The AI Gold Rush: Models Are Shovels, Data Is the Gold | RAISE Summit 2026

Panelist Introductions and Core Business Focus

• Gareth Davis (Okta/Auth0) leads an identity security company securing human, agent, and consumer workforces.
• Alex Bouzari (DDN) positions his company as the data-equivalent of NVIDIA for compute, focusing on delivering business outcomes for AI natives and sovereign entities.
• Mark Sermon (Mozilla) highlights the creation of the "Mozilla Data Collective," a fair data marketplace intended to democratize AI and foster competition through open-source principles.
• Dimitri Lascarlis (Emma) leads a cloud control plane platform streamlining global infrastructure deployment and data movement across providers.
• Will Larchek (Exa) operates a search engine optimized for AI agents, providing high-quality web and non-web data.

Strategic Thesis: The Shift from Model Dominance to Data Value

• Open source and commoditized foundation models are becoming "good enough" for most knowledge work, reducing differentiation to the quality and specificity of the data used.
• Value creation is shifting from the model layer (pre-training) to the data layer (inference, agents, and fine-tuning), requiring "shovels" to reach the "gold" of proprietary data.
• Agentic AI introduces a scaling factor where one user query may trigger 30+ agent calls, exponentially increasing compute, data, and cost requirements.
• Current enterprise adoption is constrained by the "bottleneck" of access to high-quality, legally usable, and secure proprietary data.

Critical Challenges: Security, Governance, and Trust

• Security Gap: 48% of enterprises deploying AI in production are failing to secure their agents, often utilizing hard-coded API keys or overprivileged static credentials.
• Security Evolution: The industry is moving from fine-grained authorization to "semantic intent runtime layers" that understand agent behavior and context in real-time.
• Trust Deficit: Enterprises and consumers are hesitant to unlock data due to fears of exposure and misuse; Mozilla emphasizes "choice and control" to build trust.
• Sovereign Requirements: Nations require air-gapped solutions to ensure data integrity and safety, with DDN reporting deployments in countries integrating AI into commerce, agriculture, and defense.

Economic and Operational Realities

• ROI Focus: Success depends on delivering measurable financial outcomes (e.g., Roche accelerating drug discovery, Salesforce boosting training productivity) rather than just technological capability.
• Economic Viability: Data engines must solve for economics to ensure the cost of agentic inference (which is 30x higher than standard chatbots) pencils out for enterprises.
• Integration Friction: Enterprises demand seamless integration into existing infrastructure to avoid disrupting current operations while adopting AI.
• Market Dynamics: The AI economy is hyper-competitive; differentiation will rely on "data network effects" and the speed of execution ("ferocity") rather than just intelligence.

Emerging Trends and Future Trajectories (Next 5 Years)

• Specificity over Generalization: General-purpose models will yield to custom fine-tuned models specific to industry verticals or business propositions.
• Data Monetization Markets: New mechanisms (e.g., ExaConnect, Mozilla Data Collective) will emerge to allow data owners to sell access to specific datasets, such as niche languages or sign language, on fair terms.
• Privacy-Enhancing Technologies (PETs): Federated learning and differential privacy will mature, allowing organizations to collaborate and train models without sharing raw data sets.
• Decentralization: A trend toward a more decentralized ecosystem is expected as models become commodities, preventing a monopoly by a few "model labs."

Enterprise Maturity and Readiness

• State of Adoption: Most enterprises remain in "sandbox" modes; the current shift is toward "production" systems, but maturity is uneven.
• Data Readiness: Many organizations are not yet ready to utilize unstructured, unlabeled data due to a lack of "enrichment reactors" (governance, cleaning, and security infrastructure).
• Headless Architecture: The industry is moving toward a "headless world" where AI agents replace traditional user interfaces, requiring new reference architectures for agentic enterprises.
• Risk of Disintermediation: B2B software faces credible risks of disintermediation as AI transforms the experience layer, forcing rapid adaptation to maintain relevance.

Notable Anecdotes and Quotes

• Dimitri Lascarlis cited the historical example of a California land speculator who became the state's first millionaire but died broke, illustrating that value shifts from the "shovels" (land/models) to the infrastructure (banks/transport/data engines).
• Alex Bouzari emphasized that data without enrichment is like "uranium without enrichment," requiring specific reactors and governance to generate value.
• Will Larchek noted that the total data in the world (zettabytes) is roughly 1,000x larger than the web (exabytes), suggesting massive potential for agent performance if access were unlocked.
• Mark Sermon highlighted the "computer says no" mentality in France, where Firefox offers users the ability to turn off AI features to regain control and trust.