·         AI application inflection point: Seedance 2.0 has raised the usable output rate for AI video generation from ~20% to over 90%, driven by four breakthroughs—autonomous shot planning, camera control, multi-modal referencing, and audio-visual synchronisation. This marks a shift from probabilistic outputs to industrial-grade production. Meanwhile, OpenClaw (龙虾) surpassing Linux in GitHub stars highlights accelerating real-world adoption of AI agents.

·         Policy support accelerating: “AI+” was впервые introduced as a standalone economic priority in the Government Work Report, with explicit support for cross-industry AI agent deployment. Information technology is uniquely classified as both an “emerging pillar” and “future industry” in the 15th Five-Year Plan. A sustained fiscal deficit of ~4% further underpins long-term investment in AI infrastructure.

·         Global compute demand tailwinds: Overseas leaders such as Alphabet, Broadcom, Marvell Technology, and Ciena continue to exceed earnings and capex expectations, reinforcing the global AI compute cycle. At the same time, NVIDIA’s wind-down of H200 production is accelerating substitution toward domestic AI chips.

·         ETF allocation advantage: Rapid intra-sector rotation makes stock picking challenging; a high-purity broad-based ETF offers more efficient exposure to AI beta. The GF CSI All-Share IT ETF (159939.SZ), with ~99.75% portfolio purity, provides balanced exposure across hardware, semiconductors, and software, and has delivered consistent alpha—outperforming its benchmark by 4.33pp over the past five years.

·         Valuation and upside: We project CSI All-Share IT Index EPS to reach RMB 1.38 by 2028. Applying a 55x P/E implies a target of 13,901, representing ~64% upside over the next three years.

Dual-Engine Resonance Launching a New Technology Upcycle

ByteDance’s release of Seedance 2.0 is one of the most important recent breakthroughs in AI applications. Its value lies not in incremental model improvement, but in changing how content is produced. AI video generation is moving from a luck-driven entertainment tool to an industrial tool that is predictable, repeatable, and scalable.

Seedance 2.0 improved four core capabilities: autonomous shot planning, autonomous camera movement, omni-directional multi-modal reference, and audio-visual synchronisation. This lifted the usable-take rate for 15-second 2K video from around 20% to over 90%. For a standard 90-minute content project, production costs could fall from tens of thousands of RMB to about RMB 2,000. This sharp cost reduction should first show up in lighter formats such as animated comics and short micro-dramas. China’s micro-drama market exceeded RMB 100 billion in 2025 and reached nearly 700 million users.

This matters beyond cost savings. It could reshape profitability across media and application software, while also driving new demand for AI infrastructure. As the application ecosystem expands, quality IP and content production capabilities should become more valuable. This should eventually support both upstream and midstream ETF holdings through higher compute demand and stronger software subscription revenues.

The AI Agent theme, represented by the OpenClaw/Clawdbot agent (龙虾智能体), is moving from concept-driven excitement to real industrial deployment. Since launch, OpenClaw has gained more than 248,000 GitHub stars, overtaking Linux and becoming the fastest project ever to top GitHub’s rankings. Domestic cloud providers including Tencent Cloud, Alibaba Cloud, and China Mobile Cloud have all onboarded it. Major model providers such as Minimax and Moonshot AI have also deeply integrated OpenClaw into their own systems for one-click cloud deployment. On the application side, companies such as Wondershare have launched atomic capability modules like text-to-video and music generation on the ClawHub marketplace.

There are also clear signs of traction at the application layer. Supported by the Kimi K2.5 model and Kimi Claw, Moonshot AI generated more revenue in 20 days than in all of 2025. Seedance 2.0 pricing was set at about RMB 1 per second of video, bringing AI video generation closer to mass-market use. Claude also surpassed 1 million daily new registrations, showing how quickly AI applications are spreading globally.

Policy support is also becoming more concrete. Shenzhen’s Longgang District Artificial Intelligence Bureau released draft measures to support OpenClaw and OPC development. These include a zero-cost-to-start framework for AI agent developers worldwide. This suggests local government support is moving from broad statements to real implementation.

The 2026 National Two Sessions delivered the strongest policy support the information technology sector has ever received. “AI+” was included in the Government Work Report as a standalone economic concept for the first time. AI agents were also explicitly highlighted for faster deployment across industries. Information technology was the only sector included under both the “emerging pillar industry” and “future industry” frameworks, with support extending through the full 15th Five-Year Plan cycle from 2026 to 2030.

Fiscal policy was also supportive. The deficit ratio remained at a historic high of 4% for the second straight year. Public budget spending exceeded RMB 30 trillion for the first time. In addition, RMB 1.3 trillion of ultra-long-term special government bonds were directed mainly toward technological innovation and major strategic projects. This provides long-term fiscal support for computing infrastructure.

Monetary policy also turned more supportive. The wording shifted from “timely reserve requirement ratio (RRR) and interest rate cuts” to “flexibly and efficiently deploying multiple tools including RRR and rate cuts.” This increases confidence that policy easing will continue and may help reduce valuation pressure on high-growth technology stocks.

Supply Side: The Global Compute Arms Race Is Driving Domestic Hardware Chain Buoyancy

The rise of AI applications cannot continue without more investment in underlying infrastructure. Alphabet is one example. Its 2025 annual report showed cloud revenue growing 48% year over year, and the company plans to invest USD 175–185 billion in AI infrastructure in 2026, nearly double its 2025 level. Broadcom (AVGO) reported fiscal Q1 FY2026 revenue of USD 19.3 billion, up 29% year over year, with its AI XPU/ASIC business achieving scale deployment across six major customers including Google and Meta. Marvell reported FY26 Q4 revenue of USD 2.219 billion, up 22% year over year, and raised FY2027 guidance to about USD 11 billion. Ciena reported revenue growth of 33.1%, while its order backlog rose from USD 5 billion to USD 7 billion.

Together, these results confirm strong global demand for AI infrastructure. This capex surge should flow through supply chains and support revenue growth for domestic hardware companies, giving the ETF’s hardware holdings strong earnings visibility.

According to several media reports, NVIDIA is gradually reducing H200 chip production for the Chinese market and reallocating related TSMC capacity to its next-generation AI compute platform. NVIDIA and TSMC have not fully detailed the production wind-down, but the broader direction appears clear. Even if H200 supply to China declines, total AI compute demand in China is unlikely to fall. Instead, tighter supply of premium chips may accelerate the adoption of domestic AI chips.

At the same time, industrial tools such as Seedance 2.0 should drive a sharp rise in inference demand through large-scale user usage. Under cost and supply-chain security considerations, this creates a strong opening for domestic compute infrastructure to move from merely usable to highly competitive. ByteDance’s 500-megawatt data centre partnership with 21Vianet, Samsung’s plans to raise NAND prices by about 100% in Q2, and the continued rise in memory prices all support this view.

The Stock-Selection Dilemma and the Strategic Response: Why ETFs Outperform Individual Stocks

Even with a constructive outlook for AI, investors still face one major challenge: fast rotation within the sector. Compute, applications, and software often lead at different times, and each shift tends to come with a sharp pullback in the previous leader. This makes concentrated bets hard to manage and increases concentration risk more than necessary.

The best response to these issues is to capture aggregate information technology sector beta through a broad-based ETF with high purity and broad industry coverage. This avoids both the risk of missing winners and the risk of lagging because of sub-sector rotation. Historical evidence suggests that in a high-growth thematic cycle, most excess returns come from sector beta rather than stock-specific alpha. In many cases, the extra return from stock picking does not justify the added concentration risk.

Recommended Vehicle: GF CSI All-Share Information Technology ETF (159939.SZ)

The GF CSI All-Share IT ETF (159939.SZ) was established on 8 January 2015 and tracks the CSI All-Share Information Technology Index (000993.SH). It is one of the longest-running domestic technology ETFs and one of the few passive technology products to operate steadily through a full A-share bull-bear cycle. This includes the 2015 bubble, the 2018 bear market, the 2019 rebound, the 2022 correction, and the rally from 2024 onward. GF Fund Management oversees more than RMB 1.5 trillion in assets and has one of the broadest passive product line-ups in China.

