05 Compute Network and Cloud Platform Layer
1. Industry Chain Positioning and Value Distribution
The compute network and cloud platform layer is a massive "compute distribution hub" that connects the underlying hardware (chips) with the top-level applications (large model developers).
In the "Smiling Curve," cloud platforms still occupy a very strong position on the high end of the right side. Leading cloud providers (AWS, Azure) not only earn high rents by leasing GPUs, but also lock customers firmly into their ecosystems by bundling their own MLOps tools, API interfaces, and proprietary network architectures. However, this segment is currently facing a major transformation: heavy capital-intensive investment in purchasing GPUs, with highly uncertain payback periods.
2. Core Tracks and Business Model Conflicts
Traditional cloud giants (Hyperscalers): Microsoft (微软), Google (谷歌), and Amazon (亚马逊). They have the money to buy the most GPUs, but they face the anxiety of "how to lease out these cards efficiently without losing money." To counter the GPU hegemony of NVIDIA (英伟达), they are all frantically developing their own AI chips (e.g., AWS Trainium, Google TPU) in an attempt to reduce their dependence on NVIDIA.
Emerging compute leasing platforms (Neoclouds): Represented by CoreWeave and Lambda Labs. These dark horses design network architectures specifically for AI clusters, carry no legacy burdens, and receive support from NVIDIA in the form of chip allocation quotas. With lower prices and rawer performance, they are aggressively seizing large-model startup customers from traditional cloud providers.
3. Current State of Domestic Substitution
The domestic public cloud market (Alibaba Cloud (阿里云), Tencent Cloud (腾讯云), Huawei Cloud (华为云)) is extremely competitive, and constrained by high-end chip export restrictions, domestic cloud providers face an objective gap with North America in cloud-side compute cluster scale (at the level of 10,000 or even 100,000 GPUs per cluster). At present, domestic construction of "intelligent computing centers" (AIDC) is mostly led by local governments, which then coordinate and lease the compute capacity to research institutes and enterprises.
4. Showcase of Core Benchmark Enterprises in This Layer
Kimi
Unlisted | ChinaCore Business: Develops and operates the large language model Kimi, providing ultra-long context intelligent dialogue and text processing services
⚔️ Moat: Leads in 2-million-character ultra-long context window technology, combined with high-quality product experience, building differentiated barriers
Key Competitors: ByteDance (Doubao), Baidu (ERNIE Bot), Alibaba (Qwen), Zhipu AI (ChatGLM)
Tencent Holdings Limited
0700:HK | ChinaCore Business: Leveraging its social and content ecosystem to provide AI large models and cloud AI platform services
⚔️ Moat: Consumer-side super traffic entrance + enterprise-side service ecosystem + full-stack self-developed large model capabilities
Key Competitors: Baidu, Alibaba Cloud, ByteDance
Apple Inc.
AAPL:US | United StatesCore Business: Research & development, sales, and ecosystem operation of consumer electronic devices (iPhone, Mac, iPad) integrated with Apple Intelligence, and AI services.
⚔️ Moat: Deeply integrated hardware-software ecosystem, proprietary on-device AI chip capabilities, and strict privacy protection strategies create high user stickiness and competitive barriers.
Key Competitors: Google (Pixel/Google AI), Samsung (Galaxy AI), Microsoft (Copilot+ PC)
Alibaba Cloud
9988:HK | ChinaCore Business: The largest public cloud provider in China, offering computing power rental, Qwen large model API, and Bailian AI development platform.
⚔️ Moat: With the largest cloud computing infrastructure in China, it possesses the Qwen open-source large model ecosystem and the Bailian AI platform that integrates model fine-tuning and RAG, forming a full-stack closed loop from computing power to models to applications.
Key Competitors: