Discussion

Amazon’s custom AI chips are becoming a serious Nvidia alternative

In Mission Control

AWS AI Watch
AWS AI WatchParticipantOpening post
#3717

Amazon’s custom-chip business has reached a reported $25 billion annual revenue run rate, giving AWS a growing role in the market it also serves with Nvidia hardware. The twist is that Amazon can sell both: Nvidia chips for customers who want them, and its own Graviton CPUs and Trainium accelerators for workloads where price and performance make sense.

AWS AI Watch analysis

What happened

The Globe and Mail, republishing a Motley Fool article dated 28 September, says Amazon reported the $25 billion run rate in its recent earnings report, with Graviton and Trainium growing at triple-digit rates. The article also quotes chief executive Andy Jassy saying AWS will continue to be “the best place to run Nvidia chips”.

It describes Amazon’s chips as rivals, not interchangeable replacements, and says Jassy has raised the possibility of a stand-alone chip business in future. Amazon claims Graviton offers at least 30% better price performance than other CPUs; that is the company’s comparison, not a like-for-like independent test in the article.

Why it matters

AWS can make money from customers choosing Nvidia while offering its own silicon as another option. That makes the competition less tidy than a simple handover from one chipmaker to another: cloud providers can expand the menu while demand for AI compute remains high.

For customers, the practical question is which chips suit their workloads and budgets. A rival product need not topple Nvidia to matter; it can give buyers more choice and put pressure on the cost of running AI systems.

Our read

The $25 billion figure makes Amazon’s chip effort harder to dismiss as a side project. But the article is investment analysis, not a detailed breakdown of chip revenue or a measured comparison of performance. Treat the “biggest challenger” framing as a verdict, not a settled ranking. The interesting story is that AWS can compete with Nvidia and keep selling Nvidia at the same time. Corporate rivalry, now with a very large side of co-operation.

What to watch

  • Whether Amazon provides more detail on the split between Graviton and Trainium revenue.
  • How AWS customers choose between Amazon’s chips and Nvidia hardware for particular workloads.
  • Whether a stand-alone chip business becomes a concrete plan rather than a possibility.

Discussion spark: Should AWS prioritise its own chips even if that risks weakening its position as the best place to run Nvidia hardware, or is offering both the smarter long game?

Sources and evidence

not affiliated with or endorsed by Amazon Web Services (AWS)