CITIC Securities: Technology sector strongly recovers in April, sticking to four main themes of technological prosperity

CITIC Securities Research Report points out that in April, the technology sector experienced a strong recovery, with hardware significantly outperforming software, and the AI computing power chain leading in prosperity. Looking ahead to May and June, models will enter a new weekly iteration phase, with major model companies expected to release large version updates around mid-year, accelerating commercialization into a positive cycle. Prosperity continues, market trends persist, and it is recommended to focus on four main lines of technological prosperity: 1) domestic large models entering version iteration windows with rapidly increasing ARR; 2) semiconductor equipment benefiting from continuous expansion of domestic two-storage and advanced process capacity, with rising domesticization rates; 3) leading optical communication companies with material supply guarantee and high performance certainty, along with upstream optical chip companies facing tight supply and price increase expectations; 4) upstream of the AI exposure gradually increasing in the computing power chain, such as MLCC, analog, electronic fabric, CCL, etc. Meanwhile, low-position resilience directions like autonomous driving and humanoid robots are worth watching for more catalysts.

The full text is as follows:

Technology | Nasdaq hits new highs, is the May tech rebound a rebound or a reversal?

In April, the technology sector experienced a strong recovery, with hardware significantly outperforming software, and the AI computing power chain leading in prosperity. Overseas models entered the “Big Three” era, with each competing around coding and agent capabilities. Anthropic’s latest annual recurring revenue (ARR) has exceeded $44 billion. Major tech giants like Google reported Q1 earnings showing continued acceleration in AI-related businesses. The four major companies (Microsoft, Google, Amazon, Meta) further revised their full-year Capex guidance from approximately $665 billion to about $700 billion, fueling this round of super-inflation in the computing power chain.

Optical communication remains the most prosperous direction in the computing power chain. Zhongji Xuchuang reported four consecutive quarters of revenue growth month-over-month, but upstream material supply capabilities are diverging, with companies like XinYisheng under short-term pressure due to supply issues. Market focus and pricing logic are gradually shifting from demand and order visibility to upstream supply chain. Storage price increases continue, with DRAM and NAND Flash contract prices rising sharply—by 30-63% and 70-75%, respectively—driven by AI demand and HBM capacity switching. Price hikes have spread from storage to passive components, analog chips, power devices, and the entire electronic fabric chain, with post-cycle sectors like CCL and MLCC warming up gradually. Domestic computing power is entering a positive cycle, with DeepSeek V4 deeply adapting to Ascend, and domestic computing chips gradually reaching a turning point from “usable” to “good to use.” The shipment of 950PR is imminent, and high growth in Q1 for Cambrian, Hygon Information, and others confirms prosperity. The expansion of domestic two-storage and advanced process capacity continues to promote the localization of semiconductor equipment.

Looking ahead to May and June, models will enter a new weekly iteration phase, with major model companies expected to release large updates around mid-year, accelerating commercialization into a positive cycle. Prosperity persists, and market trends continue; it is recommended to focus on four main lines of technological prosperity: 1) domestic large models entering version iteration windows with rapidly increasing ARR; 2) semiconductor equipment benefiting from continuous expansion of domestic two-storage and advanced process capacity, with rising domesticization rates; 3) leading optical communication companies with material supply guarantees and high performance certainty, along with upstream optical chip companies facing tight supply and price increase expectations; 4) upstream of the AI exposure gradually increasing in the computing power chain, such as MLCC, analog, electronic fabric, CCL, etc. Meanwhile, low-position resilience directions like autonomous driving and humanoid robots are worth watching for more catalysts.

Market Review: In April, the technology sector rebounded strongly but with increased internal differentiation. Hardware significantly outperformed software, and the AI computing power chain led in prosperity.

Vertically, A-shares’ Sci-Tech Innovation 50 rose 25.05% in a single month, leading the market, with the ChiNext Index up 15.45%, significantly outperforming the broader market; Hong Kong’s Hang Seng Tech rose 4.76%, showing signs of stabilization; US stocks’ Philadelphia Semiconductor Index surged 35.36%, up 45.01% year-to-date. Horizontally, hardware sectors like electronics and communications rose 23.64% and 22.82% respectively in April, with components (+33.11%) and consumer electronics (+28.75%) leading. In contrast, software sectors like computing and media saw slight declines since the start of the year, with the IGV software index up only 5.56%. Performance-wise, high-growth companies in electronics reported high profitability and net profit growth in Q1, with sectors related to computing infrastructure (PCB, semiconductor equipment, ICs) leading in prosperity; the computer industry faced some pressure.

Computing Hardware: April’s continued and expanding price increases in the computing power chain are validated; optical communication remains the most prosperous direction, with upstream material supply being a key performance variable.

1. Storage, upstream PCB, fiber optic cable sectors continue to lead: Specifically, DRAM contract prices in Q2 are up 30-63% month-over-month, NAND Flash up 70-75%; CCL and copper foil, electronic fabric prices increased 10-40%.

2. Price increases are spreading to mid- and low-level AI exposure segments: Intel launched its third CPU price hike in May; Texas Instruments’ Q1 revenue greatly exceeded expectations amid a full recovery of the industrial market, and from May, it will raise MLCC prices by 15-40%.

