China's AI Compute Power Enters Commercial Validation Phase: Tech Stocks Rally Strongly
Threefold Resonance Behind the Tech Sector’s Strong Rebound: China’s Domestic AI Computing Power Substitution Enters a Critical Phase of Real-World Validation
On the night of March 24, Hong Kong’s after-hours market witnessed a rare, pulse-like surge in its technology sector—the Hang Seng Index rose 2.76%, while the Hang Seng Tech Index soared 2.68%, significantly outperforming the Nasdaq (+0.83%), Nikkei 225 (+0.41%), and Euro Stoxx 50 (+0.29%) over the same period. This asymmetric outperformance was no isolated event; rather, it stemmed from a “narrative resonance” triggered simultaneously by three major developments on the same day: Xiaomi’s annual financial report release, Alibaba’s DAMO Academy hosting the XuanTie RISC-V Ecosystem Conference, and the imminent global chip event organized by Arm. At its core lies a deeper structural shift: China’s drive toward AI computing self-reliance is accelerating beyond the R&D and standards-setting stages—and entering a pivotal inflection point where commercial deployment and large-scale ecosystem validation are now underway.
Xiaomi’s Annual Report Reveals Tangible Progress in Hardware AI Integration—Consumer-End Computing Substitution Takes Concrete Shape
Xiaomi Group’s 2025 annual report—released after the close of Hong Kong trading—served as the most robust fundamental anchor for this rally. The report disclosed that Xiaomi’s full-year R&D investment reached RMB 24.5 billion, up 31% year-on-year, with over 42% allocated specifically to large AI models and on-device inference chips—the first time this share has surpassed 40%. Its flagship Xiaomi 15 Ultra series, powered by the in-house “Xuan Qiong” NPU, shipped 8.7 million units; on-device AI response latency fell below 120ms, and users engaged with AI features an average of 4.3 times per day. Crucially, Xiaomi announced it had completed joint optimization of its “Xiao Ai Large Model + Localized NPU” solution with more than 120 domestic IoT manufacturers across 17 application scenarios—including smart home systems, in-vehicle voice assistants, and industrial inspection. This signals a decisive transition: Chinese AI chips have moved beyond lab-based benchmarking contests and are now embedded within mainstream consumer electronics supply chains, generating measurable commercial returns. The market responded swiftly—Xiaomi Group’s stock surged 5.12% in after-hours trading, hitting a one-year high and triggering a broad re-rating of the entire intelligent terminal chip supply chain.
Alibaba’s XuanTie RISC-V Conference Highlights Ecosystem-Building Capabilities—Software-Defined Hardware Enters Deep Waters
Held the same day in Shanghai, the 2026 XuanTie RISC-V Ecosystem Conference elevated the technological narrative to a higher dimension. DAMO Academy announced that its XuanTie C930 processor had achieved fully independent tape-out on mature 28nm process nodes—delivering a 67% performance uplift and 39% power reduction versus its predecessor—and, for the first time, opened an “instruction-set extension interface,” enabling partners to customize AI-acceleration instructions on demand. More importantly, Alibaba released the RISC-V AI Extension Instruction White Paper alongside an open-source toolchain covering the full stack: model quantization, operator fusion, and memory scheduling optimizations. This marks RISC-V’s evolution from “functional” to “user-friendly.” With instruction-set design authority returning to developer communities, domestic chips are no longer passively adapting to foreign AI frameworks—they are proactively shaping software ecosystems tailored to their own hardware characteristics. Of the 32 chip-design firms signing agreements at the event, 19 explicitly committed to integrating XuanTie IP into edge-AI inference chips targeting sectors with strong localization demands—security surveillance, power infrastructure, and manufacturing. Accelerating ecosystem closure directly alleviates long-standing market concerns about whether domestic CPUs can truly shoulder the demands of the AI era.
Proximity to Arm’s Global Chip Event Creates Cross-Market Resonance—Signals Strengthening Global Compute Architecture Restructuring
Notably, although Arm’s global chip event—scheduled for March 24—has not yet disclosed its agenda, industry consensus expects it to focus on AI PC and edge-AI chip roadmaps. The timing’s coincidence with China’s tech-sector volatility is no accident. Today, NVIDIA’s Hopper architecture continues to dominate data centers, AMD’s MI300 series rapidly gains traction, and Arm is leveraging its energy-efficiency advantage to secure a fast-growing foothold in AI-enabled endpoints. China’s coordinated push into the RISC-V ecosystem thus strategically complements Arm’s expansion in endpoint AI—jointly squeezing x86 architecture’s remaining space in AI edge applications. This cross-market alignment confirms an emerging reality: global compute supply is shifting from “single-dominant-player governance” to a new paradigm of “multi-architecture competition and cooperation.” The strength of Hong Kong’s tech stocks reflects, at heart, the market’s forward-looking valuation of China’s structural positioning within this paradigm shift.
Profit Reallocation Risks Amid Geopolitical Friction Cannot Be Overlooked
Yet beneath the optimistic narrative flows a strong undercurrent of risk. Recent U.S.–Iran tensions have escalated sharply: Iran claimed to have sunk a U.S. oil tanker (unconfirmed by U.S. authorities), while the U.S. military denied reports of an F-15 being shot down; concurrently, the Trump administration is reportedly evaluating Iran’s parliamentary speaker as a potential negotiation counterpart. Rising geopolitical friction could trigger further tightening of U.S. semiconductor equipment export controls targeting China. Key bottlenecks remain—ASML’s immersion lithography tools and Lam Research’s etching equipment, both critical to Chinese foundry expansion—are still vulnerable to potential supply disruptions. If controls tighten, China’s substitution efforts may be forced to accelerate beyond “chip design replacement” into full-stack breakthroughs spanning equipment, materials, and fabrication—significantly elevating near-term capital expenditures and R&D costs. Historical data shows domestic foundries’ average gross margin declined 3.2 percentage points in 2023 due to rising equipment procurement costs. Consequently, today’s tech-stock valuation uplift largely reflects a “technology-breakthrough premium,” while underlying pressure on real-world profitability may gradually surface in forthcoming quarterly earnings.
Conclusion: The Inflection Point—from Conceptual Validation to Value Realization
The Hang Seng Tech Index’s 2.68% after-hours gain superficially reflects concentrated investor reaction to a single day’s events—but fundamentally, it represents the market’s phased affirmation of confidence in China’s path toward AI computing self-reliance. Xiaomi has demonstrated the commercial viability of on-device AI chips; Alibaba has showcased the engineering maturity of the RISC-V ecosystem; and Arm’s upcoming event provides a global reference frame. Together, these three developments confirm that domestic computing substitution has decisively crossed the threshold of “Does it exist?”—and entered the critical phase of “Is it good enough? Is it fast enough? Is it stable enough?”—the inflection point between conceptual validation and tangible value realization.
Still, we must remain clear-eyed: geopolitical variables are shifting from background noise to direct, material drivers. Investors should embrace technological dividends while closely monitoring progress on equipment import substitution, the maturity of domestic EDA toolchains, and marginal shifts in the U.S.–China technology rivalry. Ultimately, the true long-term winners will be those enterprises capable of both continuously shattering technological ceilings and navigating geopolitical cycles with operational agility.