As AI technology advances, the creators of these models are increasingly relying on NVIDIA's cutting-edge infrastructure. OpenAI has recently unveiled GPT-5.2, a groundbreaking model series designed for professional knowledge work. This model was trained and deployed on NVIDIA's state-of-the-art systems, including the NVIDIA Hopper architecture and GB200 NVL72 systems.
The development of AI models is a complex process, and NVIDIA's infrastructure plays a pivotal role in this journey. Three key scaling laws, including pretraining, post-training, and test-time scaling, contribute to the intelligence of AI models. Reasoning models, which tackle complex queries using multiple networks, are becoming increasingly prevalent. However, pretraining and post-training remain the foundation of AI intelligence, enabling models to become smarter and more useful.
Achieving this level of intelligence requires substantial resources. Training frontier models from scratch is a massive undertaking, demanding tens of thousands, or even hundreds of thousands, of GPUs working in harmony. This scale necessitates excellence in various aspects, including world-class accelerators, advanced networking across different architectures, and a fully optimized software stack. In essence, it's about creating a purpose-built infrastructure platform tailored for high-performance, scalable operations.
NVIDIA's GB200 NVL72 systems have demonstrated remarkable performance, delivering 3x faster training on the largest model in the MLPerf Training industry benchmarks and nearly 2x better performance per dollar compared to the NVIDIA Hopper architecture. Furthermore, the NVIDIA GB300 NVL72 architecture offers over 4x speed improvements over the NVIDIA Hopper, significantly accelerating the development and deployment of new AI models.
NVIDIA's support extends beyond language models. The company's platforms are used to train models across various modalities, including speech, image, and video generation, as well as emerging fields like biology and robotics. For instance, models like Evo 2 decode genetic sequences, OpenFold3 predicts 3D protein structures, and Boltz-2 simulates drug interactions, aiding researchers in their work. In the clinical domain, NVIDIA Clara synthesis models generate realistic medical images for advanced screening and diagnosis without compromising patient data.
Several companies, such as Runway and Inworld, have harnessed the power of NVIDIA's infrastructure. Runway recently introduced Gen-4.5, a cutting-edge video generation model that has achieved the top rating in the world according to the Artificial Analysis leaderboard. This model was developed entirely on NVIDIA GPUs, from initial research and development to pre-training, post-training, and inference, and is now optimized for the NVIDIA Blackwell architecture.
Additionally, Runway unveiled GWM-1, a state-of-the-art general world model trained on NVIDIA Blackwell. This model is designed to simulate reality in real-time, offering interactivity, control, and versatility across various applications, including video games, education, science, entertainment, and robotics. The performance of these models is evident through industry-standard benchmarks like MLPerf, where NVIDIA consistently demonstrates strong performance and versatility across all categories.
NVIDIA's Blackwell platform is widely available from leading cloud service providers, neo-clouds, and server makers. NVIDIA Blackwell Ultra, with enhanced compute, memory, and architectural improvements, is now being rolled out by server makers and cloud service providers. Major cloud service providers and NVIDIA Cloud Partners, including Amazon Web Services, CoreWeave, Google Cloud, Lambda, Microsoft Azure, Nebius, Oracle Cloud Infrastructure, and Together AI, offer instances powered by NVIDIA Blackwell, ensuring scalable performance as pretraining scaling continues.
In summary, NVIDIA's infrastructure is the backbone of AI development, enabling the creation of advanced models across various domains. From frontier models to everyday AI applications, the future of AI is being built on NVIDIA's cutting-edge technology.
