In detail
- Chip is optimized for inference, not training—similar to Apple's shift from Intel to custom processors.
- Goal: greater hardware control, better performance for specific workloads, reduced single-supplier risk.
- Trend: Google, Apple, and SpaceX are also building custom chips—Big Tech is diversifying away from pure Nvidia dependence.
Why it matters
Custom chips are a strategic move to cut costs and diversify supply chains. For you as an SME, this means: long-term AI inference costs could fall, but vendor lock-in remains a risk.
For you Watch whether custom chips drive down prices for AI APIs—that could improve margins on AI-powered products.