OpenAI Google Amazon Microsoft build custom AI chips for inference
Big names like OpenAI, Google, Amazon, and Microsoft are now making their own AI chips. They want to save money and have more control over how their systems run, especially when it comes to inference (basically, when AI answers your questions).
While NVIDIA's GPUs are still the go-to for training these models, custom chips let companies fine-tune performance for specific tasks.
Custom chips complement NVIDIA software tools
OpenAI just rolled out its Jalapeno chip (built with Broadcom and TSMC) for internal use.
Google uses TPUs, Amazon has Trainium and Inferentia, Microsoft runs Maia accelerators in Azure, and Meta's MTIA chips help boost recommendations and generative AI.
Elon Musk's xAI is also scaling up with new data centers for its Grok models.
Even though these custom chips are getting popular, NVIDIA remains essential thanks to its powerful software tools, so think of these new chips as teammates rather than replacements.