China is not sitting on US$28 trillion of deployable AI capital. What Beijing is building is potentially more consequential: a financing architecture linking banks, bonds, state funds, household savings and public markets to the pursuit of semiconductor sovereignty and AI scale.
Executive Summary
- China’s US$28 trillion narrative reflects the scale of its bond market, not a deployable AI investment fund.
- Beijing is redirecting domestic savings, public markets, state capital and low-cost credit towards strategic technology and semiconductor development.
- CXMT demonstrates both China’s accelerating industrial capability and the valuation distortions created by scarcity, low free float and policy capital.
- The US retains decisive advantages in frontier compute, HBM, lithography, private capital and hyperscaler financing.
- For global investors, the defining risk is increasingly the redistribution of semiconductor margins, capital expenditure, energy demand and geopolitical exposure across two diverging technology systems.
Can China’s Capital Markets Finance Semiconductor Independence?
China does not have US$28 trillion to spend on artificial intelligence.
The number that has become attached to Beijing’s AI ambitions corresponds most closely to the approximately RMB196.7 trillion of bonds held in custody at the end of 2025, equivalent to roughly US$28 trillion at contemporaneous exchange rates. It is principally a stock of government, financial and corporate fixed-income claims already outstanding, not a strategic investment account that policymakers can redirect into semiconductors, data centres or foundation models.
That distinction is essential.
China’s onshore listed-equity market, estimated at roughly RMB109 trillion to RMB123 trillion at the end of 2025 depending on methodology, sits largely in addition to that bond-market figure. Simply adding the two produces financial claims far above US$28 trillion. The headline therefore tells investors more about the scale of China’s financial system than the amount of capital available for artificial intelligence.
Yet the underlying thesis is more important than the headline.
Beijing is attempting to alter the way strategic technology is financed. It is connecting a bank-dominated financial system, public-equity markets, venture capital, local and national state investment vehicles, technology bonds, household savings and Hong Kong’s offshore market infrastructure to an industrial strategy centred increasingly on AI, semiconductors, advanced manufacturing and robotics.
Bloomberg’s investigation adds a useful measure of that mobilisation. Chinese technology companies raised about US$217 billion through IPOs and bond sales over the two years to 7 August 2026, according to Bloomberg data. US counterparts raised more than six times as much, approximately US$1.4 trillion.
The gap remains vast. But the mechanism through which China is trying to close it is changing.
CXMT Corp. offers the clearest illustration. China’s leading domestic DRAM producer raised approximately RMB57.92 billion before full exercise of its over-allotment option. Its shares then closed about 466% above the IPO price on their first trading day, creating a market capitalisation above RMB3 trillion despite only about 6.73% of the equity being freely tradable.
CXMT did not receive RMB3 trillion. It received roughly RMB58 billion of primary capital.
The enormous difference between capital raised and capitalisation created captures both the promise and the danger of China’s capital-market experiment.
What the US$28 Trillion Figure Actually Represents
| Component | Scale | What It Measures | Can It Be Treated as AI Funding? |
| China bond market | RMB196.7tn, about US$28tn | Outstanding fixed-income claims at end-2025 | No |
| Onshore listed equities | About RMB109tn to RMB123tn | Market value of listed companies | No |
| Mainland IPO proceeds | RMB134.1bn in 2025 | Fresh primary equity financing | Partly, where issuers are technology companies |
| China VC and PE investment | About RMB620bn in Jan to May 2026 | Combined private investment flows across sectors | Partly, but not directly comparable with pure US VC |
| National VC guidance fund | Around RMB1tn mobilisation ambition | Target for mobilising social capital | Potentially, over time |
| Technology-innovation bonds | Around RMB600bn issued or announced by mid-2025 | Targeted technology financing | Partly |
The orders of magnitude matter. Mainland IPOs raised only about RMB134.1 billion in 2025. First-half 2026 mainland proceeds were expected to reach approximately RMB105.7 billion. Hong Kong raised roughly HK$210 billion during the first half, with technology and industrial issuers accounting for a large share of activity.
These are substantial financing flows. They are not remotely equivalent to the stock of China’s bond market.
The right question is therefore not whether Beijing can deploy US$28 trillion into AI. It is whether China can convert a relatively small portion of an enormous domestic savings and financial system into sufficiently productive risk capital to accelerate technological substitution.
That is a much harder question.
Why Beijing Is Re-engineering Capital Formation
China’s traditional financial system is poorly matched to some of the industries Beijing now considers most strategically important.
