The second half of Nvidia's recent panel on open AI models shifted focus to the operational challenges facing modern enterprises. Panelists agreed that the next phase of artificial intelligence competition will center on control, governance, and scalable open infrastructure rather than sheer model size.
The Demand For Corporate Control
Mistral co-founder and CEO Arthur Mensch argued that open models must form the foundation of enterprise AI specifically to ensure corporate control. He emphasized that companies cannot afford to outsource their execution layers to unpredictable external application programming interfaces.
As AI expands into manufacturing and physical engineering, enterprises must embed their own intellectual property and internal expertise directly into these models. Mensch noted that only open models allow organizations to safely connect proprietary data sources and develop genuine operational understanding.
The Strict "Two Of Three" Rule
NVIDIA CEO Jensen Huang introduced a strict security framework for deploying autonomous agents across corporate networks. He advised that any single agent should only be permitted to perform two of three core functions: accessing sensitive data, executing code, or communicating externally.
This rule effectively dictates that robust governance and compliance mechanisms must be established long before deployment begins. Hanna Hajishirzi, a senior researcher at AI2, reinforced this by noting that user trust ultimately depends on strict data sovereignty and verifiable privacy controls.
Deploying Portfolios Of Specialists
This operational logic heavily favors deploying a portfolio of specialized agents rather than relying on a single, general-purpose model. Enterprises increasingly require purpose-built agents dedicated to distinct functions like finance, operations, healthcare, and customer service.
OpenEvidence CEO Daniel Nadler illustrated this trend by highlighting the immense administrative burden placed on modern healthcare workers. He described how specialized agents can automatically process complex insurance authorizations and draft denial appeals while a physician sleeps.
Building The Open AI Grid
As companies deploy these specialist agents at scale, critical infrastructure constraints and computational waste have rapidly come into focus. AMP founder Anjney Midha warned that enterprises are currently hoarding graphics processing units and over-provisioning for peak demand, creating massive systemic inefficiencies.
Midha proposed an "AI Grid" to provide secure, shareable open infrastructure that delivers reliable baseline computing power. Huang supported this vision, predicting the emergence of new infrastructure models that prevent enterprises from falling into traps of computational hoarding and vendor lock-in.
The Approaching ROI Inflection Point
The panel also clarified the approaching inflection point for corporate return on investment in artificial intelligence technologies. Huang stated that the market's central question is finally shifting from questioning the existence of financial returns to measuring their actual materialization.
He emphasized that coding serves as the essential mechanism for converting scattered operational processes into systematically managed digital assets. By utilizing open infrastructure and specialized agents, AI is transitioning from simple chatbots into the core operating systems of global enterprises. （Related： The Next AI Paradigm: Nvidia CEO Says The Future Belongs To Multi-Agent Systems, Not Black Boxes ｜ Latest ）
