China’s AI hardware opportunity: Global compute boom meets accelerating domestic substitution

Information technology and telecommunications lead market performance

This year's A-share rally has been highly selective rather than broad-based. Capital has not returned to the market indiscriminately; instead, it has been concentrated in AI-related industries where earnings momentum and visibility are strongest. As of 30 June 2026, the CSI 300 Information Technology, CSI 300 Telecommunications and STAR 50 indices had rallied approximately 60–70% year-to-date, significantly outperforming the broader CSI 300 Index, which gained only around 8%. Traditional sectors such as consumer and financials continued to lag, underscoring the market's clear preference for AI beneficiaries. Leadership has been concentrated in AI infrastructure segments—including advanced packaging and testing, optical modules, high-bandwidth memory (HBM), AI servers, printed circuit boards (PCBs) and semiconductor equipment—rather than the technology sector as a whole.

China's rally is also closely aligned with the global AI investment cycle. Over the same period, the Philadelphia Semiconductor Index and South Korea's equity market also delivered strong gains, driven primarily by robust demand across the AI memory supply chain. Where China differs is in its earnings exposure. Rather than benefiting predominantly from memory, Chinese companies are more leveraged to AI servers, PCBs, optical modules, domestic semiconductor design and manufacturing, and the broader localisation of the semiconductor supply chain. This gives investors exposure to a distinct segment of the global AI value chain, supported by both international AI infrastructure spending and China's accelerating domestic substitution.

Index / Underlying	YTD (approx.)	Sector characteristics / Notes
CSI 300 Information Technology	approx. +64%	AI/semiconductor direction, leading theme
CSI 300 Telecommunications	approx. +67%	Optical modules / compute supply chain
STAR 50	approx. +64%	Hard tech / high semiconductor weighting
CSI 300 (broad index)	approx. +8%	Broad market subdued; capital has not returned across the board
CSI 300 Consumer	approx. -21%	Traditional sector lagging
CSI 300 Financials	approx. -15%	Traditional sector lagging
Philadelphia Semiconductor Index	approx. +101%	Global AI capex cycle theme
KOSPI (Korea Composite Index)	approx. +100%	AI memory supply chain leading globally
SK Hynix	approx. +300%	Core beneficiary of the memory super-cycle
Samsung Electronics	approx. +173%	AI memory leader

Note: Gains are approximate for the period, as of 30 June 2026. Source: iFinD.

This rally differs fundamentally from previous technology cycles. Rather than being driven primarily by valuation expansion and market optimism, it is underpinned by improving earnings and broad-based industry fundamentals as AI investment flows through the supply chain. For investors, this argues for strategic exposure to the sector through a diversified basket, reflecting a structural earnings theme rather than a short-term momentum trade.

Dual AI ecosystem: Why US and China offer complementary AI opportunities

A common misconception is to view China's AI industry as simply a follower of the United States. We believe the global AI landscape is instead evolving into two complementary ecosystems, each built around a distinct competitive advantage and investment opportunity.

The US continues to lead the frontier compute ecosystem, driven by cutting-edge GPUs, massive hyperscaler capital expenditure and the world's most advanced foundation models. Its competitive edge lies in maximising single-chip performance and pushing the technological frontier, although future expansion is increasingly constrained by power availability and grid capacity.

China, by contrast, is developing a systems-efficiency ecosystem. Limited access to leading-edge semiconductor manufacturing has shifted the focus away from pursuing the fastest individual chip towards maximising aggregate computing performance through large-scale GPU clusters, hardware-software co-optimisation and an increasingly self-sufficient domestic AI supply chain. Rather than competing directly with the US on frontier chips, China is optimising the efficiency of the entire computing system.

Put simply, the US excels at building the world's most powerful AI hardware, while China is focused on extracting the greatest computing efficiency from its available resources. These are not competing versions of the same strategy, but two distinct approaches to AI development that address different constraints and opportunities.

From an investment perspective, the distinction is equally important. Exposure to US AI primarily captures frontier semiconductors, hyperscaler infrastructure and foundation models, whereas China offers differentiated exposure to domestic semiconductor substitution, AI infrastructure, systems engineering and the broader localisation of its technology supply chain. As these ecosystems continue to evolve, we believe they should be viewed as complementary allocations within a global AI portfolio rather than substitutes for one another.

Dimension	US ecosystem: Cutting-edge compute path	China ecosystem: Systems efficiency path
Core approach	Competing through leading-edge chip performance (frontier GPUs)	Not competing on single-chip performance, but on systems (ten-thousand-GPU clusters, cluster-level effective compute)
Driving mechanism	Capital-driven, scale expansion (large-scale hyperscaler capex expansion)	Software/hardware and algorithmic efficiency (efficient, low-cost domestic models)
Models / ecosystem	Frontier large models mostly originate in the US, currently leading at the capability frontier	Full-stack domestic substitution advancing (models + compute + applications progressively self-sufficient)
Core advantage	Strength lies in 'the most powerful brain'	Strength lies in 'systems efficiency' and full-stack self-sufficiency
Main constraint	Power and grid capacity	Leading-edge process nodes (subject to export controls)
Investment implications	Frontier GPUs, hyperscaler capex	Domestic chips, AI servers, optical interconnects, PCBs, memory, application ecosystem

Cluster-level computing can be illustrated with a simple analogy. Rather than building the world's fastest supercar, imagine coordinating thousands of well-engineered family cars to move together as a single fleet. While no individual vehicle matches the speed of a supercar, the fleet can achieve greater overall throughput and resilience through scale and coordination. China's AI strategy follows a similar principle: instead of relying solely on the most advanced chips, it seeks to maximise computing performance through large-scale clusters, software optimisation and system-level integration.

Within this global AI landscape, Japan, South Korea and Taiwan each play indispensable roles in the semiconductor value chain. Japan remains a leader in semiconductor equipment and materials, South Korea dominates the high-bandwidth memory (HBM) market, while Taiwan is home to the world's leading advanced-node foundries. These markets form the hardware backbone of the global AI ecosystem, supplying critical components to both the US and China. Unlike the US and China, however, they do not possess fully integrated AI ecosystems spanning models, computing infrastructure and applications.

For investors, the implication is clear. US AI exposure provides access to frontier computing and hyperscaler-led innovation, while China offers exposure to domestic semiconductor substitution, systems engineering and AI infrastructure deployment. Japan, South Korea and Taiwan, meanwhile, provide exposure to the critical hardware enablers that underpin both ecosystems. Together, these markets represent complementary pillars of the global AI value chain rather than competing investment themes.

Industry resonance: A triple earnings-driven logic

The sector's earnings momentum is already evident in the data. According to iFinD, constituents of the CSI All-Share Information Technology Index reported aggregate net profit growth of approximately 74% year-on-year in 1Q26, demonstrating that the current rally is supported by broad-based earnings rather than a handful of outperforming stocks. Strength was widespread across the industry's key segments, with net profit for SWS Semiconductors, Software Services, and Hardware Equipment increasing by approximately 148%, 260%, and 48%, respectively. This reinforces our central investment thesis: the sector is being driven by earnings growth rather than multiple expansion.

Three structural forces are underpinning this earnings acceleration.

First, the global AI infrastructure build-out continues to generate robust external demand. Combined AI capital expenditure by North America's leading hyperscalers is expected to reach approximately USD725 billion in 2026, with several major operators nearly doubling investment as AI data centre construction accelerates. Chinese companies participate in this spending cycle not only through component exports but also via AI servers, printed circuit boards (PCBs), optical modules, power management systems, thermal solutions and other critical infrastructure. Trade data underscores this linkage: China's integrated circuit exports rose by approximately 100% year-on-year during the first four months of 2026, with May exports alone doubling from a year earlier, reflecting strong demand for both AI-related products and cost-competitive mature-node semiconductors.

Second, rapidly expanding domestic AI adoption is driving a sustained increase in inference demand. Weekly API calls to China's leading large-language models have, at times, exceeded those of their US counterparts, highlighting accelerating enterprise and consumer adoption. This trend is likely to strengthen as AI agents become more widely deployed. Unlike traditional chatbots, AI agents perform multi-step reasoning, invoke external tools and execute complex workflows, substantially increasing inference workloads and computing intensity.

Third, the memory industry has entered an AI-driven super-cycle. High-bandwidth memory (HBM), a critical component for AI computing, commands significantly higher margins than conventional DRAM, while memory prices have risen by more than 70% year-to-date. With supply expected to remain tight through at least 2027, pricing power continues to improve across the memory value chain. At the same time, demand for AI computing capacity has pushed compute rental markets into shortage, enabling domestic cloud providers and infrastructure operators to raise prices. Together, these trends point to a structural supply-demand imbalance rather than a typical cyclical recovery.

