CXMT, Zhipu AI and Unitree: Three case studies of China’s tech rise

Property was once the most important growth engine of China’s economy, but it is now rapidly exiting the stage. In 2025, national property development investment fell -17.2% y/y; meanwhile, high-tech manufacturing is gradually becoming a new source of growth. This does not mean high tech can yet fill the gap left by the property downturn in aggregate terms, but the trend is unmistakable: high tech is moving from a relatively peripheral industrial-policy narrative towards the centre stage of economic growth and capital markets.

Chart 1: China’s Property Development Investment Has Kept Falling

Policy, of course, is one of the key drivers of China’s economic transition. From “Made in China 2025” to the Big Fund Phase III stepping up investment in new technology companies, and from the STAR Market’s registration-based system to capital-market exit reforms such as Chapter 18C in Hong Kong, all of these have created a more favourable financing and business environment for hard-tech companies.

However, attributing this transition simply to policy support or state power would be an oversimplification. The rise of hard tech equally depends on the long-term commitment of patient capital, the steady supply of engineering and science graduates from China’s education system, a complete manufacturing supply chain accumulated over decades, and the early demand created by domestic substitution. Viewing it merely as the product of “policy subsidies” makes it easy to underestimate the depth and resilience of China’s hard-tech industrial ecosystem.

This article takes CXMT, Zhipu AI and Unitree Robotics — three representative companies — as samples to illustrate the three core engines of China’s hard-tech ecosystem: the long-term commitment of state and patient capital; the combination of research commercialisation and domestic-substitution demand; and the innovative capacity nurtured by a complete manufacturing supply chain and market-driven competition.

CXMT: A Homegrown DRAM Story Underpinned by Patient Capital

ChangXin Memory Technologies (CXMT) is China’s most representative memory manufacturer and a rising latecomer in the global DRAM market. Compared with international giants such as Samsung, SK hynix and Micron, CXMT started late and had to build its technology, capacity and customer base from scratch; but precisely because of this, its rise better illustrates what “capital” and “patience” mean for hard-tech industries.

Talent and Technology

CXMT’s technological starting point came from a flow of talent and technology spanning China, Germany, the United States and South Korea. Founder Zhu Yiming is himself a returnee entrepreneur: he graduated from Tsinghua University’s Department of Physics, earned a master’s degree in electrical engineering in the US, and worked at MoSys in Silicon Valley; in 2005 he returned to China to found GigaDevice, before taking on the even higher-barrier memory industry. For CXMT, the hardest part of DRAM lies not only in design and process technology, but in the vast amount of tacit knowledge required for mass production. The company later took over the patents and technical documentation left behind after the bankruptcy of Germany’s Qimonda and brought in overseas engineers with mass-production experience, gradually building up its early technology base.

Talent and technology alone, however, are not enough to carry a memory company to real maturity. The memory industry is by nature capital-intensive, long-cycle and highly volatile: from R&D ramp-up and yield improvement to capacity expansion, it often takes years of investment with no immediate returns. In other words, to cross the long journey from technological catch-up to profitable mass production, CXMT needed more than a technological starting point — it needed long-term capital able to absorb early losses and accompany the company through the cycle.

Capital and the “Hefei Model”: Patient Capital Carrying a Long-Cycle Industry

CXMT is a textbook case of how state capital can support a capital-intensive, long-cycle and highly volatile hard-tech industry. The Hefei government has been dubbed “an investment bank disguised as a government”, with BOE and NIO among its signature deals: the former became a global leader in display panels, while the latter became a flagship of China’s EV upstarts. At the core of the Hefei model is using state capital to anchor a leading company, building supply-chain clusters around that leader, and accompanying the company’s growth with long-horizon, low-exit-pressure capital. After 2016, Hefei replicated this model with CXMT.