Banks remain dominant. Loans account for roughly 60% of Chinese banks’ assets, according to People’s Bank of China reporting cited in the research dossier, while bonds account for roughly a quarter. Banks are naturally more comfortable lending against assets and predictable cash flows than financing companies whose value depends on uncertain research programmes, semiconductor yields, software ecosystems or technologies that may require years of losses before commercial scale.
Artificial intelligence and advanced semiconductors are particularly demanding forms of industrial finance.
A leading-edge fab requires extraordinary upfront investment. AI accelerators demand sustained expenditure on architecture, software and manufacturing. High-bandwidth memory, or HBM, requires complex process integration and packaging. Foundation-model companies may consume capital long before monetisation becomes reliable. Humanoid robotics combines hardware, software, manufacturing and years of engineering risk.
Equity can absorb those losses. Conventional bank lending is less suitable.
China also faces a domestic reallocation problem. Property historically absorbed an extraordinary proportion of household wealth, local-government revenues and bank credit. As the property model weakened, policymakers acquired both an economic reason and a political incentive to redirect capital towards sectors associated with productivity and technological sovereignty.
Bloomberg estimates Chinese citizens hold around US$26 trillion in savings, the world’s largest household savings pool. Capital-market reform creates a possible channel through which some of those savings can move from deposits and property towards listed technology businesses.
The transition is not from state finance to free-market finance. It is from a predominantly bank and subsidy model towards a hybrid system.
A strategic company can increasingly move through local-government capital, national guidance funds, bank investment subsidiaries, venture investors, technology-innovation bonds, corporate credit and eventually STAR Market or Hong Kong equity issuance.
The national venture-capital guidance fund exemplifies that architecture. Its objective is to mobilise around RMB1 trillion of social capital over time, with an emphasis on seed-stage companies and hard technologies. The figure is a target, not money already deployed.
Bond markets are being pulled into the same project. By mid-2025, roughly 288 issuers had raised or announced close to RMB600 billion through technology-innovation bonds. Bloomberg reported that Chinese technology companies had sold at least US$38 billion of onshore and offshore bonds in 2026 through 7 August, the strongest comparable pace since 2016.
America’s technology sector had raised approximately US$578 billion during the same period.
The difference in absolute scale remains enormous. China, however, has a different advantage: cost.
Bloomberg calculated that major Chinese technology companies were issuing bonds at an average coupon of about 1.9%, more than 300 basis points below comparable US technology borrowers. CATL, for example, issued five-year renminbi notes at a 1.58% coupon, compared with a 5.25% coupon on a similar-maturity dollar bond from LG Energy Solution.
Low nominal funding costs do not solve technological problems. They can, however, change the economics of capacity expansion, research spending and industrial endurance.
STAR Market Reform and the New AI IPO Machine
The most visible element of China’s AI capital-market strategy is the transformation of the STAR Market from a specialist technology venue into an instrument of strategic capital formation.
In June 2025, Shanghai revived the fifth listing standard for strategically important companies that may not satisfy conventional profitability requirements. A year later, the Shanghai Stock Exchange issued guidance explicitly extending that framework towards artificial-intelligence large-model companies.
Eligible businesses are expected to have independently developed models, demonstrated large-scale applications and core operations centred on model development, services or applications.
The consequence is profound.
China is increasingly willing to list strategically important technology companies earlier in their financial maturation than conventional public-market standards might otherwise permit. By mid-June 2026, nearly 50 robotics and semiconductor companies were reportedly pursuing Shanghai or Shenzhen listings, targeting at least RMB126.1 billion in aggregate fundraising.
Hong Kong complements rather than competes with this model. It offers offshore liquidity, international price discovery and access to non-renminbi capital. First-half 2026 Hong Kong IPO proceeds reached approximately HK$210 billion, with specialist technology and A+H issuers playing a central role.
The result is an emerging dual-track financing architecture. STAR provides domestic scarcity value and strategic legitimacy. Hong Kong provides external capital and international market infrastructure.
Unitree’s August listing illustrates how far the model is moving beyond conventional semiconductors. The humanoid-robotics company priced its STAR IPO at RMB150.8 a share, implying a valuation of about RMB61 billion and seeking roughly RMB6.1 billion of proceeds.
AI model developers are moving towards the same pipeline. Z.AI and MiniMax have pursued mainland routes after Hong Kong listings. Moonshot AI has discussed an IPO timetable. DeepSeek has begun preparatory work while simultaneously pursuing private funding.
This is one of the most important differences between China and the United States.
America’s deepest frontier AI companies can remain private because US private markets are capable of funding them at staggering scale. China has less independent late-stage risk capital and consequently stronger incentives to use public markets earlier.
Public investors gain earlier access to potentially strategic champions.
