社区

□ What we now dub the "DeepSeek Trade" - sparked by the January 20th DeepSeek R1 launch and its subsequent domination of U.S. App Store rankings - marks a watershed moment in China's AI trajectory. This development, coinciding with a +23% HSI and +36% HSTECH rally (with significant AI and tech weightings), from January lows, signals a potential structural rerating of China's position in the global AI race. Key catalysts include moderated U.S. tariffs (10% vs. expected 15-20%) and DeepSeek's demonstrated capabilities, which compete with SOTA models: ChatGPT o1 (from the not[1]so-open OpenAI), Anthropic's Claude 3.5 Sonnet, and xAI's recently released Grok 3.

□ Unlike the U.S. AI narrative's evolution through multiple speculative phases, China's AI sector is moving directly to business model validation, with market focus shifting decisively from broad potential to tangible monetization paths. This transition is particularly evident in the "Six Tigers"(“六小虎”) era, where leading LLM startups face severe business model challenges - from unsustainable ToC economics (exemplified by Kimi's RMB 550Mn marketing spend) to collapsed ToB pricing (97% API price decline). Zero1's pivot to vertical specialization and efficiency-focused models may represent a viable path forward, achieving performance parity at a fraction of the cost.

□ DeepSeek's technical breakthroughs validate China's alternative path in AI development, demonstrating elite performance (88.5 MMLU, 90.2 MATH-500) at 10-14x lower cost versus global peers through architectural innovation rather than infrastructure scale. Their R1-Zero achievement proves pure reinforcement learning can enhance reasoning capabilities efficiently, while Native Sparse Attention's breakthrough (a fundamental shift in AI architecture that market narratives haven't fully processed) achieves 11.6x speedup in decoding while maintaining performance. This efficiency[1]driven approach, combined with their open-source initiative sharing production-tested infrastructure, challenges the conventional GPU-intensive development model and signals China's ability to achieve technical parity through optimization-led innovation.

□ We are witnessing an evolution from the traditional "BAT" framework to "TAP-X" (Tencent, Alibaba, PDD, plus Xiaomi). Our conviction in AI transformation is further confirmed by accelerating infrastructure investments, with BAT's CapEx/Sales ratio rebounding to historical 9% peaks. ByteDance leads 2025E investment with RMB180Bn, while Alibaba announced plans to invest RMB380Bn over the next three years. Our AI basket investment approach focuses on two key areas: cloud infrastructure providers(CSPs) benefiting from China's IaaS-heavy cloud adoption (65.7% vs U.S. 28%) and AI[1]enabled enterprise leaders in high-frequency, data-intensive sectors showing immediate ROI. Current valuations of Chinese tech leaders trade at significant discounts to U.S. peers (P/E: 20.09x vs 37.75x), creating opportunities in both domestic champions and international expansion plays as the market enters a critical sorting phase through 2025.

□ We expect enterprise adoption metrics and monetization evidence to drive the next leg of revaluation as the market transitions from AI hype to discriminating appreciation of competitive advantages. Our analysis suggests this technological convergence remains materially underpriced, particularly in companies that can effectively monetize AI capabilities while maintaining cost advantages. Maintain OVERWEIGHT on leading cloud infrastructure providers and AI-enabled vertical solutions(strongest alignment between valuation expansion and earnings acceleration). Our Deep[1]Seeking Alpha theme positions 2025 as a year of critical catalysts - from AI breakthroughs to earnings inflections- where alpha generation will require vigilance in separating signal from noise (positioning this as less a moment than a movement in redefining global tech leadership) $阿里巴巴(BABA)$ $