I think this is a pivotal strategy for $NVDA to maintain its market share as AI workloads shift from training to inference. What is important in training is raw compute power while speed, low latency, and cost efficiency take precedence in inference. As a result, AI server mix will shift from GPUs to custom ASICs over time. This means $NVDA market share will erode. Hyperscalers will be the biggest winners of this shift as they are big enough to make their own ASICs at scale. Neoclouds, on the other hand, depend on $NVDA and will keep depending on $NVDA since their customers are largely hyperscalers and they use neoclouds as a source for $NVDA capacity they need. Neocloud scale is controlled by hyperscalers as they need the hyperscaler contracts to secure the credit they require to sca
View on X ↗