They also inherit earlier-stage technology risk.
CXMT and China’s Semiconductor Capital Experiment
No company captures the new financing regime better than ChangXin Memory Technologies, better known as CXMT.
Founded in Hefei in 2016, CXMT has become China’s largest domestic DRAM manufacturer and the world’s fourth-largest supplier. Its global DRAM share reached approximately 7.7% in 2025, according to company prospectus data cited by Reuters.
That is not technological parity with Samsung Electronics, SK Hynix or Micron Technology. It is nevertheless commercially significant.
Memory markets are scale businesses. A producer does not need to equal the frontier leader in every product category before it starts altering supply, pricing and competitors’ return on capital.
CXMT priced its IPO at RMB8.66 a share and raised approximately RMB57.92 billion before full over-allotment. Its shares closed at RMB49 on the first day, after briefly reaching RMB55.03.
The roughly 466% first-day gain pushed its market capitalisation above RMB3 trillion.
But only 6.73% of its shares were freely tradable.
This creates a classic free-float distortion. When a small proportion of a company’s equity is available for trading and demand is exceptionally strong, the marginal transaction price can imply an enormous value for the entire company. That valuation is economically meaningful as a market signal, but it is not equivalent to the volume of investor capital that actually changed hands, much less the money available to management.
Bloomberg’s complete account adds another important nuance. CXMT’s journey from filing to trading took less than eight months after the company became the first issuer to use a preliminary-review pilot for strategically important firms. The mechanism allowed regulatory questions to be addressed before a formal IPO application.
China is therefore attempting to compress not just the cost of capital, but the time required to obtain it.
Yet the IPO also exposed the limitations of China’s listing regime. Conservative pricing left a considerable difference between the offer price and the market clearing price. Investors enjoyed the first-day surge, but CXMT arguably raised less capital than it might have obtained under a more aggressive book-building process.
South Korea’s SK Hynix, by comparison, recently raised US$26.5 billion in the US market.
That comparison goes to the heart of China versus US AI capital formation. China can create extraordinary scarcity valuations. America remains better equipped to convert deep investor demand into very large pools of primary corporate capital.
CXMT’s industrial progress is nonetheless real. Revenue surged during the memory upcycle, wafer capacity is expanding and global customers have begun examining Chinese memory products. Reuters reported that Apple had tested CXMT DRAM for possible use in devices, particularly in China.
The crucial weakness is HBM.
High-bandwidth memory sits beside advanced AI accelerators and feeds them data at extremely high speed. Modern AI compute increasingly depends on memory bandwidth as much as raw processor arithmetic. This is why HBM has become one of the most strategically important products in the semiconductor industry.
CXMT remains materially behind SK Hynix, Samsung and Micron in cutting-edge HBM.
That gap demonstrates the boundary between capital mobilisation and technological capability.
Money can build fabs. It can finance engineers, process tools, packaging lines and repeated experimentation. It cannot instantaneously manufacture accumulated process knowledge.
The Semiconductor Stack and the Chokepoints Capital Cannot Instantly Solve
China’s semiconductor position is best understood layer by layer.
Huawei and HiSilicon sit at the centre of the domestic AI accelerator push. Huawei’s Ascend processors are increasingly viable for Chinese workloads, particularly when domestic software developers optimise around the hardware. Demand accelerated as companies including ByteDance, Tencent and Alibaba sought alternatives to restricted Nvidia products.
Cambricon provides a publicly listed merchant-chip route. Moore Threads, Biren and MetaX expand the domestic accelerator universe further.
At the memory layer, CXMT has become relevant in DRAM and YMTC has established roughly 13% of the global NAND market as of Q1 2026, according to industry data reported by Caixin.
At the foundry layer, Semiconductor Manufacturing International Corporation remains China’s most important advanced logic manufacturer, with Hua Hong stronger in mature and speciality nodes. SMIC has demonstrated production around the 7nm class, largely using extremely complex deep-ultraviolet lithography processes.
The frontier has already moved beyond that.
The hardest barrier is extreme ultraviolet lithography, or EUV.
ASML remains the only commercial supplier of EUV systems at scale, and Dutch export controls prevent shipments of those machines to China. Without EUV, Chinese fabs must use more complicated DUV multi-patterning, which increases the number of processing steps, capital intensity, cycle time and defect opportunities.
Greater financing cannot eliminate that problem on demand.
The same applies across parts of advanced semiconductor manufacturing equipment. US controls cover categories including deposition, etching, lithography, metrology, inspection and production software.
Chinese suppliers are nevertheless making measurable progress in areas where technological substitution is more achievable.
AMEC and Naura are gaining ground in etch, deposition and related equipment categories. Reuters reported that Samsung and SK Hynix had tested Chinese semiconductor tools as a hedge against uncertainty surrounding US equipment access. Analyst estimates cited in that reporting suggested Chinese vendors could eventually capture 25% to 30% of China’s wafer-fabrication equipment market, although those figures remain estimates rather than audited shares.
KLA-class inspection and metrology remain harder to replace. Lithography is harder still.
The technology hierarchy matters for investors because export restrictions do not have the same economic effect across every semiconductor layer.
Some controls can be genuinely binding.
Others raise cost but encourage domestic scale.
This is particularly important in AI inference.
Frontier model training requires extremely powerful accelerators, enormous memory bandwidth and highly optimised clusters. Inference, where trained models actually respond to users and business systems, can offer more room for architectural substitution, lower-cost hardware, optimised software and system-level engineering.
Huawei’s approach increasingly reflects that reality. Rather than relying exclusively on one chip matching Nvidia at the transistor level, China can combine more accelerators, networking, packaging and system architecture to narrow application-level performance gaps.
The trade-off is greater power consumption and potentially worse compute economics.
Advanced packaging works similarly. Chiplets and sophisticated integration can partially compensate for weaker process nodes by joining specialised silicon into larger systems. Again, this does not erase the underlying manufacturing disadvantage. It changes how the disadvantage is managed.
Electronic design automation, or EDA, also requires nuance. Western software remains important, but restrictions have not produced a total, permanent exclusion of all EDA tools from China. The strategic vulnerability lies particularly in frontier productivity and advanced-node design workflows.
China’s semiconductor challenge is therefore not a single technology gap. It is a ladder.
China is increasingly competitive in commodity memory, optical interconnect, mature-node production, selected fabrication equipment, AI inference and manufacturing scale.
It remains materially disadvantaged in EUV, frontier HBM, leading-edge high-yield logic and selected inspection, design and process technologies.
The central investment question is whether progress on the lower and middle rungs becomes commercially disruptive before China reaches the top.
It probably can.
China Versus US AI Financing Architecture
| Financing Channel | China | United States | Strategic Implication |
| Public equity | STAR reform, Hong Kong listings, earlier access to strategic issuers | Deep public markets, but frontier AI often remains private longer | China shifts more early technology risk into public markets |
| Venture capital | Growing domestic VC, state guidance funds and policy capital | US$320bn VC in 2025, with AI about 65.4% of value | US remains much deeper in private risk capital |
| State capital | National and local guidance funds, strategic investment vehicles | More limited direct state-equity role | China can tolerate lower near-term private returns |
| Corporate balance sheets | Alibaba, Tencent, ByteDance and others expanding AI capex | Hyperscalers indicating roughly US$725bn of 2026 capex | US corporate cash generation remains a major advantage |
| Bond markets | Cheap renminbi funding, technology bonds and policy encouragement | Deep investment-grade markets and enormous tech issuance | China has cheaper coupons, US has far greater absolute capacity |
| Private markets | Expanding but less mature late-stage ecosystem | Mega-rounds, private credit, growth equity and infrastructure funds | US frontier AI can remain private for longer |
| Infrastructure financing | Banks, state capital, technology bonds and public markets | Bonds, leases, private credit, project finance and structured vehicles | AI is becoming a multi-asset financing theme in both systems |
America’s Financing Advantage Remains Formidable
China’s capital-market transformation is significant precisely because of the system it is trying to compete against.
US venture investment totalled approximately US$320 billion in 2025, with AI accounting for roughly 65.4% of value. PitchBook and the National Venture Capital Association reported more than US$400 billion of US venture investment during the first half of 2026.
Private valuations have reached levels previously associated only with the largest listed corporations. Anthropic completed a financing at a reported US$965 billion post-money valuation in May 2026. OpenAI’s referenced March transaction valued it at approximately US$852 billion.
China’s DeepSeek, by comparison, was reported to be discussing an approximately RMB500 billion, or roughly US$74 billion, valuation in its latest fundraising process.
These are transaction valuations, not intrinsic values. The comparison is nevertheless revealing.
America’s frontier AI system can fund enormous companies without requiring them to list.
Then come the hyperscalers.
Amazon, Microsoft, Alphabet and Meta were collectively indicating up to approximately US$725 billion of capital expenditure for 2026. Bloomberg’s reporting shows why access to capital remains a strategic US advantage even as financing costs rise.
The model is also evolving beyond operating cash flow.