Unlike private equity funds constrained by exit deadlines, Hefei’s state capital can withstand long investment cycles and early losses. When CXMT’s predecessor, “Project 506”, was launched in 2016, Hefei state capital funded roughly 80% of the first phase; although subsequent funding rounds diluted these holdings, they never trimmed their stakes or exited. The prospectus shows CXMT’s accumulated losses once reached RMB 36.65 billion. It was precisely this low-exit-pressure, long-horizon, companionship-style capital that helped CXMT through the memory industry’s toughest investment period.

More importantly, what Hefei provided was not just funding but a complete industrial support system. The local government brought packaging and testing, specialty gases, wafer reclaim and upstream equipment suppliers to settle around the fab. Among them, packaging and testing firm Xinhui derives more than 90% of its revenue from CXMT, and some upstream equipment makers are even directly controlled by state capital. This “anchor the leader + cluster the supply chain” approach gave CXMT, from the very start, a local industrial ecosystem with relatively controllable costs and high coordination efficiency.

We noted in "Taiwan Outlook 2H26: Taiwan’s irreplaceable AI supply chain and why the re-rating is not over" that a complete ecosystem is one of the most critical competitive advantages of Taiwan’s semiconductor industry. Likewise, for CXMT, capital support certainly matters, but what truly enabled it to start from zero and keep closing the gap with international giants is the local industrial ecosystem formed by capital, capacity and the supply chain together.

Related Article: Memory Supercycle | AI Tokenomics and the Next Memory Repricing Opportunity

Demand and Domestic Substitution: A Strategic Gap Providing Long-Term Support

CXMT, of course, carries clear strategic significance for the state. The DRAM market has long been dominated by Samsung, SK hynix and Micron; although China is one of the world’s largest memory consumption markets, DRAM localisation remains low. Against the backdrop of escalating US export controls, restricted HBM supply to China and the addition of some Chinese companies to the Entity List, a self-reliant and controllable supply chain has become a national-level strategic need.

Lifted by capital, ecosystem and demand together, CXMT has gradually grown into a major player in the global memory industry. Over the past two years, AI computing demand has driven memory prices higher; CXMT’s revenue has kept climbing and it recorded its first profit, with revenue reaching RMB 50.8 billion in Q1 2026. The company listed on the STAR Market on 27 July 2026, raising RMB 57.9 billion in its base offering — rising to approximately RMB 66.6 billion with the overallotment exercised — making it the largest IPO in STAR Market history, surpassing SMIC's previous listing.

Table 1: CXMT’s Major Shareholders (Holding 5% or Above)

Shareholder	Shareholding	Background
Qinghui Jidian	21.67%	Hefei state-owned capital background
Changxin Jicheng	11.71%	—
Big Fund Phase II	8.73%	National Integrated Circuit Industry Investment Fund Phase II
Hefei Jixin	8.37%	Includes employee incentive platform
Anhui Provincial Investment	7.91%	Anhui provincial state-owned capital
Source: CXMT and iFAST Financial Compilation Data as of 6 July 2026

Zhipu AI: A Large-Model Story Jointly Driven by Research Commercialisation and Domestic-Substitution Demand

If CXMT’s case shows the importance of state capital and patience, Zhipu AI demonstrates the combined push of China’s research commercialisation, hybrid capital backing and domestic-substitution demand. Across Zhipu’s talent, capital and customers, the joint participation of the state system and market forces is clearly visible.

Zhipu’s technology and core team came mainly from Tsinghua University’s Knowledge Engineering Group (KEG). KEG is a leading Tsinghua laboratory with a long track record in knowledge engineering, natural language processing, knowledge graphs and AI research, and one of China’s earliest research platforms to bridge foundational AI technology and industrial applications. Zhipu’s founding team also carries a strong Tsinghua pedigree, with four of its nine founding members coming from the university.

