The Model Isn't The Moat: Data, Domain And Distribution Are In The Age Of Agents
The model will keep getting cheaper and better for everyone. Plan as if it's free, then ask what you have that isn't.
Harshil Shah, Forbes Councils Member
Forbes
2 min read
6/10
Key Takeaways
Global enterprise spending on AI agents is projected to reach $37.2 billion by 2026, up from $8.6 billion in 2023.
82% of enterprise AI buyers now rank access to proprietary data as the top factor in vendor selection, according to a Sequoia survey.
Training costs for GPT-4-class models have dropped by over 60% year-over-year, accelerating model commoditization.
Palantir's AIP platform, powered by decades of defense and commercial data, generated $2.2 billion in revenue in 2025.
Salesforce's Einstein GPT agents leverage 150+ petabytes of customer interaction data, a moat competitors cannot easily replicate.
The AI model you spent millions training is now worth almost nothing – and that's exactly the point. An article published on Forbes Tech Council argues that in the age of AI agents, the model itself is no longer a defensible moat. Instead, competitive advantage shifts to data, domain expertise, and distribution. The piece, written by an industry expert, underscores a growing consensus among AI strategists: as foundation models commoditize, the real winners will be those who own unique data, deep vertical knowledge, and powerful distribution channels. Over the past two years, the cost of training large language models has plummeted. Open-source alternatives like Llama and Mistral have closed the gap with proprietary models. Meanwhile, agentic AI – systems that can autonomously plan and execute tasks – has moved from research labs to enterprise deployments. This shift forces companies to rethink where their true value lies. The article quotes a single, provocative line: "The model will keep getting cheaper and better for everyone. Plan as if it's free, then ask what you have that isn't." It cites examples of companies building moats around proprietary data sets – Palantir's defense and commercial data platforms, Tesla's fleet data for autonomous driving, and Salesforce's customer interaction data for CRM agents. Distribution moats are emerging through platforms like Microsoft Copilot and Amazon Bedrock, which give incumbent software giants a built-in user base. Domain expertise moats appear in sectors like healthcare, where Med-PaLM 2 trained on curated medical data, and legal, where Harvey uses case-law libraries. The thesis aligns with venture capital signals. Top VCs are now prioritizing "data moats" over "model moats" in pitch decks. According to a recent survey by Sequoia, 82% of enterprise AI buyers say access to proprietary data is the top factor in vendor selection. The model is becoming a race to the bottom on price; the data is where pricing power lives. Expect a wave of consolidation as companies without unique data scramble to acquire it. Watch for lawsuits over training data rights to intensify, and for "data flywheels" – where user interactions generate data that improves agent performance – to become the new standard metric. The age of agents is dawning, and the moat is no longer the model.
""The model will keep getting cheaper and better for everyone. Plan as if it's free, then ask what you have that isn't.""
Frequently Asked Questions
Foundation models are becoming commoditized as open-source alternatives close the gap and training costs drop rapidly. The real competitive advantage now lies in proprietary data, deep domain expertise, and distribution channels that cannot be easily replicated.
The moat for AI agents consists of three pillars: unique data sets that improve agent performance, domain expertise that enables specialized reasoning, and distribution networks that embed agents into user workflows. Companies like Palantir and Salesforce exemplify these moats.
Data is critical. Proprietary, high-quality data gives AI agents a performance edge that cannot be matched by generic models. A Sequoia survey found 82% of enterprise AI buyers rank access to proprietary data as the top factor in vendor selection.
Domain expertise refers to deep knowledge of a specific industry, such as healthcare or law, that is encoded in training data or model fine-tuning. It allows AI agents to handle specialized tasks with higher accuracy and compliance, creating a defensible moat.
Distribution moats arise when AI agents are embedded in widely used platforms like Microsoft Copilot or Amazon Bedrock. These built-in user bases reduce acquisition costs and create network effects as more usage generates more data to improve the agent.
