China technology is delivering clear excess returns

China A-share technology has shown distinct structural strength in recent weeks. Part of the backdrop is improving momentum in AI applications, highlighted by ByteDance’s launch of Seedance 2.0. It reinforces the sense that AI is moving from experimentation toward more practical, scalable use cases.

Against this backdrop, application software and communications equipment have become notably more active, supported by favourable industry developments. On a weekly basis, the communications equipment index is up 2.18%. That move may not look dramatic on its own. But the year-to-date picture points to more persistent relative momentum.

As of 13 February, the STAR Composite Index is up 10.54% year-to-date, clearly beating the CSI 1000 (8.03%), CSI 300 (0.66%), and SSE 50 (0.11%). In simple terms, stocks linked to the “new quality productive forces” theme — industries tied to innovation and higher-quality growth — have risen more than the overall market.

We do not view this as short-lived thematic speculation. Instead, it reflects two supportive trends happening at the same time: rising AI investment spending by global tech giants, and domestic policies that are helping the sector.

In the global AI arms race, Chinese companies — supported by a strong manufacturing base — are embedded in global supply chains. Some leading precision component makers have secured long-term orders from overseas AI server leaders. In other words, when global tech firms announce big AI spending plans, some Chinese suppliers can see that show up as real orders and revenue.

At the same time, domestic policy continues to tilt toward “anti-involution” — discouraging destructive price wars — and toward higher-quality value creation. That shift guides industries away from competing purely on price and toward competing on innovation. It also creates a more commercially viable environment for AI applications to scale profitably, setting the stage for why developments such as Seedance 2.0 matter for the broader technology value chain.

Seedance 2.0 signals the industrialisation of AI video generation

ByteDance’s newly launched Seedance 2.0 is strategically significant because it suggests AI video generation has moved from an early phase — where results were hit-or-miss — into a more industrial-grade, usable cycle.

Seedance 2.0 also fills a domestic gap in advanced video generation. More importantly, it delivers breakthroughs across four capabilities — automatic storyboarding, automatic camera movement, comprehensive multi-modal references, and audio–visual synchronisation. These features give non-professional creators more director-like control. The shift from generating isolated clips to controlling narrative addresses long-standing pain points, including weak visual continuity and difficulty maintaining character consistency.

This move toward industrial-grade controllability has clear operational value and monetisation potential. Before this, only about 20% of AI-generated videos were typically good enough to use. Creators often had to generate the same clip again and again to get a usable result, which made the process slow and inefficient. Seedance 2.0 lifts the usable rate for generating a 15-second 2K video to over 90%. This suggests AI video is becoming less about lucky outcomes and more like a reliable tool that can be used repeatedly and at scale.

Higher usability resets the cost floor for content production

A high usable rate directly changes production economics. When more of what is generated can actually be used, the effective cost per finished minute falls.

Using a 90-minute standard video project as an example, applying this technology could reduce actual production costs from the tens-of-thousands RMB level to around the RMB 2,000 range. This kind of cost drop is most likely to be felt first in lower-budget formats such as comic dramas and micro short dramas, where cutting costs and speeding up production matters most.

This matters because it can pull forward adoption. Lower costs and faster turnaround times typically support greater production volume. More production usually means more model runs, which increases the computing needed to create, polish, and distribute content at scale. Overall, this points to growth on two fronts: more activity in AI applications, and rising demand for computing capacity.

As tools spread, IP scarcity becomes more valuable

As domestic foundation models like Seedance 2.0 become more capable, it becomes easier for AI video applications to take off because the core tools are now in place. This is changing how content companies compete, because the biggest constraints on producing content are starting to shift.

When the technology becomes widely available and video production is no longer the main bottleneck, strong storytelling and high-quality IP become the real advantages. AI may materially reduce the marginal cost of producing content. But when content supply grows quickly, IP with loyal audiences and strong emotional appeal is more likely to keep winning attention.

For software and media, this also points to a change in business models. AI tools are not just a cost-saving lever. They can also increase the value of good IP by making it easier to adapt and monetise. With AI support, high-quality IP — given its broad derivative potential — can shorten the monetisation path from text to video and from static images to animated comic dramas. That can improve conversion efficiency.

In this context, platform companies with deep pools of top-tier IP, such as China Literature (Yuewen Group) and ChineseAll, may hold up better in downturns and also benefit more when earnings improve in this AI cycle. Their IP libraries may shift from being underused assets to valuable “resource mines” that AI can turn into monetisable content faster.

Micro short dramas could be an early commercialisation path

AI micro short dramas and animated content may be among the earliest areas to convert into earnings.

Using 2025 data as a reference, the domestic micro short drama market has surpassed the RMB 100bn threshold and reached nearly 700 million users, indicating substantial depth. Deeper adoption of AI video technology could systematically reduce per-episode production costs and lift gross margins.

Commercialisation may also be helped by an end-to-end loop that covers generation, creation, and distribution — for example, through the Douyin and Jimeng ecosystems. Over time, this could make content production and distribution more efficient, and strengthen barriers around audience reach and distribution. As the impact shifts from cutting costs to improving the whole value chain, profit profiles across media and application software could change meaningfully, supporting a stronger fundamental case for higher valuations.