Zhipu’s journey from research project to commercial company is closely tied to China’s system for commercialising research outcomes. Around 2018, China began allowing serving researchers at academic institutions to start businesses based on their existing research, with proceeds shared between the institution and the team. In other words, the national research system supplied the talent and technological accumulation, while the commercialisation policy opened an institutional window — and Zhipu was one of the earliest large-model companies to benefit from this arrangement and take university AI research into industry.

Zhipu’s customer structure also carries a distinct state-capital flavour. Unlike some large-model companies focused on overseas consumer applications or general-purpose model APIs, Zhipu positions itself towards customers with higher requirements for data security, self-controllability and on-premise deployment. A substantial share of these customers have state-owned or government backgrounds, including central SOEs, local state-capital platforms and government departments at various levels, forming a key source of its revenue and orders.

In 2024, as government departments and large SOEs procured large-model products and services through open tenders, Zhipu ranked among the industry leaders in the number of contracts won. Its customers come mainly from central and state-owned enterprises and government agencies, including financial institutions such as Postal Savings Bank of China and China Merchants Bank, as well as the Three Gorges Corporation’s hydropower operations-and-maintenance large-model project; in government services, Zhipu has also partnered with Huaxin Yongdao to develop vertical models for government services. The company runs a dedicated “government clients division” serving customers with higher requirements for data security, system controllability and customised deployment.

More notably, Zhipu has attracted investment from local governments and state-capital platforms across multiple regions, forming deep cooperation with related local projects. For example, Chengdu Hi-Tech Zone invested RMB 300 million in Zhipu, alongside a partnership to build the “Zhipu Zhuge large model”, with plans for a training centre and a western-China enablement platform aiming to help more than 1,000 companies pursue “AI+” transformation within five years. This state capital provides not only funding, but binds itself to Zhipu through foundation-model development, computing-centre rollouts and industrial-ecosystem incubation.

In sum, Zhipu AI’s technical team, shareholder structure and customer base all carry a clear “national team” imprint, reflecting that its rise rests not on market forces alone, but on the combined push of research commercialisation, hybrid capital backing and domestic-substitution demand.

Unitree: A Robotics Giant Born of a Complete Manufacturing Supply Chain

CXMT relied on patient capital; Zhipu on research commercialisation and SOE demand. Unitree is the most market-driven of the three: its rise came not from state capital absorbing years of losses, but from standing on the hardware supply chain China has built up over decades, breaking out quickly through productisation, mass-production capability and cost advantages.

Market-Driven Origins and Profitability

Unitree was founded in 2016 by former DJI employee Wang Xingxing, starting with quadruped robots before gradually expanding into humanoid robots. Unlike CXMT, which long depended on state capital to absorb losses, Unitree is closer to a market-driven growth story. Its prospectus shows the company already achieved a profit of RMB 600 million in 2025. This means Unitree’s valuation is not propped up by subsidies or policy resources alone — it has already proven a degree of competitiveness on both the product and commercialisation fronts.

Supply Chain Spillover Effects

Unitree’s rapid breakout is not just about making a single good product. More importantly, it stands on the complete hardware supply chain China has accumulated over the past decades. From drones and new energy vehicles to motors, batteries, sensors and precision manufacturing, China has built a hardware ecosystem of enormous scale, low cost and fast response — and this is exactly what enables Unitree’s robots to iterate rapidly and drive down costs.

This supply-chain advantage stems mainly from two parent industries. The first is the Shenzhen drone ecosystem led by DJI. To make consumer drones both cheap and reliable, DJI and the South China supply chain spent years driving down the cost and raising the efficiency of components such as small BLDC motors, electronic speed controllers, flight controllers, sensors and gimbals, nurturing more than three thousand drone component suppliers. Although these components initially served drones, once the robotics industry took off, the technologies and suppliers could be redeployed quickly.

The second is the new energy vehicle industry, represented by BYD. In 2024, China produced 31.3 million vehicles, of which around 40% were NEVs. The sheer volume pushed BLDC motors, drives, encoders, power batteries and precision mass-production processes towards scale and low cost. In other words, many of the core hardware capabilities robots require had already been honed over and over in the drone and NEV industries.

