摩根士丹利：AI Agents正在敲响行业大门

Investment Rating - The report assigns an "Attractive" industry view for the software sector in North America [4] Core Insights - The emergence of Agentic AI applications is driven by the increasing power of deep reasoning models, enabling automation of broader business functions and unlocking significant value [2][6] - The transition to Agentic AI is viewed as an evolutionary journey, requiring software companies to adapt their business and pricing models over time [2][6] - The Agentic AI market is projected to represent a $52 billion opportunity today, expected to reach $102 billion by 2028, with a compound annual growth rate (CAGR) of 26% [18][46] Market Opportunity - A narrow view of the Agentic AI market indicates a $6 billion opportunity today, projected to grow to $20 billion by 2028 [46] - The market for System of Record (SoR) AI Agents is currently valued at $16 billion, expected to reach $33 billion by 2028, while System of Engagement (SoE) AI Agents represent a $36 billion opportunity today, projected to grow to $69 billion by 2028 [11][76] Key Drivers - Increased demand from enterprises for process automation and ongoing improvements in foundational models are driving further demand for Agentic AI [48] - The declining price of intelligence enhances the return on investment (ROI) for AI initiatives, further fueling demand for agents [48] Company Positioning - Companies identified as Agent Beneficiaries include Microsoft, Amazon, Google, CyberArk, Okta, Salesforce, and HubSpot, which are well-positioned to monetize AI Agents or benefit from agentic adoption [26][64] - Agent Contenders are those in favorable categories but require higher execution and market maturity, while Wildcards are companies where the impact of agentic adoption is uncertain [64][87] Pricing Models - Current pricing for AI Agents is fragmented, with various approaches including activity-based, outcome-based, fixed, and variable pricing structures [27][28] - Traditional seat-based models are still prevalent, but there is a shift towards usage-based components and workflow/outcome-based models gaining traction [29][32] Value Accrual - The highest value capture potential is seen in hyperscalers and AI infrastructure providers, followed by AI model providers, security and governance, data infrastructure, and workflow automation [20][59] - The report emphasizes that the most value likely accrues to the hyperscalers and AI infrastructure layer due to significant compute resource utilization from AI Agent deployments [59][81]