🇪🇺 The four things I learned at NVIDIA GTC 2025 🇺🇸
1. Inference at Scale – The Silent Giant
We once believed that training was the heavy lift, but inference - especially for reasoning models is rewriting that narrative. Models are no longer just delivering one-shot answers; they’re engaging in contextual, multi-step reasoning. This shift demands up to 100x more tokens in inferencing workloads, putting immense strain on compute resources and creating a surge in demand for optimized data center and edge deployments. Inference is now a full-stack, high-throughput challenge NVIDIA is ready to deliver on.
2. Private Data – The Final Frontier
With public data maxed out; medical records, private knowledge bases, and enterprise databases/files are emerging as the newest fuel for AI. These are private, regulated, and often hyper-sensitive data sets. They require dedicated infrastructure and sovereign deployment models. AI must come to the data, not the other way around. Private data is the next front for training, fine-tuning and RAG, requiring an unimaginable scale of FLOPS.
3. Robotics – AI Meets The Streets
Robots are evolving from pre-scripted tools into intelligent agents, combining LLMs with hardware in ways that allow them to understand, reason, and interact naturally. We are witnessing the fusion of NLP, real-time perception, and physical motion into a single intelligent entity. They live at the edge, in our homes, on the streets and in the factories.
This isn’t just warehouse automation or household gadgets. The long-term vision? Cars as robots. Hospitals staffed with AI-assisted tools. Elder care supported by autonomous companions. Home-assistants. Everything that's plugged in will be chatting with (and assisting) us in natural language.
4. European Private AI
With geopolitical instability rising, concerns over FISA and the U.S. Cloud Act, the risks of using U.S. controlled cloud services are escalating. These laws enable subpoenas and surveillance over any data housed on American owned clouds no matter where it physically resides. For governments, hospitals, and corporations in Europe and beyond, this is simply not an acceptable risk.
The only option remaining is open-source software and sovereign AI infrastructure. These AI solutions can be fully owned, operated, and secured by the organization itself.
NVIDIA is leaning into this shift, opening new opportunities for regional partners and infrastructure providers to build secure, high-performance environments for training and inference on private data.
What Comes Next
The future isn’t about whether AI will integrate into our lives, it’s about how quickly it will do so. From data centers to edge devices, from boardrooms to bedside monitors, from cloud APIs to sentient machines, AI will touch everything. The only limits will be creativity, manufacturing capabilitis and energy.
And finally, the BIGGEST question. Who will pay for this multi-trilion dollar tectonic shift. 💰
