4 ms·

I think the takeaway here is a little different. In a world where everyone is desperate for GPUs, Oracle has spare GPUs to rent out to competitors. What does

• mike_hearn 3y ago
  Apparently the demand is large, as Microsoft is not a small customer by any measure. The actual cause here is something else: NVIDIA have been allocating GPUs according to long term corporate priorities rather than just auctioning them off and letting price signals do the work. The reason is twofold: 1. They want a diversity of customers for GPUs, not just one or two LLM mega-players that everyone else invokes via high level cloud APIs. That would put NVIDIA in a bad negotiating position. 2. They don't want GPU supply to be dominated by companies that are simultaneously designing their own silicon. Microsoft is. [1] Oracle has been getting large allocations because they aren't designing their own AI hardware (or at least, if they are, this isn't public), and because they are in the business of selling CUDA and not hardware abstracted model access. [1] https://www.theverge.com/2023/4/18/23687912/microsoft-athena-ai-chips-nvidia https://www.theverge.com/2023/4/18/23687912/microsoft-athena...