According to new data from SemiAnalysis InferenceX reported on the NVIDIA AI Blog, the NVIDIA Blackwell Ultra platform delivers up to 50x better performance and 35x lower costs for agentic AI applications compared to previous generations.
The original NVIDIA Blackwell platform has already been adopted by leading inference providers including Baseten, DeepInfra, Fireworks AI, and Together AI, achieving cost reductions of up to 10x per token, according to NVIDIA.
The Blackwell Ultra platform represents the next evolution, specifically optimized for agentic AI workloads and coding assistants. According to NVIDIA, this new platform builds on the momentum of the original Blackwell architecture while delivering substantially greater performance improvements for AI agents.
The SemiAnalysis InferenceX benchmarking data provides independent validation of the platform’s capabilities for inference workloads, though specific details about the testing methodology were not provided in the source material. The performance gains target the growing market for AI agents and development tools that require efficient, cost-effective inference at scale.