Executive Summary
NVIDIA announced that leading AI inference providers, including Baseten, DeepInfra, Fireworks AI, and Together AI, are using its Blackwell platform to dramatically lower the cost of running large, open-source AI models. By combining the Blackwell architecture with optimized software stacks, these partners are achieving up to a 10x reduction in cost per token compared to the previous Hopper platform. This improvement in "tokenomics" makes scalable AI applications more economically viable for businesses across industries such as healthcare, gaming, and customer service.
Key Takeaways
* Performance Leap: The NVIDIA Blackwell platform delivers up to a 10x reduction in the cost per token for AI inference compared to the prior NVIDIA Hopper platform.
* Key Partners: Leading inference providers Baseten, DeepInfra, Fireworks AI, and Together AI have adopted Blackwell to power their services for open-source models.
* Enabling Technologies: Cost savings are driven by a full-stack approach, including Blackwell GPUs, low-precision NVFP4 data formats, and software like NVIDIA TensorRT-LLM.
* Industry Use Cases: The announcement highlights significant cost reductions for end-customers:
* Healthcare (Baseten): Sully.ai cut inference costs by 10x for medical documentation.
* Gaming (DeepInfra): Latitude reduced cost per token by 4x for its AI Dungeon game.
* Agentic AI (Fireworks AI): Sentient Foundation achieved 25-50% better cost efficiency.
* Customer Service (Together AI): Decagon lowered its cost per query by 6x for AI voice agents.
* Future Roadmap: NVIDIA teased its next-generation Rubin platform, which aims to deliver another 10x performance and cost-per-token improvement over Blackwell.
Strategic Importance
This announcement reinforces NVIDIA's dominance in the AI hardware market by demonstrating Blackwell's significant total cost of ownership (TCO) advantages for inference. By making powerful open-source models cheaper to run, NVIDIA accelerates enterprise AI adoption and positions itself as the critical infrastructure provider for the entire AI economy.