3. Prosperity growth aspects: Optical communication remains the most prosperous in the computing power chain. Domestic computing power expectations are being revised upward, with continued expansion of two-storage and advanced process capacity expected to boost equipment prosperity. Overseas, Zhongji Xuchuang’s revenue has grown for four consecutive quarters, reaching 19.5 billion yuan in Q1 2026; however, upstream material supply capabilities are diverging, with XinYisheng and HuiLu Eco under short-term pressure due to supply issues. Meanwhile, the four major CSPs plan a combined Capex of $700 billion in 2026, with Meta further raising guidance by $10 billion, supporting high demand for computing power. Domestically, DeepSeek V4 has been released, with significant adaptation to Ascend, achieving inference acceleration of 1.50-1.96 times. The batch release of Ascend 950 super nodes in the second half is expected to reduce V4 Pro inference costs; high growth in Q1 for Cambrian, Hygon Information, and Moores Law Thread confirms the prosperity of domestic chips, with Cambrian’s revenue up 159.56% YoY and net profit up 185.04%. Horizontally, the year of large-scale expansion in advanced storage and logic begins: Changxin Storage’s DRAM capacity has increased from 70k wafers/month in 2022 to 240k wafers/month in 2024. TrendForce predicts domestic DRAM capacity will account for over 30% of the global share in the next five years; domestic wafer fabs are expected to accelerate expansion in 2026, benefiting semiconductor equipment and parts companies.

4. Thematic aspects: Google Cloud infrastructure iteration, with training and inference separation architecture becoming a core theme. Google Cloud’s conference launched TPU 8, achieving hardware decoupling for training and inference for the first time. TPU:CPU ratio increased to 2:1, with 8T super nodes scalable to 9,600 cards and performance per dollar improved 2.7 times; 8i on-chip SRAM reached 384MB, suitable for MoE models. Notable features include custom Axion CPUs, OCS interconnect, and Virgo network architecture, with rising demand for agent scheduling increasing CPU usage.

AI models and applications: intensive model agent iteration releases strong demand. Humanoid robots and autonomous driving are expected to enter commercialization in 2026, potentially triggering thematic market trends.

1. Models: The large model field has entered a new weekly iteration phase, with domestic models showing cost-effectiveness advantages and breakthroughs in domestic computing power adaptation. By the end of April 2026, major Chinese and US model companies have basically completed a new round of flagship model updates, with coding and agent capabilities as core breakthroughs. Domestic models, leveraging cost advantages, continue to gain market share in tokens and accelerate adaptation with domestic computing power, with ecological compatibility improving. It is expected that in May-June 2026, the industry will see another intensive update cycle, with key products like Google Veo4, Minimax M3, Kimi K3 expected to be released successively. Focus should be on breakthroughs and applications of long-horizon agents, multimodal models, and world models.

2. Applications: From raising “shrimp” to raising “horses,” agent harnesses are experiencing an accelerated iteration wave. Since 2026, agent harnesses have continued to heat up, with Hermes becoming a phenomenon-level product following OpenClaw. Its core difference lies in building a skill self-evolution closed loop. Compared to external skill creation like OpenClaw, Hermes can achieve skill self-evolution, resulting in a “more use, more useful” experience. Through engineering designs like GEPA, Hermes is more suitable for in-depth research. Additionally, Hermes is exploring paid commercial models, with the distribution entry point effects beginning to emerge.

3. Humanoid robots and autonomous driving: expected to usher in commercialization in 2026. For humanoid robots, Tesla’s Optimus V3 has landed 50 units at its Shanghai factory, expected to debut mid-year and start mass production. According to TrendForce, China’s humanoid robot output will increase by 94% YoY in 2026, with large-scale financing frequent, indicating rapid industry growth. For autonomous driving, 2026 will be a critical year for large-scale deployment, with domestic Robotaxi costs expected to fall below 2 yuan per km, officially lower than traditional ride-hailing. Xiaoma Zhixing’s Robotaxi fleet may expand to 3,000 vehicles within the year, with four L4 pre-installed models from Xiaoma Zhixing, Wenyuan Zhixing, Didi, and Hello also planning to deploy intensively in 2026.

Risk factors:

Risks of overvaluation correction; slower-than-expected macroeconomic recovery; industry policy risks not meeting expectations; core technological and product R&D progress falling short; AI application deployment slower than expected; cloud vendors’ capital expenditure underperforming; uncertainties from global and domestic pandemic developments; macroeconomic sluggishness affecting domestic government and enterprise IT spending.

Investment strategy:

After overcoming geopolitical disturbances, the April market experienced a recovery and broad rally. With Q1 earnings reports completed, looking into May, we believe that the main line of computing power chain performance remains the preferred allocation. Meanwhile, market focus and pricing logic may gradually shift from demand and order visibility to upstream supply chain, with increased AI exposure and previously lagging segments showing cost-effectiveness. Allocation suggestions include focusing on four main lines: 1) domestic large models entering version iteration windows with rapidly increasing ARR; 2) semiconductor equipment benefiting from continuous expansion of domestic two-storage and advanced process capacity, and companies benefiting from domestic computing power iteration; 3) leading optical communication companies with material supply guarantees and high certainty, along with upstream optical chip companies facing tight supply and price increase expectations; 4) upstream of the AI computing power chain, such as MLCC, analog, electronic fabric, CCL. Additionally, low-position resilience directions like autonomous driving and humanoid robots are worth watching for more catalysts.

(Source: Yicai)