Major US technology companies issued roughly US$194 billion of bonds through early July 2026, according to Reuters reporting contained in the research dossier. Future data-centre lease commitments among Microsoft, Meta, Oracle, Amazon and Alphabet were estimated at roughly US$1.09 trillion.
AI has therefore moved onto corporate balance sheets, bond markets, real-estate finance and private credit.
China’s model is materially cheaper but smaller.
Bloomberg’s comparison of technology debt illustrates the trade-off: Chinese technology issuers had raised about US$38 billion of bonds in 2026 through 7 August, versus US$578 billion in the US, but average Chinese coupons were around 1.9%, more than 300 basis points below US peers.
China does not need to recreate Silicon Valley exactly.
It needs to finance substitutes cheaply enough, for long enough and at sufficient industrial scale to change market economics.
When Strategic Capital Becomes Speculative Capital
The strongest argument against the bullish China AI capital-markets thesis is not that Beijing lacks capital.
It is that capital may become too abundant relative to technologically productive opportunities.
CXMT’s first-day rise is the obvious warning. A 466% closing gain combined with a 6.73% free float suggests that scarcity and policy significance contributed heavily to price discovery.
The wider market displays similar features.
At the end of 2025, the STAR Market comprehensive index traded at a price-to-earnings multiple around 67 times, compared with roughly 16.3 times for Shanghai’s main market.
Chinese VC and PE investment reached approximately RMB620 billion during the first five months of 2026, almost 60% above the prior-year period according to ChinaVenture data cited by Reuters. Capital has increasingly crowded into robotics, quantum technology, fusion, commercial space and other policy-designated industries.
China has seen this pattern before.
State support helped build formidable electric-vehicle, battery and solar industries. It also helped create extraordinary capacity, intense price competition and margin compression.
Technological sovereignty and shareholder returns are not the same objective.
A government can rationally accept low private returns in exchange for employment, supply-chain resilience, technological independence or national security. A private investor cannot assume those same strategic benefits will accrue to minority shareholders.
The memory sector offers a particularly important risk.
CXMT and YMTC do not need to surpass every global competitor technologically to affect returns. They only need to add enough incremental supply.
SK Hynix, Samsung and Micron currently benefit from extraordinary AI-related memory demand, especially HBM. If Chinese capacity expands into a later period when demand growth normalises, commodity memory could again confront severe oversupply.
Industrial success for China could therefore mean weaker profitability for the industry as a whole.
The same tension applies to fabrication equipment.
AMEC and Naura can gain domestic share while forcing Applied Materials, Lam Research and other incumbents to surrender parts of the Chinese market. Yet policy-driven expansion can eventually create too many domestic competitors, reducing margins even among the firms that succeed technologically.
The correct framework is not that China’s AI equity market is fictitious.
It is a genuine technology boom containing substantial valuation distortions.
Global Semiconductor Winners, Losers and Cross-Asset Transmission
The US-China AI race is already redistributing economic rents.
Nvidia faces the clearest strategic tension. Domestic substitution by Huawei and Cambricon reduces its potential Chinese addressable market. Yet extraordinary AI spending outside China can more than offset that pressure at the global level.
AMD faces a similar, though smaller, trade-off.
Micron, Samsung and SK Hynix face a more direct long-horizon challenge because Chinese memory capacity can affect global pricing even without technological parity.
ASML occupies an unusual position. Its EUV monopoly gives it one of the most consequential technological moats in global industry. Export restrictions simultaneously preserve that strategic scarcity and prevent ASML from fully monetising Chinese demand.
TSMC remains another paradox. It is arguably the most indispensable manufacturing company in the frontier AI economy, yet its Taiwan concentration creates one of the largest geopolitical tail risks in global portfolios.
Applied Materials, Lam Research, KLA and Tokyo Electron sit between AI capex growth and substitution pressure. Their global businesses can benefit from the semiconductor investment boom while local Chinese competitors gradually reduce Western share within China.
The semiconductor cycle itself remains powerful. The Semiconductor Industry Association reported global semiconductor sales of approximately US$791.7 billion in 2025, up 25.6%, and expected the industry to approach US$1 trillion in 2026. That is an industry forecast, not a guaranteed outcome.
The larger cross-asset story begins beyond equities.
Data-centre construction requires power generation, transmission, cooling, transformers, switchgear and enormous quantities of electrical equipment.
The International Energy Agency expects global data-centre electricity demand to more than double towards approximately 945 TWh by 2030.
Copper therefore becomes connected to AI through transmission networks, substations, power distribution and data-centre electrical systems. Natural gas and nuclear power enter the equation because compute requires increasingly reliable dispatchable generation. Cooling infrastructure becomes a capital-intensive technology category of its own.