Taken together, these three drivers reinforce one another. Global AI investment expands infrastructure demand, rising domestic AI adoption increases utilisation, and the memory super-cycle supports pricing and margins across the upstream supply chain. The combined effect is stronger revenue and earnings growth for companies involved in AI servers, semiconductor equipment, PCBs, memory and domestic chip design.

This distinction is crucial for investors. Unlike previous technology rallies that relied primarily on improving sentiment and valuation re-rating, the current cycle is anchored by tangible earnings growth. As a result, the sector's long-term performance is likely to depend less on multiple expansion and more on whether companies continue to deliver on earnings expectations—a considerably more durable foundation for sustained returns.

Source: iFinD, Bloomberg, General Administration of Customs, and other public data.

China's unique structural endowments: Domestic substitution, policy, and energy

The previous section explained where demand comes from. The more important question is why China is well positioned to capture that demand. We believe the answer lies in three enduring structural advantages: accelerating domestic substitution, sustained policy support and a growing energy advantage. Together, these factors reinforce one another to create a competitive ecosystem that is difficult for other markets to replicate.

Domestic substitution and system-level breakthroughs

China's response to semiconductor export controls has evolved beyond simply replacing imported chips. Instead, it is reshaping the way AI computing is built. Huawei's proposed "Tao's Law" illustrates this shift. Rather than relying solely on traditional transistor scaling under Moore's Law, Tao's Law emphasises system-level optimisation—using software, networking and hardware coordination to improve overall computing efficiency. According to Huawei's public disclosures, the latest-generation Kirin chip delivers approximately 55% higher transistor density and 41% better energy efficiency per generation, reflecting continued progress despite manufacturing constraints.

Put simply, AI performance no longer depends exclusively on building a faster individual chip. Increasingly, it can also be achieved by coordinating thousands of chips more efficiently through distributed computing, cluster scheduling and high-speed optical interconnects. This represents China's comparative advantage. Instead of competing head-on in frontier semiconductor manufacturing, China is building leadership in systems engineering and large-scale AI infrastructure.

Paradoxically, tighter US export controls may reinforce this trend over time. While restrictions remain a near-term headwind, they are also accelerating investment across domestic semiconductors, AI servers and the broader local supply chain, strengthening China's long-term technological self-sufficiency.

Policy anchoring under the 15th Five-Year Plan

China's AI ambitions are also supported by an unusually strong policy framework.

For the first time, "AI+" has been identified as a standalone national economic priority, with the government calling for broader deployment of AI applications across industries. Information technology has also been designated as both an "emerging pillar industry" and a "future industry", signalling that AI development is a long-term industrial strategy rather than a short-term stimulus measure.

Implementation has been equally comprehensive. The State Council has prioritised the expansion of intelligent computing infrastructure, hyperscale computing clusters and the Eastern Data, Western Computing initiative, while the National Development and Reform Commission is encouraging domestic AI models to optimise for locally developed semiconductors. Several provincial governments, including Beijing and Guangdong, have also established targets for domestic AI infrastructure, with locally developed technologies expected to account for at least 70% of incremental computing capacity by 2027.

Collectively, these initiatives provide a durable demand pipeline for domestic semiconductor manufacturers, AI server providers and the broader AI infrastructure ecosystem.

Energy advantage — a frequently underappreciated factor

While much attention has focused on semiconductor technology, the next major constraint on AI is increasingly electricity rather than chips.

Across North America and Europe, hyperscale data centre expansion is facing growing delays due to grid congestion, lengthy connection queues and limited power availability. China, by contrast, benefits from significantly greater power generation capacity, continued investment in new electricity infrastructure and abundant renewable energy resources in western provinces. The Eastern Data, Western Computing strategy further strengthens this advantage by relocating energy-intensive AI workloads to regions with lower electricity costs and greater power availability, improving both scalability and operating efficiency.

Taken together, these structural advantages form a reinforcing ecosystem. Domestic substitution drives technological self-sufficiency. Industrial policy creates sustained demand. Abundant energy enables continued expansion of AI computing capacity. This combination gives China a differentiated competitive position within the global AI value chain and helps explain why its semiconductor sector is emerging as a structural, rather than cyclical, investment opportunity.

Source: Huawei public disclosures, State Council and National Development and Reform Commission public documents, IEA, and other public sources.

Risk Factors

•    AI commercialisation and Tao's Law industrialisation falling short of expectations (highest sensitivity): If AI application adoption or the industrialisation of Tao's Law proceeds more slowly than expected, this would directly affect sector earnings delivery and weaken the valuation re-rating thesis.

