Beyond the hype: how NVIDIA GTC Paris 2025 trends are shaping industry

NVIDIA GTC Paris 2025, hosted within one of Europe’s largest startup and tech events, Viva Technology, showcased AI's practical application and industrialisation. Stretching across an expansive exhibition floor, NVIDIA's vision revealed an unprecedented scale and breadth of innovation, solidifying its leadership far beyond traditional GPU markets. From quantum computing to autonomous systems, industrial digital twins, and advanced robotics, the emphasis was clear: not on predicting the future of AI, but on demonstrating how existing technologies are being put to work today.

I attended the event to extract key trends that matter for our partners leading data and AI initiatives. This article distils the key takeaways—from industrial digital twins and generative agents to AI sovereignty and compute architecture—and explains what they mean for accelerating innovation today. A more in-depth exploration is also available on this second blog article for those seeking comprehensive technical coverage.

The Industrial Revolution Powered by AI and Digital Twins

Jensen Huang’s keynote underscored how NVIDIA’s comprehensive, end-to-end stack – spanning hardware, software, and cloud services – is unlocking powerful synergies across diverse domains. One of the most compelling narratives from GTC Paris was the impact of AI and digital twins on industrial operations and manufacturing.

Omniverse-Powered Physical Simulation & Digital Twins: NVIDIA Omniverse emerged as a unified platform for realistic physics and behaviour simulation. It's enabling the creation of intricate digital twins – virtual replicas of factories, vehicles, and consumer products – using the Universal Scene Description (USD) format. Some concrete examples:

• BMW's Factory Transformation: BMW showcased how it transformed its physical factory into a digital twin using 3D scans, which were imported into NVIDIA Omniverse. This virtual environment unlocked approximately 4,000 use cases, delivering a 35% return on investment and allowing 70% of factory planning to be executed virtually. This demonstrates how companies can optimise production, test layouts, and identify efficiencies before committing to physical changes.
• AI Agents for Intelligent Inspection: Omniverse also facilitates the integration of AI agents, powered by Large Language Models (LLMs) and Retrieval-Augmented Generation (RAG). These agents can reason over the virtual model, allowing users to query the virtual factory naturally to identify non-connected components, detect missing safety installations (like ramp railings), or perform guided virtual walkthroughs. This capability redefines design review, safety compliance, and operational efficiency.
• Consumer Goods Visualisation: Beyond heavy industry, Omniverse is being applied to consumer goods. Nestlé, for instance, launched an AI-powered content workflow with Accenture, converting 4,000 products into 3D twins with a target of 10,000 within two years. This highlights the potential for rapid content creation and real-time marketing visualisation, even extending to the fashion and art industries (an entire floor of the conference was dedicated to an Omniverse art gallery).

Advanced Robotics: The event also detailed advancements in robotics, with Newton, an open-source physics engine co-developed with Google DeepMind and Disney Research. Coupled with Jetson Thor AI inference modules and Omniverse simulation, Newton enables high-fidelity robot learning in virtual environments. This means industrial robots can be trained to perform complex manipulations or navigate varied terrains virtually, significantly accelerating their development and deployment in the real world.

Industrial AI Clouds: NVIDIA announced Europe's first Industrial AI Cloud. Based in Germany and powered by Omniverse, this dedicated cloud ecosystem provides manufacturers such as Mercedes-Benz, Bouygues Télécom, and BMW with a turnkey solution for digital twins. This enables them to achieve breakthroughs in real-time design, advanced simulation, and wind tunnel analysis, directly driving efficiency on the shop floor.

This aligns with a growing trend where major telecom operators, including Orange, T-Mobile, Telenor, and Swisscom, are positioning themselves as sovereign cloud providers, offering the full NVIDIA stack directly from their data centres to meet strict data sovereignty and regulatory compliance needs within the EU.

NVIDIA Grace Blackwell Architecture

At the heart of much of the showcased innovation was NVIDIA’s new Blackwell-architecture GPUs. These immense chips, featuring 208 billion transistors and manufactured on TSMC’s 4NP process, comprise dual dies connected by a staggering 10 TB/s interconnect. When combined with Grace CPUs in a Blackwell-Grace module, they form supercomputer-scale racks, offering 100 TB/s NVLink connectivity and liquid cooling. Jensen Huang highlighted that weekly production of 1,000 such modules will fuel the next generation of AI factories and on-premises supercomputers, scaling down from large data centres to individual workstations and laptops.

