Conference Presentation, Product Demonstration
Powering Europe's Industrial Renaissance | Paolo Guglielmini, NVIDIA | RAISE Summit 2026
- Current State of AI Adoption: There has been a distinct spike in "agentic AI" adoption and a peak in urgency for European industry specifically within the last few weeks.
- Software Development Metrics: GitHub data indicates a significant surge in software output rather than headcount, with 180–200 million total developers and 27–28 million professionals contributing, driven by coding AI agents.
- The "Five Layers" Ecosystem Framework: NVIDIA defines the AI industry as five synchronized layers: Energy, Chips, Infrastructure, Models, and Applications (specifically the agentic layer), which must all scale in unison for massive ROI.
- Europe's Strategic Challenge: Europe risks becoming a consumer rather than a builder of AI if it faces constraints at the foundational layers (energy and compute); the goal is to transition from spectator to protagonist in the fourth industrial revolution.
- NVIDIA's Open Model Strategy: NVIDIA is fully committed to the Nemotron open model family, ranging from 5 billion to 550 billion parameters, aiming to enable domain-specific, culturally relevant AI for the European market.
- Industry-Specific Open Models:
- Cosmos: The leading open model for physical AI and robotics.
- Isaac Groot: A platform for humanoid robotics and development.
- BioNeMo: A solution for digital biology, genomics, and protein development.
- European Deployment: Partners in France (Lina Gora), Germany (Sophie), Israel, Poland, and pan-European initiatives (Euro LLM) are already deploying on the Nemotron stack.
- Compute Capacity Disparity: Europe currently accounts for only 5% of global installed AI compute capabilities.
- Capacity Growth Targets: Following the deployment of 800 exaflops since GTC Paris, there is a projected 10x improvement in AI compute capacity by 2027.
- Hardware Efficiency Requirements: Given Europe's energy constraints, the strategic imperative is to maximize "tokens per megawatt," with NVIDIA platforms running 90% of current AI factories.
- Software Optimization (DSX): The new DSX operating system for AI factories is projected to deliver a 2x improvement in productivity by the end of next year through optimized inference and reliability.
- Next-Gen Hardware (Vera Rubin):
- The Vera Rubin architecture, launching between late 2024 and 2025, features seven new chips and five new systems (CPUs, GPUs, networking, LPUs).
- This architecture targets a 35x improvement in tokens per megawatt compared to previous generations.
- Vera Rubin is specifically designed to handle the increased computational demands of agentic workloads.
- Strategic Recommendations for Europe:
- Build, Don't Consume: The region must prioritize building its own AI ecosystems rather than merely consuming external technology.
- Ecosystem Collaboration: Success requires deep association with NVIDIA's partner network, as no entity can build the necessary stack alone.
- Financial Viability: Investments must focus on financially sound, energy-efficient computing platforms to ensure affordability in an energy-constrained environment.
- Event Call to Action: The speaker invited attendees to GTC Berlin in October to further collaboration on building these capabilities.