Technology sector performance in February indicated that while U.S. cloud providers collectively raised capital expenditures, increased market concerns over capital returns and cash flows pressured certain cloud service and SaaS segments. Narrative and valuation focus has shifted toward computing power, advanced process nodes, equipment, storage, CPO, and liquid cooling.
Looking ahead to March, with both overseas and domestic computing demand continuing to exceed expectations, upstream sector prosperity and price increases are expected to persist. This remains the most reliable growth theme within the technology sector for "high-growth, high-prosperity" allocations. On one hand, recent updates from OpenAI and Anthropic continue to drive cloud computing and token demand beyond expectations. Intense competition among large language models is boosting both training and inference demand, leading cloud service providers to continuously revise investment upwards. However, uncertainties around return on investment and cash flow persist, making the upstream supply chain more attractive due to clearer earnings visibility. On the other hand, domestic large models are accelerating iteration, with models like GLM-5, KIMI K2.5, and Seedance2.0 gradually narrowing the gap with overseas counterparts. Some models have achieved usability and implemented price increases in application areas such as coding and video generation, reflecting extreme tightness in computing capacity. Prices are rising across the entire industry chain, from cloud services and tokens/APIs to storage, advanced manufacturing, optical communication, liquid cooling, and power.
Regarding key events, the NVIDIA GTC and OFC conferences may validate new technology trends such as CPO/NPO and LPU. Starting in March, the next generation of overseas large models will enter a密集 release period, while domestic developments like DeepSeek V4+ and Ascend 950 are anticipated. Attention should be paid to these new technological trends and domestic computing power advancements. Additionally, the market's narrative regarding "AI吞噬软件" appears overly pessimistic. The current relationship between models and software demonstrates more integration than substitution, suggesting potential investment opportunities in undervalued leading application software companies.
**Market Review: Tech Sector Diverged in February, CSP and Software Stocks Under Pressure.** Significant divergence was observed within the technology sector since February. From January 1, 2026, to February 27, under pressure from "high Capex and uncertain ROI," Amazon, Microsoft, and Google saw cumulative declines of -17%, -9%, and -2% respectively, indicating the market has already discounted expectations regarding the sustainability of capital expenditures and return paths for the CSP sector. Meanwhile, downstream AI applications and SaaS companies faced greater valuation repricing pressure. During the same period, according to Wind data, the IGV software index fell 21.8%, with representative stocks like Salesforce and AppLovin experiencing maximum declines exceeding 30%. The CITIC Media Index declined 5% in February, and the Hang Seng Tech Index fell 10.6%.
Triggering factors include the Q4 2025 earnings season, where although companies generally provided positive guidance on AI investment and revenue, it was insufficient to alleviate medium-to-long-term concerns about "AI restructuring the application layer and compressing the value capture space of traditional SaaS." Furthermore, the new version of Anthropic's Claude demonstrated an ability to "autonomously operate complex software interfaces." Combined with events like Claude Cowork and OpenClaw, this fueled narratives around "Agent acceleration," "models吞噬 applications," and "AI吞噬 software," significantly elevating market concerns about the long-term positioning and sustainable pricing power of SaaS companies.
The upstream supply chain, with its stronger earnings growth certainty, gained market favor. Continued upward revisions to AI Capex by CSPs have created robust demand pull for the entire chain, including GPU/TPU, HBM, CPO optical modules, PCBs, high-speed interconnects, power supplies, and liquid cooling systems. According to Wind data, Micron's stock price rose 45.36% in February. In the A-share market, the semiconductor equipment index and liquid cooling server index led gains. Regarding advanced processes and equipment, based on company earnings calls, TSMC raised its 2026 capital expenditure forecast to a range of $520–560 billion. Coupled with rising demand for memory, advanced packaging, and hybrid bonding, this pushed stock prices of global leading semiconductor equipment companies higher.
**Investors are advised to focus on technology sub-sectors with sustained positive momentum:** 1. **Dual-driven by Training and Inference, Computing Demand Remains Strong.** On the training side, continuous iteration and competition for state-of-the-art status among Chinese and U.S. large models, with releases like Doubao-Seed-2.0, Qwen3.5, Kimi K2.5, GLM-5, and Minimax M2.5 around the Chinese New Year, indicate robust training compute demand. On the inference side, penetration of application scenarios like Agents, Coding, and multimodality is driving exponential growth in token consumption. According to OpenRouter, since January 2026, token calls for mainstream models have maintained rapid growth for several consecutive weeks, with domestic models accounting for approximately 40% of the total. Token consumption by leading vendors also shows exponential growth; for example, ByteDance's daily token consumption by the end of 2025 grew over 400 times compared to early 2024, according to Volcano Engine conference data. As AI applications accelerate towards scaled deployment, token call growth in the first half of 2026 is expected to significantly exceed expectations. Furthermore, Zhipu's GLM-5 increased package prices by over 30% due to exploding demand in Coding and Agent scenarios, with the adjusted packages selling out quickly, further validating the current tight supply-demand dynamics for computing power.
2. **Widespread Price Increases in Upstream Components Driven by Sustained AI Demand and Capacity Constraints.** Based on expected price hike ranges for Q1 2026 across various segments, and evaluating subsequent beneficiary potential based on price hike drivers, AI exposure, and industry concentration, recommendations favor memory, PCB upstream materials, and fiber optic cables. Passive components are also suggested for monitoring.
**Forward Outlook:** Looking towards March 2026, key events for the computing power产业链 include the GTC conference, OFC conference, and the Ascend 950 PR release. Key points for GTC include the Feynman architecture based on 3D-stacked LPUs, the A16 process with HBM4, and the Rubin Ultra orthogonal backplane. The OFC focus will be on CPO progress, aligning with recent Alibaba Cloud NPO solutions and domestic switching chip roadmaps. The core aspects for the Ascend 950 PR release are its Prefill capability and revenue outlook expectations. Attention is recommended on SRAM, PCBs, CPO/NPO, and domestic switching chips.
**Overseas Review: "Capex ROI" Sparks Market Worries; "Software吞噬 Model" Narrative Leads to Misplaced Selling.** U.S. tech stocks experienced significant volatility early in 2026, raising doubts about the sustainability of the tech rally. On the computing power side, based on quarterly earnings calls, the 2026 Capex for the four major CSPs (Google, Meta, Amazon, Microsoft) exceeded expectations, totaling $6.65 trillion, a year-on-year increase of 58%. However, the market began expressing concerns about potential overheating in AI investment. On the application side, the release of AI Agent products like OpenClaw and Claude Cowork led markets to anticipate the arrival of an "AI吞噬 software" era, contributing to the IGV index's 23% decline since the start of the year.
Looking forward, from a medium-to-long-term industry perspective, the U.S. software sector appears structurally oversold and subject to misplaced selling pressure. In the short term, substitution concerns remain largely narrative-driven, weighing on the valuation of North American software stocks. Long-term, the relationship between Agents and SaaS is expected to rebalance. SaaS providers are likely to deliver enterprise-grade solutions featuring "upper-layer Agent orchestration with underlying SaaS as infrastructure." Furthermore, underlying infrastructure software essential for Agent operation, such as databases and cybersecurity, is poised to benefit significantly from the rapid development of the Agent industry, presenting new growth opportunities.
**Risk Factors:** Potential risks include slower-than-expected macroeconomic recovery;相关政策不及预期; slower-than-anticipated progress in core technology and product R&D by companies; slower-than-expected adoption of AI applications; cloud provider capital expenditures falling short of expectations; uncertainties related to global and domestic pandemic situations; and risks associated with weaker macroeconomic growth leading to lower-than-expected IT spending by domestic governments and enterprises.
**Investment Strategy:** For March, the view is that: 1) Regarding sector momentum, with both overseas and domestic computing demand exceeding expectations, upstream sector prosperity and price increases are likely to continue, representing the most certain growth主线 within the tech sector. 2) Thematic events like the NVIDIA GTC and OFC conferences may validate new technology trends such as CPO/NPO and LPU. The overseas next-generation large model release cycle intensifies from March, while domestic launches like DeepSeek V4+ and Ascend 950 are expected. Focus should be on these new tech trends and domestic computing power. 3) Additionally, market concerns over "AI吞噬软件" seem excessive. The current dynamic between models and software suggests integration rather than replacement, highlighting potential opportunities in mispriced leading application software companies.
Comments