Landscape
Landscape
Berkeley Compute's platform offers a versatile solution for a wide range of AI and compute needs. It enables AI startups to access GPU resources for model training without large capital investments, allows mobile app developers to integrate low-latency AI features, and provides investment opportunities in the AI infrastructure market through GPU NFTs. The platform also benefits datacenters looking to expand their GPU offerings, large tech companies seeking to monetize idle GPU time, and academic researchers needing short-term access to high-performance computing resources.
Training done on GPU clusters (100-1000s of GPUs). $152B CapEx from Google, MSFT, Meta only in 2024!
Inference smaller open-source models done on distributed computing (1-2 GPUs). Best $/TFLOP and TFLOP/W. Gen AI infra spend $18.3B in 2024
Fine tuning on AI server (1-8 GPUs), Confidential computing. SMB not experts in running AI servers