More critically, this supply chain is no longer the advantage of any single company, but of the entire industrial ecosystem. Today, multiple Chinese provinces host makers of spec-compatible reducers and high-torque motors, and roughly 200 humanoid robot companies share the same supply chain, in turn expanding the industrial network. This cost and delivery advantage has even spilled over into global markets: several US robotics startups have chosen to base their hardware teams in China, and even Tesla’s Optimus sources some components from the Chinese supply chain. Unitree, in other words, is not fighting alone — it stands on an industrial chain that even global competitors can hardly bypass.

On top of this ecosystem, Unitree has gone a step further with vertical integration, developing in-house core components such as BLDC motors, planetary reducers, LiDAR and depth cameras — segments where many Chinese peers still rely mainly on external purchases. Unitree’s self-produced motors can cost as little as 30–40% of comparable Western products, and it has launched one of the world’s cheapest humanoid-robot reducers. Scale production has also strengthened its upstream bargaining power, lifting its quadruped-robot gross margin from 42.36% to 55.49%.

Unitree’s moat, therefore, is not the performance of any single product, but a systemic advantage formed by China’s hardware supply chain, low-cost components, fast delivery capability and its own vertical integration.

A Decade-Long Industrial Strategy Enters Its Proving Period

As the property engine exits the stage, can hard tech truly take over? CXMT, Zhipu AI and Unitree seem to offer an answer. The rise of Chinese companies does not rest on policy subsidies or state capital alone, but on an industrial ecosystem jointly supported by patient capital, research commercialisation, domestic-substitution demand and a complete manufacturing supply chain.

The three companies represent three paths within this ecosystem: CXMT is the patient-capital story with the deepest policy involvement, showing how state capital can nurture capital-intensive, long-cycle yet strategically significant industries; Zhipu AI is driven jointly by research commercialisation, hybrid capital and government-and-enterprise demand; Unitree is the most market-driven, breaking out quickly on the back of China's hardware supply chain, low-cost mass production and vertical integration.

As early as 2015, "Made in China 2025" listed next-generation information technology and robotics among its priority areas; although such language later gradually faded from high-frequency official discourse, its industrial direction has carried through to today's "new quality productive forces". CXMT, Zhipu and Unitree all sit within these strategic tracks. With Zhipu and CXMT now both having reached the capital markets, and Unitree expected to follow in due course, this may mark the point at which a decade-long industrial strategy formally enters its results-validation period. Whether more new hard-tech companies will emerge is worth the market's continued attention.

These three companies are examples chosen to show how China's hard-tech companies actually get built: through patient state capital, through research turning into a business, and through a strong supply chain. That same set of mechanisms is producing opportunities well beyond these three companies alone — from the equipment makers, materials suppliers, and component manufacturers that sit around CXMT and Unitree, to the other model developers and AI-service providers that share Zhipu's research-commercialisation and domestic-demand story.

Of the three, only Zhipu is currently accessible through a product discussed here — as a constituent of the iShares Hang Seng TECH ETF (HKEX:3067) since June 2026. CXMT, though now listed on the STAR Market, is not held by either 3067 or the GF CSI All-Share Information Technology ETF (SZSE:159939); 159939 offers broader exposure to China's onshore semiconductor and AI-hardware supply chain instead, without capturing CXMT specifically. Unitree remains privately held and is not yet accessible through either product.

Declaration:

This research report was prepared with the assistance of artificial intelligence (AI) tools. iFAST Financial Pte Ltd does not rely exclusively on AI for content generation; the content of this report – including all investment theses, ratings, price targets and conclusions – has been independently reviewed and verified by the research analyst(s) to ensure accuracy and professional integrity.

For specific disclosure, at the time of publication of this report, IFPL (via its connected and associated entities) and the analyst who produced this report hold a NIL position in the abovementioned securities.