Vista Equity Partners just published their latest Charterbook on Agentic AI –https://www.vistaequitypartners.com/insights/agentic-ai-is-here-what-investors-need-to-know/. As we do periodically on this blog, lets run a quick analysis of the key themes in their report.
The Infrastructure Paradox: $5 Trillion in Capital, Commodity Returns
The projected global expenditure of over $5 trillion on AI-specific data center infrastructure by 2030 represents one of history’s most aggressive technological build-outs. Yet this staggering investment obscures a critical strategic reality: infrastructure spending does not equal value capture. For investors and enterprise leaders, fixating on the compute race fundamentally misreads where durable economic moats will be constructed.
The most defensible, high-margin value in the Agentic AI revolution will not accrue to those building the infrastructure layer (Wave One), but to those controlling the application layer (Wave Three)—specifically, the enterprise software platforms that own mission-critical workflows and proprietary data repositories.
This represents a structural shift in the AI value chain, where competitive advantage migrates from raw computational capacity (increasingly commoditized) to contextualized intelligence (increasingly proprietary). Understanding this migration is essential for capital allocation decisions in the AI era.
Wave One: The Compute Race as a Necessary but Undifferentiated Cost
The Hyperscale Infrastructure Build-Out
The massive capital expenditure fueling GPU procurement, data center construction, and high-performance networking by hyperscalers (AWS, Microsoft Azure, Google Cloud) provides the raw computational power underpinning all AI capabilities. This infrastructure layer enables:
- Training large language models (LLMs) requiring thousands of GPUs operating in parallel
- Real-time inference at enterprise scale for millions of concurrent users
- Multi-modal AI systems processing text, image, video, and sensor data simultaneously
- Continuous model fine-tuning on fresh data to maintain accuracy and relevance
Yet despite its technical necessity, for the typical enterprise, this infrastructure remains a commodity to be consumed, not a moat to be built. Several dynamics explain this commoditization:
Why Infrastructure Doesn’t Create Enterprise Moats
Capital Intensity Creates Barriers, Not Moats: The billions required to build competitive AI infrastructure create barriers to entry for new hyperscalers, but they do not create defensible competitive advantages for enterprises consuming these services. A healthcare company gains no strategic differentiation by building its own GPU clusters versus accessing equivalent compute through AWS or Azure.
Rapid Commoditization Through Competition: As multiple hyperscalers compete aggressively, AI compute pricing follows a deflationary trajectory. OpenAI’s GPT-4 API pricing dropped over 90% within 18 months of launch. This competitive dynamic ensures infrastructure costs trend toward marginal cost, compressing margins for Wave One players while benefiting Wave Three application layer companies.
Abstraction Through Platform Services: Modern cloud platforms abstract the complexity of infrastructure through managed services (Amazon SageMaker, Azure AI Studio, Google Vertex AI). Enterprises access cutting-edge capabilities without managing the underlying complexity, further cementing infrastructure as undifferentiated heavy lifting.
The Strategic Imperative for Enterprises
For most organizations, the strategic approach is straightforward and financially prudent:
Avoid Direct CAPEX on AI Infrastructure: Do not attempt to replicate the AI infrastructure of cloud providers. The capital requirements are prohibitive, the expertise rare, and the opportunity cost enormous. A regional bank building a private GPU cluster diverts resources from its actual competitive advantages (customer relationships, regulatory expertise, regional market knowledge).
Focus on Intelligent Consumption: Access AI capabilities through platforms and software providers who have already absorbed the enormous capital costs and operational complexity. This approach converts CAPEX into predictable OpEx while accessing continuously improving capabilities.
Invest in Proprietary Data and Workflow Integration: Redirect capital toward Wave Three opportunities—building proprietary datasets, embedding AI into core workflows, and creating AI-native business processes that competitors cannot easily replicate.
This dynamic means the multi-trillion dollar infrastructure spend serves primarily as a sunk cost base for Wave One players (hyperscalers), while for end enterprises, it translates into a manageable OpEx line item for licensing finished capabilities through software platforms. The infrastructure becomes the table stakes, not the game itself.
Wave Three: The Moat Multiplier Effect of Enterprise Software
Structural Advantages in the Application Layer
The true competitive advantage and economic value of Agentic AI is realized only when advanced intelligence is injected into the operational heart of the business—the mission-critical enterprise platforms that orchestrate daily work. Companies providing CRM (Customer Relationship Management), ERP (Enterprise Resource Planning), HCM (Human Capital Management), and vertical-specific systems hold structural advantages over both new AI-native startups and the infrastructure layer players.
These advantages stem from exploiting three interconnected assets that create compounding moats:
- Proprietary Data: The Essential Fuel for Domain-Specific Intelligence
Decades of Contextualized, Unique Data: Established enterprise software companies possess irreplaceable datasets accumulated over decades of customer usage:
- Salesforce CRM: Contains detailed interaction histories, deal progression patterns, customer sentiment evolution, and win/loss analytics across millions of sales cycles
- SAP ERP: Holds comprehensive supply chain relationships, procurement patterns, production schedules, and financial transactions that reveal operational DNA
- Workday HCM: Stores employee performance trajectories, skill development patterns, compensation benchmarks, and organizational network effects
This data is contextualized (tagged with business meaning), longitudinal (showing trends over time), and proprietary (unique to each customer). Raw compute power is functionally useless without this quality, domain-specific fuel.
The Data Flywheel: As Agentic AI systems operate within these platforms, they generate new layers of behavioral data—how users interact with agents, which AI recommendations prove effective, what decision patterns emerge. This creates a powerful flywheel: better data → more effective agents → more usage → richer data → increasingly valuable agents.
Competitive Inaccessibility: New entrants or horizontal AI companies cannot replicate this data advantage. A startup building “AI for sales” cannot access the historical deal data, customer communication patterns, and win/loss intelligence locked within a customer’s Salesforce instance. This data moat is structurally defensible.
- Embedded Workflows: Dictating How Work Gets Done
Software as the Arbiter of Business Processes: Enterprise platforms don’t just store data—they dictate operational workflows. When Salesforce defines how sales reps progress opportunities through pipeline stages, or when SAP defines how procurement requests flow through approval chains, the software becomes the procedural memory of the organization.
Embedding Agentic AI directly into these workflows creates indispensability:
- Salesforce Agentforce: AI agents that autonomously qualify leads, draft personalized emails, schedule meetings, and update CRM records become integral to the sales motion, not peripheral tools
- SAP AI Agents for Supply Chain: Agents that predict stockouts, automatically reorder inventory, and renegotiate supplier terms become load-bearing elements of operations
- Workday AI for Talent Management: Agents that match candidates to roles, draft job descriptions, and predict attrition risks become embedded in workforce planning
The Workflow Lock-In: When business-critical processes are mediated by AI agents within the platform, replacing the platform requires reconstructing not just the data, but the AI-augmented workflows. This dramatically amplifies switching costs.
- Amplified Switching Costs: Creating Durable, High-Margin Revenue
Baseline Enterprise Software Switching Costs: Even before Agentic AI, enterprise software enjoyed high switching costs due to:
- Data migration complexity and risk
- Employee retraining requirements
- Process reengineering overhead
- Integration dependencies with other systems
- Contractual and financial commitments
The Agentic AI Multiplier: Agentic AI exponentially amplifies these already-high switching costs through several mechanisms:
Institutional Memory Embodied in Agents: AI agents trained on a company’s historical data and workflows become repositories of institutional knowledge. A Salesforce agent that “knows” which email subject lines resonate with specific customer personas, or a Workday agent that “understands” which skill combinations predict high performance, embodies years of organizational learning. Switching platforms means losing this accumulated intelligence.
Network Effects Within Agent Ecosystems: As multiple specialized agents (sales agent, marketing agent, customer service agent) interact within a platform, they develop coordinated behaviors and cross-functional intelligence. A CRM agent that learns to pass qualified leads to a marketing automation agent, which then coordinates with a customer service agent, creates inter-agent dependencies that are platform-specific.
Customization and Fine-Tuning Investment: Enterprises invest significant resources in fine-tuning AI agents to their specific business context—training on proprietary data, configuring decision rules, and optimizing agent behaviors. This investment is platform-specific and represents sunk cost that amplifies switching barriers.
Perceived Risk of AI Transition: Beyond technical complexity, switching platforms when Agentic AI manages critical functions introduces perceived execution risk. CFOs and CIOs become reluctant to disrupt AI-powered workflows that are delivering measurable business value, even if alternative platforms offer marginal improvements.
The Value Migration: From Undifferentiated to Contextualized
The fundamental value shift in the Agentic AI era moves from:
Undifferentiated Infrastructure → Contextualized Intelligence
- Wave One (Infrastructure): Provides necessary but commoditized compute. High CAPEX, competitive pressure, margin compression.
- Wave Three (Application Layer): Delivers proprietary, workflow-embedded intelligence. Asset-light, high switching costs, expanding margins.
For investors, this framework clarifies where to allocate capital: software companies whose Agentic AI integrations are strategically designed to exploit proprietary data, embedded workflows, and amplified switching costs will capture disproportionate value.
Investment Implications: Identifying Wave Three Winners
Criteria for Defensible Agentic AI Positioning
Not all enterprise software companies will equally benefit from Agentic AI. Investors should evaluate companies based on:
- Data Asset Quality and Exclusivity: Does the platform accumulate unique, longitudinal, contextualized data that creates compounding advantages for AI agents? (e.g., Salesforce’s sales interaction data, Workday’s workforce analytics)
- Workflow Centrality and Criticality: How mission-critical are the workflows the platform orchestrates? Platforms managing core business processes (revenue generation, financial close, compliance) have stronger positioning than peripheral tools.
- Agent Ecosystem Potential: Can the platform support multiple, interacting agents that create network effects? Platforms enabling “agent swarms” for complex processes have stronger moats than single-agent applications.
- Customer Stickiness Metrics: What are current net revenue retention rates and gross retention rates? High baseline retention amplifies the Agentic AI switching cost multiplier.
- Platform Control and Extensibility: Does the company control the platform foundation, or is it building on someone else’s infrastructure? Platform owners capture more value than platform tenants.
The Margin Expansion Thesis
As Agentic AI becomes table stakes, Wave Three companies should experience margin expansion through several mechanisms:
- Pricing Power: Platforms can price based on business value delivered (outcomes) rather than per-seat licensing, capturing more value from AI-driven productivity gains
- Operating Leverage: AI agents reduce customer support costs while increasing platform engagement and feature adoption
- Reduced Churn: Amplified switching costs translate to higher retention and longer customer lifetime values
- Land-and-Expand: AI capabilities accelerate seat expansion within existing accounts as more employees gain access to AI-augmented workflows
Companies successfully executing this playbook should demonstrate expanding gross margins (AI features delivered at software economics, not services economics) and accelerating revenue growth (from AI-driven product differentiation).
Conclusion: Capital Allocation in the Agentic AI Era
The Agentic AI revolution presents a clear strategic fork for enterprises and investors:
For Enterprises: Avoid the seductive trap of competing on infrastructure. The winning strategy is aggressive consumption of commoditized compute combined with aggressive investment in proprietary data, workflow integration, and AI-native business process redesign. Partner with Wave Three software platforms that embed intelligence where your organization creates unique value.
For Investors: The $5 trillion infrastructure build-out is necessary but undifferentiated—a cost base for hyperscalers, not a moat for enterprises or a source of outsized returns for most investors. The durable, compounding returns will accrue to enterprise software platforms that control critical workflows, possess irreplaceable data assets, and systematically amplify their switching costs through embedded Agentic AI.
The value chain is clear: Infrastructure enables, but Application Layer captures. In the Agentic AI economy, focus capital where defensibility compounds—in the software that dictates how work gets done.
Featured image design by Freepik