There has been a recent trend of large, well-funded AI startups amassing a significant number of NVIDIA GPUs.
Nvidia's investment arm, NVentures, has been actively investing in AI startups, including Cohere, Hugging Face, and Inflection, among others. These investments aim to support the development of advanced AI models that require substantial computing power.
Examples of this trend are included below:
Inflection AI
Along with its partners CoreWeave and NVIDIA, Inflection AI is building the largest AI cluster in the world comprising 22,000 NVIDIA H100 Tensor Core GPUs. In just over a year, Inflection AI has developed one of the most sophisticated large language models in the market to enable people to interact with Pi, your Personal AI (pi.ai), in the most simple, natural way and receive fast, relevant and helpful information and advice.’
Anthropic
Anthropic estimates its frontier model will require on the order of 10^25 FLOPs, or floating point operations — several orders of magnitude larger than even the biggest models today. Of course, how this translates to computation time depends on the speed and scale of the system doing the computation; Anthropic implies (in the deck) it relies on clusters with “tens of thousands of GPUs.”
Cohere
Through the partnership, Cohere will train, build, and deploy its generative AI models on OCI. OCI is uniquely positioned to run AI workloads as it delivers the highest performance and lowest cost GPU cluster technology, with scale of over 16K H100 GPUs per cluster, and very low latency and the highest bandwidth RDMA network in the cloud. This will enable the acceleration of large language models (LLM) training while simultaneously reducing the cost.
Source: https://txt.cohere.com/oracle-to-deliver-powerful-and-secure-generative-ai-services-for-business/
Imbue
Models. We pretrain our own very large (>100B parameter) models, optimized to perform well on internal reasoning benchmarks. Our latest funding round lets us operate at a scale that few other companies are able to: our ~10,000 H100 cluster lets us iterate rapidly on everything from training data to architecture and reasoning mechanisms.