Inflection AI Plans to Build a Supercomputer Cluster with 22,000 Nvidia GPUs

Inflection AI, a startup backed by Microsoft and Nvidia and founded by the former head of DeepMind, is set to deploy a supercomputer cluster powered by 22,000 Nvidia H100 GPUs. This ambitious venture is funded through a recently secured $1.3 billion from industry heavyweights in the form of cash and cloud credit. The supercomputing system they aim to construct will have peak theoretical compute power comparable to the Frontier supercomputer.

Inflection AI currently operates a cluster running on 3584 Nvidia H100 compute GPUs in Microsoft Azure cloud. The proposed supercomputing cluster would offer about six times the performance of the current cloud-based solution.

The company’s product is a generational AI chatbot named Pi, designed to act as an AI-powered personal assistant, similar to the generative AI technology akin to ChatGPT. This tech allows Pi to interact with users via dialogue, making it possible to ask queries and provide feedback.

It is noteworthy that, while FP64 peak performance is crucial for many scientific workloads, this proposed system will likely operate much faster for AI-oriented tasks. The cost of the cluster remains unknown, but considering the Nvidia H100 compute GPUs retail for over $30,000 per unit, it is speculated that the GPUs for the cluster would cost hundreds of millions of dollars.