Technology stocks staged a strong recovery in April, with hardware significantly outperforming software, and the AI computing supply chain leading the rally. The global AI model landscape has consolidated into a "big three" era, with ongoing competition focused on coding and agent capabilities. Anthropic's latest annual recurring revenue (ARR) has surpassed $44 billion, while major tech giants like Google reported accelerating AI-related business in their Q1 earnings. Four major cloud service providers (Microsoft, Google, Amazon, and Meta) collectively raised their full-year capital expenditure guidance from approximately $665 billion to around $700 billion, fueling the current super-inflation cycle in computing infrastructure.
Optical communication remains the most robust segment within the computing chain. Zhongji Innolight Co.,Ltd. reported sequential revenue growth for the fourth consecutive quarter, though performance diverged based on upstream material supply capabilities. Companies like Sunic faced short-term pressure due to supply chain constraints. Market focus and valuation logic are gradually shifting from demand and order visibility to upstream supply chain stability. The memory price rally continues, with DRAM and NAND Flash contract prices rising significantly, driven by both AI demand and HBM capacity shifts. Price increases have expanded beyond memory to passive components, analog chips, power devices, and electronic fabrics across the board, with later-cycle segments like CCL and MLCC gradually heating up. Domestic computing capabilities are entering a virtuous cycle, with DeepSeek V4 achieving deep compatibility with Ascend processors. Domestic computing chips are approaching an inflection point from "usable" to "effective," with the 950PR nearing mass production. Strong Q1 earnings from Cambricon and Hygon confirm sector momentum, while domestic memory and advanced process expansion are driving increased localization rates for semiconductor equipment.
Looking ahead to May-June, AI models are entering a new phase of weekly iterations, with major version updates expected from leading model companies by mid-year, accelerating commercialization into a positive feedback loop. As long as sector momentum persists, the rally should continue. Investors should maintain focus on four key technology themes: 1) Domestic large model companies approaching version update cycles with rapidly growing ARR; 2) Semiconductor equipment firms benefiting from domestic memory and advanced process expansion; 3) Optical communication leaders with secure material supply and high earnings certainty, alongside optical chip companies facing tight supply and potential price increases; 4) Upstream computing chain segments with increasing AI exposure, such as MLCC, analog, electronic fabrics, and CCL. Additionally, autonomous driving and humanoid robotics present undervalued opportunities awaiting catalysts.
Market Review: April saw a strong technology sector recovery with intensified internal divergence. Hardware significantly outperformed software, with AI computing chains leading growth. Among major indices, China's STAR 50 gained 25.05% monthly, while the ChiNext Index rose 15.45%, substantially outperforming the broader market. Hong Kong's Hang Seng Tech Index increased 4.76%, showing signs of stabilization. The Philadelphia Semiconductor Index surged 35.36%, bringing its year-to-date gain to 45.01%. Sector-wise, hardware segments electronics and communications rose 23.64% and 22.82% respectively in April, with components (+33.11%) and consumer electronics (+28.75%) leading gains. In contrast, software sectors computers and media posted minor year-to-date declines, with the IGV Software Index rising only 5.56%. Earnings validation shows electronics led in Q1 profit growth and high-growth company ratios, particularly in computing infrastructure segments (PCB, semiconductor equipment, integrated circuits). Computer sector performance remained relatively constrained.
