Tech • AI • Robotics • Game
NVIDIA GTC Berlin 2026 | Live Keynote Pregame
7/10TL;DR
Europe’s AI push is accelerating around open models, sovereign infrastructure, industrial deployment and robotics, but leaders say the region still needs far more compute capacity to compete at global scale.
KEY POINTS
Europe’s AI strategy is shifting from research to full-stack buildout
Industry leaders framed Europe as a region with strong engineering, science and industrial depth that now must translate those advantages into AI infrastructure, software and deployment. The central argument was that Europe already has the talent and industrial demand, but needs to build the “whole stack,” from models and data centers to enterprise applications and edge networks.
Open models are emerging as a strategic priority
Mistral AI chief executive Arthur Mensch said the company’s latest open model, Magistral, is performing at state-of-the-art level in areas including industrial engineering, cyber capabilities and geospatial imaging, in some cases outperforming closed models. He argued that open models are critical for Western competitiveness because they give companies portability, cost control and the ability to keep AI close to sensitive data.
Sovereign AI is being defined as control, continuity and cost discipline
Executives described sovereign AI less as national isolation and more as the ability for companies and governments to retain control over infrastructure, intellectual property and business continuity. That includes the option to move workloads across providers, keep critical data local and avoid dependence on centralized model vendors. The concept was also tied to affordability, since open models can reduce the premium associated with proprietary systems.
German telecom infrastructure is becoming a key AI battleground
Deutsche Telekom chief executive Tim Höttges said the company’s first 10,000 GPUs in Munich sold out within six months, prompting a larger expansion and plans for a future AI gigafactory. He said the company is building a sovereign AI stack in Germany and Europe, combining network access, local data centers, NVIDIA compute and cloud services to support developers and enterprises that do not want sensitive workloads hosted abroad.
Industrial AI is moving from concept to measurable factory economics
Dassault Systèmes chief executive Pascal Daloz said virtual twins are now being used to design AI factories, including by NVIDIA, with the goal of reducing time to deployment and speeding “time to first token.” He argued that industrial AI must be grounded in science, simulation and physics, not just language capability, because manufacturers need to know whether materials, components and processes will work before committing capital.
Factory downtime is becoming a prime AI use case
In manufacturing, AI’s clearest return on investment is emerging in predictive maintenance and operational planning. Mensch said AI agents can detect when equipment is likely to fail, coordinate maintenance with customer downtime and protect revenue tied to service-level agreements. For companies running expensive industrial systems, that can save hundreds of millions by keeping machines operational.
European companies are reorganizing around AI rather than layering it on
Höttges said AI is not being treated as another software tool but as a redesign of operating models and governance. At Deutsche Telekom, 90% of mobile network traffic is already managed by AI-driven autonomous systems, while software development teams are seeing productivity gains of up to 60%. He also said 26% of the company’s roughly 200,000 employees are now trained as AI experts.
The biggest bottleneck remains compute
Investor Jeannette zu Fürstenberg of General Catalyst said Europe’s application startups can scale globally, but infrastructure remains constrained by insufficient compute. She called for 20 gigawatts of AI capacity in Europe to support local champions and said the region risks underperforming its potential unless capital, procurement and industrial partnerships move faster.
AI deployment is becoming multi-model and multi-agent
Leaders rejected the idea that one model will handle every enterprise task. Mistral AI said it runs around 30 token-intensive internal AI applications, while Deutsche Telekom said it has 38 models in production, including multimodal systems and embeddings. The dominant architecture described was a combination of large general-purpose reasoning models with smaller specialized models and agents embedded in workflows, software systems and robotics.
Autonomy, healthcare and humanoids are being pitched as the next large societal gains
Mercedes-Benz chief executive Ola Källenius said about 1 million people die in road accidents globally each year and argued that autonomous driving could drive that figure toward zero while giving users back hours of time. In healthcare, AI was highlighted for early cancer detection, diagnosis and treatment planning. In robotics, executives said humanoids are likely to appear in factories before homes, with broader household use potentially arriving in three to five years.
CONCLUSION
Europe’s AI race is no longer centered only on model quality. The decisive question is whether the region can rapidly build enough compute, infrastructure and industrial deployment capacity to turn technical strength into lasting economic power.
Ask a question