•    Sector valuation digestion: Sector valuations are no longer cheap, and further upside depends on continued earnings validation; if the pace of delivery falls behind market expectations, valuation mean-reversion could weigh on sector performance for a period.

•    Geopolitics and export controls: Should US chip export controls on China be expanded further, or should structural shortages of core components intensify, this could cause periodic disruption and weigh on earnings for hardware constituent stocks.

•    High volatility from single-sector concentration: This product is heavily concentrated in the single information technology sector, with volatility and drawdowns significantly higher than broad-based indices; allocation size should be determined based on an investor's individual risk tolerance.

Recommended investment vehicle: GF China Securities All-Share Information Technology ETF (159939.SZ)

For investors seeking to capture China's AI infrastructure and semiconductor opportunity, we believe the GF China Securities All-Share Information Technology ETF (159939.SZ) offers one of the most effective implementation vehicles. Rather than relying on individual stock selection, the ETF provides diversified exposure to the key beneficiaries of both China's domestic semiconductor substitution and the global AI infrastructure investment cycle.

Since we first highlighted the ETF on 13 February 2026, its net asset value (NAV) has risen from 0.942 to approximately 1.476 as of 30 June 2026, representing a cumulative return of around 56.7% and closely tracking the strength of China's AI hardware sector. While past performance is not indicative of future results, our scenario analysis suggests that the ETF's underlying benchmark—the CSI All-Share Information Technology Index—still offers approximately 23% potential upside over the next three years, subject to earnings delivery and market conditions.

Why this ETF? Launched in 2015, the ETF is one of China's longest-established technology-sector funds and maintains close to 100% sector exposure, making it a high-conviction vehicle for expressing the investment thesis outlined in this report.

More importantly, its portfolio captures both structural drivers of China's AI opportunity. On one side, it provides exposure to domestic semiconductor substitution, including semiconductor equipment, wafer foundries, domestic CPUs, AI accelerators and memory manufacturers. On the other, it participates in the global AI infrastructure cycle through holdings in AI server manufacturers, printed circuit boards (PCBs), memory interface providers and other AI hardware suppliers. As a result, the ETF is well positioned to benefit from both rising global AI capital expenditure and China's ongoing localisation of its semiconductor ecosystem.

Top 10 holdings	Track / value-chain segment	% of NAV	Ecosystem thread
Luxshare Precision	Consumer electronics · AI hardware	3.82%	Global capex
Foxconn Industrial Internet	AI servers · cloud equipment	3.11%	Global capex
Naura Technology	Semiconductor equipment	2.96%	Domestic substitution
Hygon Information Technology	Domestic CPU / DCU	2.95%	Domestic substitution
SMIC	Wafer foundry	2.87%	Domestic substitution
Cambricon Technologies	AI chips	2.76%	Domestic substitution
GigaDevice Semiconductor	Memory · MCU	2.46%	Domestic substitution
Victory Giant Technology	PCB · AI compute boards	2.31%	Global capex
Montage Technology	Memory interface · DDR5	2.17%	Global capex
BOE Technology	Display panels · OLED	2.17%	Global capex
Top 10 total	—	approx. 27.6%	Both threads represented

Note: Holdings reflect the fund's most recent quarterly report; the 'ecosystem thread' classification is our own categorisation based on the dual-ecosystem framework in this note. Source: iFinD, fund quarterly report.

The portfolio's recent earnings performance reinforces this positioning. Among companies aligned with the domestic substitution theme, Cambricon reported approximately 160% year-on-year revenue growth in 1Q26, while Hygon Information Technology delivered growth of around 68%. Within the global AI infrastructure supply chain, Foxconn Industrial Internet, a key contract manufacturer for Nvidia's GB-series AI servers, recorded approximately 102% year-on-year growth in net profit attributable to shareholders.

The ETF also provides meaningful exposure to China's competitive strengths in AI hardware manufacturing. Mainland China accounts for approximately 56% of global PCB production value, making it the world's largest manufacturing base. One of the ETF's holdings, Victory Giant Technology, briefly became the world's largest supplier of AI and high-performance computing (HPC) PCBs and is deeply involved in advanced AI server platforms, including Nvidia's GB200 ecosystem.

Note: Individual stock financials are from 1Q26 reports; PCB-related data verified against Bloomberg. Source: Bloomberg, iFinD.

We believe a diversified ETF is a more effective way to capture this structural investment theme than selecting individual companies. First, leadership within the AI supply chain can rotate rapidly between semiconductors, memory, AI servers, PCBs and application software, making successful stock selection increasingly difficult. Second, the current investment opportunity is driven by sector-wide earnings growth rather than a handful of exceptional companies. As AI investment broadens across the value chain, a diversified portfolio is better positioned to capture the overall expansion of the ecosystem. Finally, individual semiconductor companies face elevated risks, including export controls, customer concentration, technology disruption and product execution. These company-specific risks are significantly reduced within a diversified ETF, allowing investors to participate in the structural growth of China's AI industry without relying on the success of any single business.

Conclusion

China's information technology sector is attractive not because it has outperformed, but because its fundamentals continue to strengthen. What began as a policy-supported theme has evolved into a structural earnings story, underpinned by the convergence of global AI infrastructure spending, accelerating domestic semiconductor substitution and sustained policy support. Together, these forces are translating into improving revenue visibility and earnings growth across China's AI hardware ecosystem.

Unlike previous technology rallies driven primarily by sentiment and valuation expansion, the current cycle is anchored by tangible earnings momentum. At the same time, China's differentiated approach to AI—centred on systems engineering, domestic supply chain development and an increasingly important energy advantage—creates a distinct investment proposition within the global AI landscape rather than a replication of the US technology sector.

For investors seeking to express this theme, we believe the GF China Securities All-Share Information Technology ETF (159939.SZ) provides a compelling implementation vehicle. Its high sector purity and broad exposure across semiconductor equipment, AI chips, servers, memory, PCBs and other AI infrastructure beneficiaries make it well positioned to capture the structural growth opportunities discussed in this report. In our view, the ETF is best suited as a medium- to long-term strategic allocation, rather than a vehicle for short-term tactical trading.

More broadly, we believe investors should view China's AI opportunity through two complementary investment pillars. The first is Hang Seng TECH, which provides exposure to internet platforms, AI applications and the ongoing recovery in valuations. The second is China's AI hardware ecosystem, represented by the GF China Securities All-Share Information Technology ETF, which offers exposure to domestic semiconductor substitution and the global AI infrastructure cycle. Together, these two themes provide more balanced exposure across China's evolving AI ecosystem than either allocation alone. As an illustration rather than a portfolio recommendation, investors seeking dedicated China technology exposure could consider allocating broadly across both pillars—for example, with an approximately 50:50 split between Hang Seng TECH and the hardware-focused ETF. The appropriate allocation should ultimately reflect each investor's objectives, investment horizon and risk tolerance.

Related articles

•    《AI Application Boom Meets Policy Resonance: The Optimal Path to Capturing Technology Sector Beta》

•    《Market Update: US and Iran Reach Ceasefire Agreement, Strait of Hormuz Reopens》

This report was prepared by the iFAST China Research Team and is intended for general informational reference only; it is published openly for general investors. This report does not constitute, and should not be construed as, an offer, solicitation, or investment advice to buy or sell any security, fund, or investment product, and does not take into account the investment objectives, financial situation, or specific needs of any particular recipient.

The information, opinions, forecasts, and valuations contained in this report are based on public information that our team believes to be reliable, together with our independent judgement; however, the Company makes no express or implied warranty as to their accuracy, completeness, or timeliness. Forward-looking statements in this report are subject to uncertainty, and actual outcomes may differ materially from them.

Past performance is not indicative of future performance. The value of investments and the income derived from them may fall as well as rise, and investors may not recover their original investment principal; where an investment is denominated in a foreign currency, exchange rate movements may also adversely affect its value.

Before making any investment decision, investors should independently and carefully assess the relevant risks, and where necessary, seek independent professional financial, legal, or tax advice tailored to their individual circumstances.

The Company and relevant licensed entities within the iFAST Group (including iFAST Platform Limited (IFPL), among others) are regulated by the relevant regulatory authorities in their respective jurisdictions. For further information about the Company and this report, please visit the official iFAST website.

Analyst Disclosure

As at the date of publication of this report, the author, Ian Li, CFA, and his immediate family hold NIL positions or interests in the securities or related investment products discussed in this report; the author's compensation is not directly linked to the specific views expressed in this report.

AI Usage Disclosure

AI-assisted tools may have been used in the drafting, data compilation, and chart-preparation stages of this report; all related outputs have been independently reviewed and verified by our research team.