Computing Hardware: April confirmed continued price increases and their expansion across the computing chain. Optical communication remains the strongest segment, with upstream material supply becoming the key performance differentiator. 1) Memory, PCB upstream, and fiber optics continue leading growth: DRAM Q2 contract prices rose 30%-63% quarterly, while NAND Flash increased 70%-75%. Price hikes of 10%-40% materialized for CCL, copper foil, and electronic fabrics. 2) Price increases are spreading to segments with lower AI exposure: Intel initiated a third round of CPU price increases in May, while Texas Instruments reported significantly better-than-expected Q1 revenue with broad industrial market recovery, announcing 15%-40% MLCC price increases effective May. 3) Growth segments: Optical communication remains the computing chain's strongest segment, with revised domestic computing expectations and ongoing memory/advanced process expansion likely to boost equipment demand. International chain: Zhongji Innolight Co.,Ltd. achieved four consecutive quarters of sequential revenue growth, reaching RMB 19.5 billion in Q1 2026. However, material sourcing capabilities created performance divergence, with companies like Sunic and Huilv Ecology facing earnings pressure from supply constraints. Meanwhile, four major CSPs plan combined 2026 Capex of $700 billion, with Meta raising guidance by an additional $10 billion, supporting robust computing demand. Domestic chain: DeepSeek V4 launch shows significant compatibility progress with Ascend, achieving 1.50-1.96x inference acceleration. Mass production of Ascend 950 super-nodes in H2 is expected to reduce V4 Pro inference costs. Strong Q1 earnings from Cambricon, Hygon, and Moore Thread validate domestic chip momentum, with Cambricon's Q1 revenue growing 159.56% YoY and net profit up 185.04%. Advanced memory and logic expansion are accelerating: CXMT's DRAM capacity increased from 70,000 wafers/month in 2022 to 240,000 wafers/month in 2024, with TrendForce projecting China will capture over 30% of global DRAM capacity within five years. Domestic wafer fab expansion should accelerate in 2026, benefiting semiconductor equipment and component suppliers. 4) Thematic developments: Google Cloud's infrastructure iteration centers on training-inference separation architecture. Google Cloud Conference introduced TPU 8, achieving hardware decoupling for the first time, with TPU:CPU ratio improving to 2:1. 8t super-nodes scale to 9,600 cards with 2.7x performance per dollar, while 8i's on-chip SRAM reaches 384MB suitable for MoE models. Custom Axion CPU, OCS interconnects, and Virgo network architecture are noteworthy, with clear trends toward increased CPU usage for agent scheduling.
AI Models and Applications: Models and agents are iterating rapidly, releasing strong demand. Humanoid robotics and autonomous driving may reach commercialization inflection points in 2026, potentially driving thematic rallies. 1) Model side: Large models have entered a weekly iteration phase, with domestic models demonstrating cost advantages and achieving compatibility breakthroughs with domestic computing. By end-April 2026, leading Chinese and U.S. model companies had largely completed flagship model updates, with coding and agent capabilities as key breakthrough areas. Domestic models are gaining token market share through cost advantages while accelerating compatibility with domestic computing, improving ecosystem integration. The industry expects another密集 update cycle during May-June 2026, with key products like Google Veo4, Minimax M3, and Kimi K3 likely launching. Investors should monitor progress in long-horizon agents, multimodal models, and world models. 2) Application side: Agent Harness is accelerating from basic to advanced capabilities. Since 2026, Agent Harness has gained momentum, with Hermes emerging as a phenomenon following OpenClaw. The key differentiator is Hermes' self-evolving skill闭环, enabling "improvement through use" compared to OpenClaw's externally created skills. Through engineering designs like GEPA, Hermes naturally suits deeper research applications. Additionally, Hermes is exploring paid business models, demonstrating early platform distribution effects. 3) Humanoid robotics and autonomous driving may reach commercialization milestones in 2026. Tesla's Optimus V3 has deployed 50 units at Shanghai factory, with mid-year unveiling and mass production expected. TrendForce projects China's humanoid robot production will surge 94% YoY in 2026, with frequent large financings indicating rapid scaling. Autonomous driving enters critical mass deployment phase in 2026, with domestic Robotaxi costs potentially falling below 2 RMB/km, undercutting traditional ride-hailing. Pony.ai plans to expand its Robotaxi fleet to 3,000 vehicles this year, with four L4 vehicle models from Pony.ai, WeRide, DiDi, and Hello scheduled for密集 deployment in 2026.
Risk Factors: Potential decline in high-valuation stocks; slower-than-expected macroeconomic recovery; policy implementation risks; delays in corporate technology and product development; slower AI application adoption; cloud provider capital expenditure shortfalls; pandemic uncertainties; and constrained government/enterprise IT spending due to macroeconomic weakness.
Investment Strategy: After overcoming geopolitical disruptions, markets experienced broad recovery in April. With Q1 earnings season complete, we believe performance-driven computing chains remain primary allocation targets for May, while market focus may shift from demand/order visibility to upstream supply chains. Segments with increasing AI exposure and previous underperformance offer attractive valuations. Investors should maintain focus on four technology themes: 1) Domestic large model companies approaching version updates with rapid ARR growth; 2) Semiconductor equipment firms benefiting from domestic memory/advanced process expansion and localization; 3) Optical communication leaders with secure material supply and high earnings certainty, plus optical chip companies facing supply constraints and potential price increases; 4) Upstream computing chain segments with growing AI exposure like MLCC, analog, electronic fabrics, and CCL. Autonomous driving and humanoid robotics present undervalued opportunities awaiting catalysts.
Comments