NVIDIA announced exciting news about the future of artificial intelligence at the GTC conference in Paris. Here are seven key announcements worth highlighting.
First is the NVL72 system, equipped with 72 Blackwell GPUs. This platform has a throughput of up to 130 terabytes per second and is designed for large-scale AI solutions. It features high inference speeds and supports popular models such as DeepSeek-R1, Llama 3.1 with 405 billion parameters, and Llama 3.3 with 70 billion. It includes the second generation of the Transformer Engine, supporting the FP4 format, and the TensorRT Model Optimizer allows for quantization of models to this format.
Talking about AI factories, CEO Jensen Huang described them as “factories of intelligence.” These are large computing centers with Blackwell, NVLink, and liquid cooling, operating around the clock. The plan is to build a network of such factories across Europe so that local ideas can quickly evolve into global innovations.
Another significant trend is the creation of digital twins for everything happening in the real world. All processes and objects are first translated into a virtual environment, opening new possibilities for modeling and analysis.
The next stage is the development of Agentic AI. These systems not only observe and reflect but also act, continuously learn, and rethink their decisions—kind of agent capable of adapting and evolving.
Special attention is given to the CUDA-Q platform, which runs on the Grace Blackwell GPU. This open-source solution for developing hybrid quantum applications combines GPU, CPU, and QPU into a single system. The platform is “qubit-agnostic,” supporting any type of qubits and quantum computing architectures. In simpler terms, it allows combining classical and quantum computations in one pipeline, achieving up to 2,500 times acceleration in simulations on GPU. CUDA-Q is equipped with top-tier compilers and tools for operation and integrates with AI and HPC systems. It supports both real and emulated quantum processors.
Initiatives to develop independent AI infrastructure are underway across Europe—countries like France, Germany, the UK, Finland, Italy, and Spain are collaborating with NVIDIA in this area.
Finally, NVIDIA, in partnership with Hugging Face, launched DGX Cloud Lepton—a service providing instant access to powerful GPUs with a single click. Training large language models has become even easier with integration into the Training Cluster as a Service platform.
These are just some of the most notable innovations demonstrating how rapidly the AI landscape is evolving.