Rarely have two statements stood so close together – and so far apart. On CNBC last week, Sam Altman praised the efficiency leaps of ever-larger LLMs; a day earlier, Yann LeCun in Paris declared the very same architecture a dead end and championed World Models that understand the physical world.
Who will be proved right is open. What can be measured is what the race costs even today: energy on the scale of entire countries, as Google's environmental report reveals. And what it risks: forced pauses for the strongest models, autonomous ransomware, export controls now from Beijing too. Read here how industry turns all of this into manageable engineering.
Your HANNOVER MESSE robotics team
AI by blueprint
Google consumes power like Denmark. Siemens and NVIDIA deliver the blueprint for the efficient AI factory.
19 model-free days
Export control, return, chip deal: how two news items raise the AI industry's question of power.
AI, no monoculture
Anyone planning their AI architecture today treats model sourcing as a supply chain risk.
Sun Valley: Altman adds it up
On CNBC a few days ago, the OpenAI chief explains why companies suddenly ask about token efficiency – and what Washington checked before the green light.
LeCun: words are not enough
On the RAISE stage in Paris, the Turing Award winner shows why LLMs are stuck in a dead end – and why machines first have to learn the physics.
© 2026 Deutsche Messe AG
