrained, industrial-policy-supported domestic market that Nvidia can no longer adequately serve. The Chinese market is expected to grow at a CAGR exceeding 50% and is likely to surpass US$200 billion by 2030.
If Hygon secures a 10% market share, its annual revenue could reach US$20 billion, 10 times its 2025 level. In the following table, we lay out some of their key metrics. We assume a 50% CAGR for China’s AI infrastructure market until it exceeds US$250 billion per annum by 2030. In our scenario, China’s localisation rate hits 70%, while Hygon captures 14% of the total domestic market, or 20% of the portion served by domestic suppliers only. We assume Nvidia grows 50% in 2026, then slows to a 30% CAGR for 2027–2030.
The power of cheap power: infrastructure and cost advantages are driving a surge in “token exports”. In February, the volume of Chinese model requests on OpenRouter surpassed that of US models for the first time, fuelled by a decisive cost-per-token advantage. While US AI companies struggle with an ageing power grid and high utility rates, China enjoys rates that are 50%–60% lower, underpinned by its Eastern Data, Western Computing (EDWC) national megaproject and direct access to green energy sources.
EDWC rebalances the country’s digital economy by shifting data processing from the saturated eastern coast to the resource-rich western interior. Launched in 2022 and entering a high-utilisation phase under the 15th Five-Year Plan (2026-2030), the initiative treats computing power as a strategic national resource, on par with energy or water. It establishes eight “national computing hubs” — including Inner Mongolia, Ningxia and Guizhou — to serve as the backbone for heavy-duty processing, such as AI model training and big data analysis. By leveraging the West’s cooler climates and abundant renewable energy, EDWC aims to lower the operational costs and carbon footprint of China’s tech sector.
As of 2026, the focus has evolved toward “Computing-Electricity Synergy,” ensuring direct, green energy sources power these massive data centre clusters to maintain grid stability and meet net-zero targets. That is where the 2026 National Energy Administration (NEA) Green Power Direct Connection (GPDC) policy comes in. It permits dedicated transmission lines to link renewable energy sources directly to high-load industrial sites, such as data centres.
By bypassing the public grid, these projects provide 100% physical “green energy traceability.” In international markets, this is known as a Direct Power Purchase Agreement (PPA) — a long-term contract in which a buyer purchases electricity directly from a renewable energy developer rather than a traditional utility provider. Unlike standard PPAs that rely on the national grid for delivery, this “behind-the-meter” approach uses a literal, physical wire to connect the energy source (e.g., a wind or solar farm) to the facility. This ensures that electricity never passes through the public utility’s measurement system or congested long-distance lines, guaranteeing both carbon neutrality and price stability.
The following table illustrates the cost advantage each project offers Chinese players over electricity costs in America. With electricity accounting for 40%–60% of total operating costs, Chinese AI services like Kimi, MiniMax, and DeepSeek can deliver tokens at a mere 10% of the price of overseas competitors.
For investors finding it hard to pay US$4.4 trillion for a stock, Hygon is an attractive alternative positioned at the intersection of three powerful trends: domestic substitution, the surge in demand for AI inference, and the export of Chinese AI infrastructure.
China Mobile vs Verizon
China Mobile and Verizon are the leading telecommunications service providers of the two largest economies in the world. Both must invest tens of billions in infrastructure each year. In the past year, China Mobile spent US$21.9 billion for 5G-Advanced and 6G infrastructure, whilst Verizon invested US$17 billion, primarily to complete its 5G C-Band buildout and expand its fixed wireless access (FWA).
Apart from this functional parity, these two telecommunications behemoths have little in common. As a Chinese State-Owned Enterprise (SOE), China Mobile is operated in part as an extension of national industrial policy that prioritises universal domestic access. This results in China Mobile having nearly 1 billion users, of whom over 600 million are 5G users, and charging a phenomenally low Average Revenue Per User (ARPU) of US$7 per month.
In contrast, Verizon is a shareholder-driven entity that prioritises profitability. Despite its smaller scale, it could charge an ARPU of US$148 on its 147 million subscribers by aggressively bundling mobile services with residential broadband. Despite a much higher ARPU, its ROA is half that of China Mobile’s.
Investors view China Mobile as a stable dividend stock. Guided by SOEs’ recent emphasis on shareholder returns, it has maintained a steady payout ratio above 70%, delivering dividend yields of 8%-9%. Such a generous dividend policy has been made possible by its strong free cash flows of US$12 billion per annum and a net cash position of US$30.5 billion.
Verizon investors also buy the stock for dividends. The dividend payout ratio and dividend yield are slightly lower, at 66% and 6.7%, respectively.
Verizon relies primarily on debt to fund its infrastructure investments and spectrum acquisitions, whilst China Mobile uses its cash flows. As a result, Verizon carries over US$162 billion in net debt.
Verizon traded at 19 times 2025 P/E ex-net cash, whilst China Mobile traded at just 10 times 2025 P/E. Despite a steady track record of earnings growth, a higher ROA, a stronger balance sheet, and a much larger user base, China Mobile trades at half of Verizon’s two times P/B.
One can say with a high degree of certainty that investors have consistently misvalued the two stocks, despite China Mobile’s clear fundamental superiority. Strangely, the two stocks have similar market caps: China Mobile at US$222 billion and Verizon at US$214 billion.
Not surprisingly, China Mobile’s share price has significantly outperformed Verizon’s over the last five years, but the mispricing remains stark, as evidenced by the metrics in the table below.
A tale of two valuations
This paper examines the seemingly different valuation methodologies used to price tech companies on either side of the Pacific Ocean. Or, to put it another way, the risk premiums used for the two sets of companies differ disproportionately. Admittedly, the fortunes of tech companies like Nvidia and Intel can change unexpectedly. Therefore, one can argue that it is never easy for analysts to estimate future earnings growth rates. Be that as it may, we used China Mobile and Verizon to show that investors have mispriced this pair of stocks for the last five years. And the same mistake is still being made. The two companies’ industry dynamics and business fundamentals are repeatedly analogous, and therefore so is their earnings predictability.
Chinese stocks have outperformed US stocks since September 2024. Will China stocks continue to be re-rated, or will they remain dogged by the high risk premium attached to them?
Wong Kok Hoi is the founder, executive chairman and chief strategist of APS Asset Management