Computing power and infrastructure remain the foundation

Scaling applications ultimately depends on continued expansion of underlying infrastructure. Here, forward-looking global capex guidance suggests Big Tech’s AI investment is not weakening. In fact, it continues to surprise on the upside.

Taking Alphabet’s latest 2025 annual report disclosure as an example, its cloud revenue grew 48% year-on-year, with annualised revenue surpassing USD70bn. More importantly, Alphabet plans to invest USD175bn–185bn in 2026 for AI infrastructure, nearly doubling versus 2025. This global “computing arms race” provides upstream hardware supply chains with stronger earnings visibility and a supportive demand environment.

China’s inference ramp adds another leg of demand

Domestically, the focus is shifting from “catching up” on model training to ramping up inference — the computing used when models are actually run in real applications. While supply constraints on high-end chips such as NVIDIA H200 for China appear to have eased marginally, this does not change the long-term strategic commitment to technological self-reliance.

In the near term, access to advanced overseas computing can still help speed up domestic model iteration and application development. At the same time, industrial-grade video tools like Seedance 2.0 can generate large and sustained inference demand as usage scales and models are called more frequently.

With both cost-effectiveness and supply-chain security in focus, rising inference demand may create a favourable environment for domestic AI chips to gain share more quickly. Over time, this can create a positive feedback loop, where broader usage helps domestic computing platforms improve from “usable” to “highly usable”.

Hardware beneficiaries: global resonance and faster localisation

The domestic semiconductor segment may benefit from two reinforcing drivers: stronger links to global supply chains and faster localisation at home.

On one hand, globally competitive leaders have positioned successfully within overseas supply chains. For example, optical module leader Zhongji InnoLight (SZSE:300308), supported by technical barriers in 800G and 1.6T products, has maintained a solid share among overseas major clients. Its 2025 net profit is guided to grow by 89.5%–128.17%.

On the other hand, domestic computing leaders such as Hygon Information Technology (SSE:688041) and Cambricon (SSE:688256), as well as key component suppliers such as Jiangsu Maixinlin (SSE:688685), may benefit directly from the surge in domestic inference demand and the expansion of China’s push to replace foreign technology with domestic alternatives (also known as the Xinchuang initiative, 信创).

The link from application demand to orders across the hardware supply chain is clear enough to support continued strong growth in relevant segments through 2026.

Valuation: earnings recovery supports further upside

Based on our valuation model, the risk–reward profile of the CSI All-Share Information Technology Index looks attractive.

Based on index weights, semiconductors account for about 35.1%, electronic equipment/instruments/components about 30.7%, and software about 11.7%. This “hard tech + core software” mix allows the index to capture broad-based benefits from the AI era. The application breakout is only one visible part of a broader recovery across the value chain, supported by semiconductor foundations and hardware equipment as the self-reliance narrative continues.

We assign a fair 55x PE ratio to the index. This is based on the weighted average of constituent EPS forecasts, combined with the index’s average P/E over the past five years during upcycles. On this basis, we estimate 64% upside potential. With faster AI commercialisation supporting earnings recovery in software and IT services, higher growth expectations may support a gradual rise in valuations.

CSI All-Share IT Index	2025	2026E	2027E	2028E
PE Ratio (x)	68	48	39	34
EPS Growth	45%	40%	25%	15%
EPS (RMB)	0.6850	0.9590	1.1988	1.3786
Potential upside over the next 3 years (based on 55x fair PE)				64%
Target Price				13,901

Strategy: capture broad tech beta with a full value-chain tool

An application breakout can make stock selection harder and increase single-stock volatility. In that environment, one way to capture technology-sector beta is to use a broad-based index product that reflects the full value chain.

Against that backdrop, we suggest focusing on  GF CSI Shrs Tech Info ETF (SZSE:159939). Launched in 2015, the fund’s latest size is around RMB 1.183bn. It has solid liquidity and lower transaction costs from trading impact.

Its key advantage is balanced exposure across hardware, software, and semiconductors. Its top ten holdings span the ecosystem — from Apple supply-chain leader Luxshare Precision and server leader Industrial Fulian, to domestic computing core Cambricon and SMIC.

This structure aligns closely with the “new quality productive forces” theme. On one hand, application breakthroughs such as video generation can keep lifting upstream computing demand. On the other hand, the “Xinchuang” initiative and localisation narratives add structural support, helping the product benefit as the AI build-out moves from infrastructure spending to broader application rollout.

With the “Two Sessions” policy observation window approaching and global asset volatility rising, using a liquid tool as a core holding may help reduce the risk of over-rotating into a single sub-theme. It may also help investors look past near-term sentiment and stay focused on the longer-term allocation case for a technology recovery cycle.

Risk factors

• AI iteration and rollout underperform expectations. If model upgrades slow or application-side commercialisation takes longer, the sector’s valuation could come under pressure.
• Policy, regulatory and copyright compliance risks. Copyright ownership, data security, and regulation for generative AI are still evolving. Legal risks may affect sentiment.
• Geopolitics and supply-chain security risks. Structural shortages in key semiconductor components or changes in external trade policies may temporarily constrain localisation and overseas expansion.

Declaration: