Who is Safe in SaaS? The Lens Via our Data Moat Scorecard

Why AI Deepens Data Moats

There is a physical infrastructure dimension to the AI and software story that the market-level debate entirely overlooks. In conversations we had with Michael Dell, and other Dell executives last summer, a thesis emerged that we believe will reshape how enterprises interact with software over the next three to five years. Michael Dell’s perspective is “why bring your data to AI, meaning move data to the cloud, when you can bring AI to your data, meaning edge and on-premise.” We think this is a compelling perspective and one that may be the forcing function of AI workloads back on premise and allow for some, not all, SaaS solutions to be custom made by enterprises, uniquely for their enterprise.

Where Data Sits

Global data creation is approaching 200 zettabytes annually, and roughly three-quarters of enterprise-generated data is now created and processed at the edge, outside traditional centralized data centers or cloud environments, up from roughly 10% less than a decade ago. This changes the fundamental calculus of enterprise AI deployment. If your data sits in the cloud, you bring your data to AI. But if your data sits at the edge, where it is increasingly being stored and created, you bring AI to your data. The implications, both strategic and economic, for enterprise software should seem obvious. When AI inference runs on-premise or at the edge at 50 to 75% lower total cost of ownership than cloud alternatives, and when agents can sit close to that data and interact and update in real time, the value proposition of paying a cloud SaaS vendor to process data that never needed to leave the building erodes structurally. We believe the architecture that emerges is a hybrid agent model, where an AI agent on an edge device like a PC can work with local data, coordinate with agents running on on-premise servers, and selectively connect to cloud-based systems of record when the task requires it. This does not kill SaaS. It accelerates what we see as the central dynamic in enterprise software today: a bifurcation.

The Big Picture

Before we get into the thesis, consider what the data actually shows. Numerous CIO and CTO surveys we have seen show contract volumes for AI tools will be additive to the millions and tens of millions they spend on enterprise software today. Enterprise GenAI spending is estimated at roughly $65 billion today and is on track to exceed $600 billion (conservatively) by the end of the decade. Yet fewer than 15% of Fortune 500 companies have deployed enterprise-wide AI initiatives, and large enterprise AI tool adoption runs at roughly 5 to 7%. The market is pricing in mass disruption from a technology that the vast majority of large enterprises have barely begun to deploy. And here is the detail that challenges the all SaaS is dead narrative: when we look at which companies are consuming the most AI compute, roughly half are B2B SaaS companies. The supposed victims of the AI wave are its single largest category of users.

At face value, if agents can build custom applications from a prompt, the value proposition of packaged enterprise software erodes. Software stocks should be de-rated accordingly. It seems the market has acted, to a degree, on this narrative. Software valuations have compressed roughly 30 to 40% over the past two to three quarters. The sector is now trading at multi-decade valuation lows relative to semiconductors, a complete inversion of the highs it reached in 2021. There is more to this onion to peel away.

It is our current view that AI will not kill SaaS, at least not all of it. What AI will do, and what the edge computing shift accelerates, is split the software market into two fundamentally different categories. There is a consensus view on this point that the first consists of system-of-record platforms: ERP, core CRM, financial planning, HR, and infrastructure orchestration. These companies sit on decades of proprietary customer data, carry prohibitive switching costs, network effects, and benefit from integration complexity that makes replacement impractical. AI does not weaken these companies. It strengthens them. Whether the data lives in the cloud or on-premise, AI agents still need to connect to the system of record to access the proprietary data that makes them useful. The system of record is the gravitational center that every agent orbits around. Multiple institutional estimates project the total software addressable market expanding from roughly $1 trillion today to between $2.5 and $3 trillion over the next decade. The market is not shrinking. It is growing, and the growth rate accelerates with AI adoption rather than contracting.

The second category consists of commodity software: single-function cloud tools that live outside an organization’s core workflow. This is where the data locality thesis hits hardest. There are now well over 1,500 AI-native applications competing in this space, and the number is growing every quarter. When AI reduces the cost of building a basic application to near zero, and when a coding agent sitting on local data can build a replacement tool without ever sending proprietary information to a third-party cloud service, the value proposition of paying $15 to $50 per user per month for commoditized functionality erodes structurally. This is where the death of SaaS narrative has real substance.

The most telling evidence comes from how agents actually behave in production. Every documented enterprise agent deployment we have reviewed sits on top of incumbent systems rather than displacing them. C.H. Robinson, an $18 billion logistics company, deployed over 30 AI agents that automated 3 million shipping tasks annually and delivered 35%+ productivity gains over two years. Those agents run inside C.H. Robinson’s proprietary logistics infrastructure. They amplify the system of record. They do not replace it. OpenAI launched Frontier, its enterprise AI platform, built explicitly to integrate with Salesforce and Workday rather than replace them. We believe Anthropic is poised to create an ISV ecosystem, which will deepen hooks to cloud-based agentic software and platforms. The most AI-capable company in the world chose to embrace incumbent software vendors, not displace them. And the one company that did try to kill SaaS, Klarna, eliminated 1,200 SaaS vendors and replaced 700 customer service agents with AI, only to publicly reverse course when customer satisfaction collapsed. Its CEO admitted the company went too far. The pattern is clear: agents that sit on top of incumbent systems succeed. Attempts to replace those systems wholesale do not.

When you put all of this together, the picture that emerges is similar to what we see in semiconductors and other sectors experiencing structural transformation. The value is not disappearing. It is migrating to different layers of the stack and accruing to companies with data advantages, platform breadth, and the ability to serve as the orchestration layer for an AI-powered enterprise. The shift of data creation to the edge adds a physical infrastructure dimension that reinforces this migration: system-of-record platforms become the gravitational center that edge agents connect to, while commodity cloud tools face displacement from agents that sit closer to where the data actually lives. The software industry will look very different in five years. It will also be bigger.

THE DATA MOAT SCORECARD

To make this thesis actionable, we built a proprietary scoring framework we call the Data Moat Scorecard. It evaluates any enterprise software company across five dimensions: proprietary data accumulation, workflow embeddedness, integration density, AI amplification potential, and pricing power. Each dimension is scored on a 1 to 5 scale, producing a composite score out of 25 that maps directly to where a company sits in the bifurcation.

The scoring bands are explained in detail in the report. Companies scoring 21 or above sit firmly on the system-of-record side. AI deepens their moat, their data advantages compound with usage, and the current valuation compression represents a mispricing rather than a rational repricing of risk. Companies at 15 or below sit in the commodity or workflow-utility tier, where AI erodes the value proposition and a disruption discount is warranted. The 16 to 20 range is the gray zone, where outcome depends on execution. The two dimensions that carry the most weight in determining which direction a company moves are proprietary data accumulation and AI amplification potential.

In the full analysis, we apply this scorecard to over 25 public enterprise software companies by name, classify the market into a four-tier taxonomy, stress-test the AI gross margin impact across archetypes, and lay out the hybrid agent architecture that we believe will define enterprise AI deployment for at least the next three to five years. The goal is to give subscribers a repeatable framework they can apply to any name in the sector, updated monthly as new adoption data and channel signals emerge. We can run any SaaS company through our scoring matrix and will keep updating in real time going forward.

WHAT SUBSCRIBERS GET IN THE FULL ANALYSIS

• Actionable conclusions with catalyst timeline: where to lean in, where to stay cautious, and the three observable phases through which the bifurcation thesis resolves, from adoption data maturation through financial divergence to multiple rerating

• The four-tier taxonomy: a precise classification of enterprise software into systems of record, systems of engagement, workflow utilities, and AI-exposed commodity tools, with the measurable characteristics that determine which tier a company occupies

• The bifurcation map: composite Data Moat Scorecard scores for over 25 real public companies across all tiers, including SAP, Oracle, Salesforce, Workday, ServiceNow, Adobe, Databricks, HubSpot, Atlassian, Snowflake, Palantir, Zendesk, Zoho, Shopify, and others

• Enterprise agent deployment analysis: how production AI agents at C.H. Robinson, OpenAI Frontier, and Klarna validate and pressure-test the system-of-record thesis with the most detailed public case studies available

• The orchestration layer thesis with bear case: why MCP creates a new infrastructure moat that incumbents control, plus the credible risks of protocol commoditization and security governance friction that could slow enterprise agent adoption

• Data locality and the hybrid agent architecture: why three-quarters of enterprise data now lives at the edge, how on-premise inference delivers 50 to 75% cost advantages over cloud, and why the three-tier hybrid agent model accelerates commodity SaaS displacement while reinforcing the system-of-record moat

• AI Gross Margin Stress Test: inference cost math per transaction across system-of-record and commodity archetypes, showing why the margin impact is absorbable for platforms with pricing power and potentially fatal for tools without it

• Pricing model evolution: the tension between outcome-based pricing and cost predictability, why Salesforce moved back from consumption to seat-based AI licensing, and what the broad industry shift toward usage-based pricing means for margin trajectories

• The build versus buy debate updated with OpenAI Frontier’s explicit embrace of incumbent vendors, plus the Klarna case study: the most aggressive SaaS replacement attempt in corporate history, what worked, what failed, and why the CEO publicly reversed course

• The Data Moat Scorecard: a five-dimension scoring framework with detailed scoring criteria investors can apply to any enterprise software company, plus worked examples for Salesforce, Adobe, and commodity tools showing how composite scores map to structural resilience

• Tracking dashboard with specific bull and bear signposts, including the catalysts that would change the thesis in either direction