Huang also addressed the transition from Blackwell to Blackwell Ultra, noting extraordinary demand despite early production challenges. Nvidia's GPUs have been central to the AI boom, with the first-generation Hopper chip becoming highly sought after. However, competition from lower-cost models like Chinese startup DeepSeek's has raised questions about infrastructure investment for advanced LLM development. Huang highlighted the growing demand for compute, citing an example where DeepSeek's R1 model required significantly more tokens than traditional models for tasks like creating a wedding seating chart.
Key takeaways:
- Nvidia's GTC 2025 AI conference features CEO Jensen Huang discussing next-generation chipsets Blackwell Ultra and the Vera Rubin platform.
- Nvidia is partnering with General Motors to build custom AI systems using Nvidia accelerated compute platforms for vehicles, factories, and robots.
- The transition from Blackwell to Blackwell Ultra is expected to launch later this year, with high demand anticipated despite early production hiccups.
- Huang highlighted the growing demand for compute, noting that models like Deepseek's R1 require significantly more tokens than traditional LLM models.