Post by Alison Andrews Reyes

Just back from NVIDIA GTC 2026, and I've got some initial thoughts for enterprise leaders. There was *a lot* of focus on agents: OpenClaw, NemoClaw, and inference economics. BUT, Jensen Huang also talked in his keynote about "the ground truth of AI": your structured (& unstructured) enterprise data. Agents are transforming work, but for enterprises to see real value from agentic tasks, reasoning, and automation, data and knowledge graphs are paramount. The ecosystem is converging on a key theme: companies that own and structure their data estate will own the AI advantage. Everyone else will rent someone else's intelligence and wonder why the results don't reflect their business. If you're an enterprise leader watching this space, here's the sequence I'd encourage you to think about: 1. Data foundations. Get your structured and unstructured data in order. Clean it, index it, govern it. Not glamorous, but everything else depends on it. 2. Knowledge layer. Build the bridge from your data to your AI systems — RAG pipelines, knowledge graphs, domain-specific context. This is where your enterprise intelligence actually lives. Not in the foundation model. Not in the agent framework. 3. Agents. Once your data is structured and your knowledge layer is connected, agents become genuinely powerful. Without that foundation, they're just fast, confident systems making decisions disconnected from your operating reality. A few other GTC signals worth watching through this lens: > Agent security is real and urgent. 48% of cybersecurity professionals now rank agentic AI as their top attack vector, and only 29% of organizations feel ready. When agents are accessing sensitive enterprise data, governance has to be designed into the knowledge layer, not bolted onto the agent after the fact. > Sovereign infrastructure is maturing fast. NVIDIA Cloud Partners doubled their global AI factory footprint year over year. Microsoft Foundry explicitly targets data sovereignty requirements. On-prem and controlled-environment options are scaling for regulated industries and critical infrastructure — meaning you can keep your data estate and the intelligence derived from it within your governance boundaries. > Inference economics will reshape your budget. Vera Rubin promises 10x performance per watt. AI-Q's hybrid routing cuts query costs 50%+ by matching workload complexity to model tier. But inference-heavy agentic workloads at scale demand a FinOps discipline most organizations haven't built yet. If you don't have a plan for this, you'll feel it in 12 months. At MorganFranklin Cyber, we are already delivering solutions like this for our client. Data foundations, knowledge architecture, security and governance, then agentic capability. We work with organizations running critical infrastructure and regulated environments where there's no room for agents operating on bad data or outside governance boundaries. #GTC2026 #DataSovereignty #EnterpriseAI #AIStrategy