China possesses another potential strategic advantage here. It has extraordinary manufacturing capacity in electrical equipment and is expected by the IEA to account for almost half the increase in global electricity demand through 2030.
The US is building the world’s most capital-intensive frontier AI infrastructure.
China may have a superior ability to industrialise parts of the physical supply chain surrounding it.
Fixed income is also moving to the centre of the theme. Hyperscaler bonds, long-duration leases, infrastructure credit and private lending mean investors cannot treat AI merely as an equity-duration story.
The renminbi poses a different question.
Successful Chinese technological upgrading could improve productivity expectations, support capital inflows and reduce selected import dependencies. None of this implies imminent reserve-currency displacement of the US dollar.
Foreign institutions held only about 1.8% of China’s bond market at the end of 2025. Capital controls remain meaningful. Global reserve management, collateral markets and cross-border financing remain overwhelmingly dollar-centric.
Hong Kong provides the compromise. It permits selective international access to Chinese growth without requiring full capital-account liberalisation.
Gold sits further out on the transmission chain. AI demand itself does not create a direct gold thesis. Escalating sanctions, Taiwan risk, reserve diversification and financial fragmentation can.
The investment regime is therefore multi-asset by construction.
What China’s AI Capital-Market Transformation Means for Global UHNW Portfolios
For UHNW investors and family offices, the first task is not to determine whether China or America will “win” the AI race.
It is to identify where exposure already exists.
A globally diversified portfolio benchmarked to major equity indices may already contain significant implicit concentration in Microsoft, Nvidia, Alphabet, Amazon, Meta and other businesses whose earnings and valuations increasingly depend on sustained AI capital expenditure.
That creates a hidden common factor.
A portfolio can appear diversified across software, cloud computing, semiconductors and internet platforms while remaining heavily exposed to one underlying proposition: that extraordinary AI infrastructure investment will ultimately generate adequate returns on capital.
China creates another dimension of that risk.
A-share exposure provides access to semiconductor equipment, memory, robotics, optical networking and advanced manufacturing, but introduces policy sensitivity, retail participation, capital controls and occasional free-float distortions.
Hong Kong provides a more internationally accessible route to Chinese internet and technology companies, as well as an increasingly important listing venue for specialist AI and semiconductor businesses.
Indirect Chinese exposure can be just as relevant. A global portfolio may carry China sensitivity through Korean memory, Dutch lithography, Taiwanese foundries, Japanese equipment, copper, power systems or companies whose Chinese revenues depend on the future scope of export controls.
Semiconductor diversification therefore requires more than owning companies from several countries.
A nominally global portfolio can still depend on US chip architecture, Taiwanese fabrication, Korean HBM, Dutch lithography and Chinese end demand simultaneously.
Private markets add another asymmetry.
US frontier AI remains disproportionately private. Reported valuations approaching US$1 trillion create unusually high entry valuations and liquidity risk for private investors.
China is transferring selected frontier companies into public markets earlier, creating the opposite problem: liquidity may be better, but scarcity premiums and immature earnings can produce extreme public valuations.
The financing layer deserves equal attention.
AI infrastructure private credit investment risks include leverage, data-centre obsolescence, counterparty concentration, power availability, lease economics and residual asset values. A facility designed for one generation of accelerators may not retain its economic value if power density and cooling architectures change faster than expected.
Family offices also need to consider liquidity.
A portfolio containing private AI funds, long-duration infrastructure credit, concentrated semiconductor equities and illiquid real assets may possess far more common risk than traditional asset-class labels suggest.
Currency management belongs in the same framework. Renminbi exposure, Asian semiconductor-linked currencies and the US dollar respond differently to trade restrictions, capital flows, interest-rate differentials and geopolitical shocks.
Geopolitical hedging is similarly about portfolio architecture rather than forecasting conflict. The relevant questions include dependence on a single semiconductor geography, access to liquidity during market dislocation, cross-border custody, options-based protection, commodity exposure and the ability to rebalance across jurisdictions.
For sophisticated private capital, visibility across these exposures increasingly matters as much as directional conviction. Bancara’s multi-asset infrastructure, including BancaraX, MetaTrader 5 and TipRanks, is designed around the ability to monitor equities, indices, currencies and commodities within a broader cross-market risk framework rather than treating technology exposure in isolation.
