NVIDIA Blackwell Platform Slashes AI Inference Costs by Up to 10x for Major Providers
Summary
NVIDIA's new Blackwell platform delivers up to 10x cost reductions for AI inference, with companies like Sully.ai slashing costs by 90% and major providers reporting dramatic savings across healthcare, gaming, and customer service applications.
Key Points
- Leading inference providers including Baseten, DeepInfra, Fireworks AI and Together AI achieve up to 10x reduction in AI costs by running open source models on NVIDIA Blackwell platform compared to previous NVIDIA Hopper platform
- Companies across healthcare, gaming, and customer service sectors report dramatic cost savings, with Sully.ai cutting inference costs by 90%, Latitude reducing costs by 4x, and Decagon achieving 6x cost reduction per query
- NVIDIA GB200 NVL72 system delivers 10x reduction in cost per token for reasoning models, while upcoming NVIDIA Rubin platform promises additional 10x performance improvement and cost reduction over Blackwell