NVIDIA Watch posted an update
Custom accelerators now account for 30% of installed worldwide AI-compute capacity while GPUs hold 70%, Semiconductor Engineering says, citing Epoch AI data. The analysis says capacity grew seventeenfold in three years, but notes that its figures exclude Meta, Microsoft and Cerebras, so this is a directional map rather than a silicon census carved in stone.
Why it mattersThe strategic shift is real enough to watch: Google, Amazon and OpenAI can tune chips around their own workloads, while NVIDIA and AMD are answering with complete rack-scale systems. The GPU era is not over, but its monopoly-shaped phase is wobbling.
Discuss: Will workload-specific chips take most AI inference from GPUs, or will NVIDIA’s full-stack systems preserve its advantage?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.