This raw computational power is critical. NVIDIA is partnering with European infrastructure leaders like Siemens and Schneider Electric to build sovereign “AI factories”. This pan-European rollout, including new tech centres and sovereign AI hubs in France, Germany, Italy, Finland, the U.K., Spain, and Sweden, underscores a strong regional commitment to digital sovereignty and building local AI expertise.

Powering Distributed AI with DGX Cloud Lepton

Crucially, these AI factories and distributed AI deployments are being brought to life through platforms like NVIDIA DGX Cloud Lepton. More than just a cloud service, it's a "virtual global AI factory" and a unified compute marketplace that connects developers to tens of thousands of GPUs across a federated network of cloud providers worldwide. For businesses operating globally, Lepton addresses the critical need for reliable, high-performance GPU resources by unifying GPU compute across various cloud providers into a single, developer-friendly platform. It simplifies multi-cloud AI deployment, enabling enterprises to prototype in one region and scale globally while adhering to data-governance requirements.

Features include:

• Instant access to NVIDIA APIs, serverless endpoints, and Blueprints
• Region-specific GPU resource selection for sovereign AI
• Marketplace integration with NVIDIA Cloud Partners (NCPs)

Agentic AI and Generative Models

Jensen Huang framed AI's progress in distinct waves, with "Agentic AI" (involving perception, understanding, reasoning, planning, and action) now at the forefront, followed by "Physical/Robotics AI" for motion generation and dynamic adaptation. NVIDIA's Nemotron family of multimodal models provides state-of-the-art agentic reasoning capabilities, enabling sovereign LLMs to be deployed on-premises or via the NVIDIA Cloud.

A key takeaway was the strong European partnership in this space. Companies like Siemens, Novo Nordisk, and Shell are actively building their own agentic AI agents using NVIDIA’s Blueprints and NeMo Agent toolkit, specifically tailored to regional languages and data sovereignty requirements. Furthermore, significant partnerships were announced: Mistral AI is deploying 18,000 Blackwell GPUs in a new Paris-area facility, and Perplexity is localising reasoning models across Europe, generating synthetic data for low-resource languages. These developments mean digital companies can leverage highly sophisticated, context-aware AI agents while respecting local data regulations.

Emerging Frontiers: Quantum Computing's Practical Dawn

While still an early-stage technology for most business applications, Jensen Huang noted that quantum computing has reached an "inflection point—shifting from theoretical promise to near-term practical applications." This means laying the groundwork for solving problems that even today’s largest AI models struggle with.

NVIDIA is facilitating this transition with CUDA‑Q, its open‑source, qubit‑agnostic platform. CUDA‑Q enables hybrid quantum‑classical programming, orchestrating GPUs, CPUs, and QPUs from a unified interface. It also provides GPU‑accelerated simulation of quantum circuits when physical QPU hardware isn’t available—a major tool for accelerating quantum error correction, logical qubit management, and backend validation.

Real-world examples highlighted at GTC include:

• Denmark’s supercomputer Gefion using CUDA-Q to simulate quantum algorithms in fluid dynamics, accelerating workflows on real-world problems.
• French startup Alice & Bob integrating CUDA-Q with its Dynamiqs library to speed quantum-state simulations by up to 75x, aiding in fault-tolerant qubit development.
• France’s quantum innovator Pasqal joining forces with NVIDIA to connect its QPUs to CUDA-Q for enhanced programming and simulation capabilities.

Conclusion

This conference was truly impressive. The scale of the exhibition floor, an entire level of the venue dedicated to art created in Omniverse, the extensive booths showcasing internal R&D and real-world applications, and the quality of discussions during the expert sessions all reflect how far GTC has come. It’s becoming a must-attend event for anyone who wants a clear view of what’s technically possible today.

What stood out most was NVIDIA’s ability to demonstrate the depth and breadth of its innovation, spanning GPUs, HPC, CUDA, 3D simulation engines, reinforcement learning, robotics, agentic AI, computer vision, and even weather modelling. You name it. Even without our Euranova lab full of A100s, it’s clear that NVIDIA remains one of the most visionary companies in the tech ecosystem.

And NVIDIA GTC Paris 2025 underscored a clear message: AI is a practical, industrialised force. If you are looking for guidance to confidently embrace these advancements, whether it’s launching your own digital twin or integrating agentic reasoning into your product workflows, we're passionate about helping businesses navigate those paths. Do not hesitate to reach out to exchange further!