AI Agent Value Chain Analysis

In early June, Morgan Stanley released a research report on AI Agents, outlining the value chain of AI Agents and analyzing which sectors could occupy high-value niches after AI reconstructs the value chain. The main conclusions are shown in the figure below. Let's take a detailed look at each layer:

At the top of the pyramid are Hyperscalers/AI infrastructure: providing real-time inference computing power to support deep inference agents. The shovel sellers always make steady profits. In this field, there are only three companies: $Microsoft(MSFT.US) $Alphabet(GOOGL.US) $Amazon(AMZN.US)

Next are AI Model providers: enhancing agent capabilities by continuously expanding the boundaries of inference and intelligence. This strategic positioning allows them to extend into areas such as agent frameworks, data processing, and integration, eventually evolving into modern application platforms. The leaders in LLMs are OpenAI, Anthropic, Gemini, and Xai. The other three are not yet listed, and the only one available in the secondary market is $Alphabet(GOOGL.US)

3rd Security & Governance: There are significant commercial opportunities in improving the trustworthiness, reliability, accuracy, and security of agent applications. Agent architectures expand the attack surface, including risks such as data theft, supply chain attacks, prompt injection, and open-source contamination. Morgan Stanley recommends these two companies: $Okta(OKTA.US) $CloudFlare(NET.US) But I believe $Palantir Tech(PLTR.US) 's Gotham platform for the military also falls into this category, emphasizing data security.

4th Data Infrastructure: For AI agents to be truly practical, they must have governance permissions to access critical data sources in real time. Data infrastructure providers face two major opportunities: first, upgrading customer data assets and improving data quality to support agent architectures; second, enabling fast, reliable, and accurate information retrieval during runtime. Data infrastructure is a given—it's the foundation. $Palantir Tech(PLTR.US) $Snowflake(SNOW.US)

5th Workflow Automation: When AI agents are embedded into workflow automation systems, they can more fully automate unstructured segments of common business processes. Because these platforms span enterprise core systems, they are naturally suited to act as coordination layers to manage internal and external agents. $ServiceNow(NOW.US) $Palantir Tech(PLTR.US) PLTR's Ontology does exactly this.

6th End-to-End AI Lifecycle Platforms: Agents in these platforms can automate the entire process of AI application development and deployment, including building data pipelines, integrating enterprise data with large models, model training and evaluation, and data visualization. Additionally, these agents can be embedded into external applications to enhance customer interactions and create highly interactive experiences. $Palantir Tech(PLTR.US) On the foundation of Ontology, AIP acts on real-world workflows—this is the value of AIP.

Further down, starting with traditional SaaS, including Monitoring/IT tools, the value of these areas is continuously eroded, falling into the category of disruption.

After reading this, I am even more convinced of my views on PLTR and Google. Google firmly occupies the two highest-value niches in AI infra and Model, while PLTR provides industry-leading products in Data Infra, Security, Workflow, and E2E platforms. For the AI application sector, I will continue to hold long-term positions in $Palantir Tech(PLTR.US) $Alphabet(GOOGL.US)