The fund has had two portfolio managers: founding manager Lu Zhiming from January 2015 to June 2021, and current manager Huo Huaming from April 2017 to the present, including a co-management period. Since inception, the fund has delivered an annualised return of 4.88%. Since Huo Huaming began independent management in April 2017, annualised return has risen to 5.37%, beating the benchmark by a cumulative 15.98 percentage points. This suggests the current management team has steadily improved its index-tracking effectiveness.

The peer ranking of 152/342 should be viewed carefully, as it covers a mixed group of active technology funds and passive products. That makes comparisons less consistent. For a passive ETF, the more relevant test is excess return versus the benchmark. On this basis, the fund has delivered positive alpha every year since Huo Huaming took sole charge, outperforming the benchmark in each year from 2017 to 2025. This places it among the stronger passive technology products in the A-share market. As of 27 March 2026, the fund’s tradable market capitalisation was about RMB 988 million, with 1.158 billion tradable shares. Equity holdings made up 99.75% of NAV as of 31 December 2025, giving investors near-full exposure to the information technology sector.

The fund’s quarterly asset allocation since inception shows that equity exposure rose quickly from about 95% at launch and has remained in the 98% to 100% range over time. Bond and cash holdings have stayed close to zero. This confirms the fund’s role as a high-purity beta vehicle. In practice, the fund stays almost fully invested across market conditions, so investors are not diluted by the manager holding excess cash.

The CSI All-Share IT Index holds about 100 A-share information technology blue chips. Sub-sector weights are roughly 35% in semiconductors and semiconductor equipment, 31% in electronic equipment, instruments, and components, 12% in software and IT services, and 22% in technology hardware and storage. This gives the fund broad exposure across the three main layers benefiting from the current AI cycle.

•     Upstream compute layer (semiconductors): This includes core domestic compute and wafer fabrication names such as Cambricon and SMIC. These companies should benefit directly from rising inference demand and faster domestic substitution.

•     Midstream infrastructure layer (hardware / AIDC): This includes AI server assembly players and Apple supply-chain leaders such as Foxconn Industrial Internet and Luxshare Precision. These firms should benefit from continued strong growth in domestic and overseas data centre capex.

•     Downstream application software layer: As applications like Seedance 2.0 and OpenClaw spread, AI-native software companies may see stronger earnings visibility and greater valuation upside. Representative holdings include iFlytek and Yonyou Network, both leading players in China’s AI software and enterprise services markets.

The table below shows the fund’s top 10 holdings as of 31 December 2025. Together, they account for 29.61% of NAV, while the remaining roughly 70% is spread across smaller holdings. This shows a high degree of diversification.

Note: In the Position Change column, ↑ denotes an increase and ↓ a decrease relative to the prior period; the figure represents the change in share count held. Qtr. Price Return refers to the price performance of each stock during Q4 2025. Position changes and price returns have no necessary causal relationship: the former reflects the manager’s portfolio activity and the latter reflects market performance.

Source: iFinD, iFAST Research | Data as of 27 March 2026

The chart below shows how the top 10 holdings’ combined share of NAV has changed over time. It has stayed broadly stable in a 15% to 31% range. The recent rise to about 31% in Q3 2025 mainly reflects the natural appreciation of AI leaders during the rally, which has increased portfolio convexity.

Source: iFinD, iFAST Research | Data as of 27 March 2026

The table below summarises the fund’s performance over key time periods as of 27 March 2026. The fund has generated positive excess returns over the benchmark over the past one year (+30.17%), three years (+35.03%), and five years (+35.18%). Cumulative alpha has also widened with longer holding periods, from +0.88 percentage points over one year, to +2.60 over three years, and +4.33 over five years. This suggests stable tracking quality over medium- to long-term periods.