UHNW Portfolio Exposure Map
| Exposure | Opportunity | Principal Risk | Key Indicator |
| China hard-tech equities | Domestic semiconductor and AI substitution | Valuation, policy and free-float risk | STAR fundraising and earnings delivery |
| Hong Kong technology | Offshore access to Chinese AI and technology | Geopolitics and earnings execution | IPO activity and Stock Connect flows |
| US mega-cap AI | Frontier technology and global scale | Capex returns and concentration | Free cash flow relative to capex |
| Global semiconductors | AI infrastructure growth | Cyclicality and geopolitical concentration | Orders, inventories and pricing |
| Asian memory | HBM and memory demand | CXMT and YMTC competition | HBM pricing and Chinese market share |
| Semiconductor equipment | Critical manufacturing chokepoints | Export controls and domestic substitution | Chinese equipment localisation |
| AI infrastructure credit | Contracted cash flows and structural demand | Leverage, obsolescence and lease risk | Credit spreads and utilisation |
| Power and grid assets | Data-centre electricity growth | Regulation and capital intensity | Grid connections and capacity additions |
| Copper and electrification | Transmission and infrastructure demand | Global growth sensitivity | Grid and data-centre capex |
| RMB and Asian FX | Productivity and regional growth exposure | Capital controls and geopolitical flows | Foreign holdings and cross-border flows |
Four Scenarios for the Next Phase of China-US AI Competition
Bull case: capital formation produces commercially competitive substitution
Chinese capital markets repeatedly fund companies capable of gaining durable market share. CXMT and YMTC expand memory presence, Huawei and Cambricon improve system economics, domestic fabrication-equipment suppliers gain qualification and advanced packaging compensates for part of the lithography disadvantage.
The result would not require China to achieve absolute frontier parity. Commercially viable substitutes could be sufficient to reduce Western share inside China, sustain infrastructure demand and strengthen selected RMB assets at the margin.
Base case: selective success with persistent chokepoints
This is the scenario most consistent with the evidence available at the 10 August 2026 research cut-off.
China becomes increasingly competitive in inference accelerators, commodity memory, mature-node semiconductors, networking, robotics and selected equipment categories. EUV, frontier HBM, leading-edge logic yields and parts of the design and inspection stack remain serious constraints.
Returns within Chinese technology become highly dispersed. Some companies evolve into genuine global competitors. Others retain strategic importance without ever earning attractive returns on invested capital.
The United States retains the frontier but loses monopoly-like economics in selected layers below it.
Bear case: capital grows faster than technological productivity
Public valuations outrun earnings. Local governments and guidance funds finance duplicated capacity. Low free floats amplify speculative price discovery. The global AI capital-expenditure cycle slows just as new semiconductor and memory supply enters the market.
The result could be valuation compression across Chinese AI equities, weaker memory pricing, disappointing returns on semiconductor capacity and a broader reassessment of AI infrastructure spending globally.
China could still gain technological sovereignty under this scenario while shareholders suffer.
Geopolitical stress case: financial and technological bifurcation accelerates
Technology restrictions broaden. Capital restrictions intensify. Taiwan risk raises the global semiconductor risk premium. China responds with greater domestic substitution or restrictions involving strategic materials and supply chains.
Global semiconductor investment becomes less efficient because production capacity is duplicated across jurisdictions. The renminbi faces greater external pressure while geopolitical hedges become more important.
The cost of technological sovereignty rises for both sides.
The Strongest Counter-Thesis
The bullish interpretation of China AI capital markets deserves serious resistance.
- First, a US$28 trillion bond market is not a US$28 trillion technology fund.
- Second, capital cannot simply purchase EUV systems when export restrictions prohibit access.
- Third, the US financing advantage is greater than public-market comparisons suggest because frontier AI is financed through venture capital, private equity, corporate cash flows, investment-grade debt, infrastructure finance and private credit.
- Fourth, low free floats can create market capitalisations that overstate the depth of liquid investor demand.
- Fifth, state-directed capital can reduce failure discipline.
- Sixth, earlier public listings may transfer technology risk from venture investors and government vehicles onto retail and institutional shareholders before the underlying business model has matured.
- Seventh, China’s capital markets remain far less internationally integrated than those of the United States.
- Eighth, geopolitical restrictions can create a persistent valuation discount even for technologically successful companies.
- Ninth, industrial policy success and shareholder success must be evaluated separately.
China may be capable of achieving semiconductor sovereignty faster than it achieves attractive semiconductor shareholder returns.
That is the central counterweight to the entire investment thesis.
What Investors Should Monitor
The next phase will be determined less by headline market capitalisation and more by operating evidence.
CXMT’s DRAM share matters, but margins and return on invested capital matter more.
Chinese HBM production should be watched for verified commercial volume rather than laboratory claims.
Huawei and Cambricon adoption must be evaluated through real customer deployment, software maturity and system economics.