Source: iFinD, iFAST Research | Data as of 27 March 2026

The negative since-inception alpha of -9.63% mainly reflects poor timing at launch, rather than a structural issue with long-term management. The fund was launched in January 2015, near the end of a leveraged A-share bull market. Because of its ramp-up period, it was unable to fully participate in the run-up before launch and missed the final bubble-stage rally in the first half of 2015. More importantly, once the fund became fully operational, the market peaked and then collapsed in June. This meant the fund effectively missed the upside and then absorbed the downside. That starting-point disadvantage still weighs on since-inception figures.

If we adjust for this early timing distortion and focus on Huo Huaming’s record since taking sole charge in April 2017, the picture is much clearer. The fund has delivered positive alpha relative to the benchmark in every year from 2017 to 2025. Five-year cumulative excess return stands at +9.72%, while three-year cumulative excess return is +8.59%. NAV recovery has also been steady and consistent.

The two charts below provide more detail. The upper chart shows cumulative return since inception, while the lower chart shows annual ETF return, benchmark return, and alpha from 2016 onward. Notably, the fund generated positive alpha every year from 2017 to 2025. This is a strong result in the volatile A-share technology sector.

4.5  Risk-Return Profile: Superior Returns vs. Peers, with Relatively Higher Volatility

Based on three-year data from 30 March 2023, the table below compares the fund’s key risk-return metrics with the passive index fund peer group. The fund’s annualised return of 10.52% is well above the peer average of 6.05%, while its Sharpe ratio of 0.44 also exceeds the peer average of 0.33. At the same time, annualised volatility of 30.74% and maximum drawdown of -42.46% are both above peer averages. This mainly reflects the fund’s concentrated exposure to one sector: information technology.

Investors should recognise that higher return potential comes with higher volatility. This fund is therefore better suited to investors with higher risk tolerance and a constructive medium- to long-term view on China’s technology sector.

Source: iFinD, iFAST Research | Data as of 27 March 2026

Valuation and Three-Year Return Potential

We built the following three-year earnings forecast framework using the constituent-weighted EPS of the CSI All-Share IT Index (000993.SH), guided by the earnings recovery path implied by accelerating AI commercialisation. Industrial deployment of AI applications such as video generation and AI agents should continue to drive upstream compute demand, supporting earnings growth for semiconductor and AIDC constituents in 2026. At the same time, the domestic technology self-reliance theme should provide a stable government procurement tailwind for software, making the overall EPS growth path relatively predictable.

Source: iFAST China Research | Data as of 28 March 2026 | E = Forecast | Target based on 55x fair P/E assumption

Rationale for 55x P/E: The IT sector’s forward 2026 P/E is about 48x. With EPS growth of 40%, the implied PEG is about 1.2x, which is within a historically reasonable range. During the previous technology upcycles from 2019 to 2021, the CSI All-Share IT Index traded at 50x to 70x P/E during periods of strong earnings growth. A 55x multiple therefore looks like a conservative neutral assumption.

EPS Growth Drivers: The 40% EPS growth forecast for 2026 is mainly driven by faster commercialisation of AI applications such as video generation and AI agents, alongside rising domestic inference demand and support from the Xinchuang government procurement cycle. Growth is expected to slow in later years to 25% and then 15%, mainly because of base effects. Even so, the longer-term structural growth case remains intact. Applying a fair P/E of 55x to our FY2028 EPS forecast of RMB 1.3786 gives a target level of 13,901 points for the CSI All-Share IT Index, implying about 64% upside over three years from current levels.

Risk Factors

•     AI technology progress and application commercialisation may fall short of expectations: If model upgrades slow, or if commercial adoption of applications such as video generation and AI agents is weaker than expected, sector EPS forecasts may be revised down and valuations may come under pressure. This is the most sensitive assumption in this report.

•     Geopolitical and supply-chain risks: Further U.S. export controls on AI chips to China could create downside risk to semiconductor earnings and temporarily slow the pace of domestic substitution. This could limit the fund’s NAV performance.

•     Macro conditions and domestic demand recovery may disappoint: If policy stimulus takes longer to feed into the real economy, enterprise IT spending may recover more slowly than expected. Manufacturing PMI remained below the 50 expansion threshold in February 2026. This could delay software earnings recovery and weigh on the overall EPS path.

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