SMIC’s progress should be assessed through yield, cost and power efficiency, not simply nominal process-node labels.
AMEC, Naura and other Chinese equipment suppliers need to demonstrate wider qualification across critical fabrication steps.
STAR Market fundraising should be judged alongside aftermarket performance and institutional participation.
A-share margin financing is an important signal of whether technology enthusiasm is becoming leverage-driven.
Stock Connect flows help reveal whether offshore investors are embracing Chinese hard technology or retreating as geopolitical risks rise.
The US side requires equal scrutiny. Hyperscaler capex must eventually translate into revenue, operating leverage and free cash flow. Data-centre lease obligations and credit financing should be evaluated as part of the effective AI capital stack rather than relegated to footnotes.
Finally, power matters.
The durability of the AI investment cycle increasingly depends on whether grid capacity, generation, transformers, cooling infrastructure and electrical equipment can be delivered at acceptable cost.
The winners in the AI race may be determined as much by power economics and financing capacity as by benchmark scores.
The Real Race Begins After the Capital Is Raised
China’s US$28 trillion AI narrative is compelling because it compresses a complicated story into a single number. It is also financially misleading.
China does not have US$28 trillion of deployable AI capital. It has an approximately US$28 trillion bond market, an onshore listed-equity market worth roughly US$15 trillion to US$18 trillion depending on date and methodology, an enormous bank-dominated financial system, high household savings, expanding venture and private capital, state guidance funds and a government increasingly willing to redesign capital-market rules around strategic technology.
That architecture matters.
CXMT demonstrates how quickly state-supported industrial ambition can be converted into public financing and enormous market value. Huawei, YMTC, SMIC, AMEC and Naura demonstrate that export restrictions can create captive demand for domestic substitutes. STAR Market reform shows that Beijing wants public markets to absorb more of the risk traditionally carried by banks and government balance sheets.
But finance is only one input.
China still faces real constraints in EUV lithography, frontier HBM, leading-edge manufacturing economics, inspection and parts of the software stack. The United States retains exceptional advantages in private capital, hyperscaler cash generation, frontier accelerators, advanced semiconductor manufacturing and global financial-market depth.
The most plausible future is therefore not a clean Chinese victory or a successful American containment strategy.
It is a progressively bifurcated AI-industrial system in which China becomes stronger across selected layers while remaining constrained at several technological frontiers.
For global investors, the consequences will emerge through semiconductor margins, equity valuations, bond issuance, private credit, currencies, power infrastructure, copper, data centres and geopolitical risk.
The decisive question is not whether capital can help China catch up.
It can.
The question is whether China can convert strategic capital abundance into sustainable technological productivity and attractive economic returns before capital abundance itself creates the next cycle of overcapacity.
Works Cited
- https://www.pbc.gov.cn/en/3688006/5876310/2025121510120178451/index.html
- https://english.sse.com.cn/news/newsrelease/c/c_20260618_10822848.shtml
- https://english.sse.com.cn/news/newsrelease/voice/c/c_20250619_10782399.shtml
- https://www.bloomberg.com/news/features/2026-08-09/china-bets-on-ai-stocks-as-it-races-against-us-for-chip-tech-dominance
- https://www.reuters.com/world/asia-pacific/china-memory-chipmaker-cxmt-set-shanghai-debut-after-asias-biggest-ipo-2026-07-26/
- https://www.reuters.com/world/asia-pacific/what-is-cxmt-how-did-it-become-chinas-dram-champion-2026-07-27/
- https://www.reuters.com/world/china/big-chinese-tech-firms-scramble-secure-huawei-ai-chips-after-deepseek-v4-launch-2026-04-29/
- https://www.reuters.com/world/china/samsung-sk-hynix-test-chinese-chip-tools-hedge-against-us-risks-2026-08-05/
- https://www.caixinglobal.com/2026-06-04/chinese-chipmaker-ymtc-claims-13-of-46-billion-global-nand-market-102450726.html
- https://www.bis.gov/press-release/commerce-strengthens-export-controls-restrict-chinas-capability-produce-advanced-semiconductors-military
- https://home.treasury.gov/policy-issues/international/outbound-investment-program
- https://www.government.nl/latest/news/2025/01/15/klever-export-controls-on-advanced-semiconductor-manufacturing-equipment-to-be-tightened
- https://nvca.org/press_releases/nvca-releases-2026-yearbook-charts-a-venture-industry-in-transition/
- https://www.semiconductors.org/global-annual-semiconductor-sales-increase-25-6-to-791-7-billion-in-2025/
- https://www.iea.org/reports/energy-and-ai/energy-demand-from-ai