Nvidia to Sell 1 Million GPUs to Amazon Web Services by End of 2027
Key Takeaways
- ▸Nvidia will supply 1 million GPUs to AWS by 2027, with shipments beginning in 2024
- ▸The deal includes a diverse mix of Nvidia chips beyond GPUs, including Spectrum networking chips and newly released Groq chips
- ▸AWS plans to deploy a seven-chip combination optimized for AI inference, emphasizing that efficient inference requires multiple specialized chip types
Summary
Nvidia has announced a major supply deal with Amazon Web Services (AWS) to deliver 1 million graphics processing unit chips, along with additional hardware including Spectrum networking chips and Groq chips, through 2027. The sales will commence this year and extend over the multi-year period, representing a significant commitment from AWS to expand its AI infrastructure capabilities. While financial terms were not disclosed, Nvidia VP Ian Buck indicated the transaction encompasses a diverse portfolio of chips optimized for AI inference workloads.
The deal is particularly notable for its focus on inference efficiency, with AWS planning to deploy a combination of Nvidia's Groq chips and six other Nvidia chips to optimize AI systems' ability to generate answers and perform tasks for users. Additionally, AWS will integrate Nvidia's Connect X and Spectrum X networking equipment into its data centers alongside custom networking infrastructure, marking a collaborative expansion of AWS's AI infrastructure. This partnership aligns with Nvidia CEO Jensen Huang's projection of a $1 trillion sales opportunity for the company's Rubin and Blackwell chip families through 2027.
- Nvidia's networking equipment (Connect X and Spectrum X) will be integrated into AWS data centers alongside AWS's custom networking infrastructure
Editorial Opinion
This mega-deal underscores Nvidia's dominant position in supplying AI infrastructure to cloud hyperscalers and validates the company's multi-chip strategy for tackling the computationally intensive inference problem. The inclusion of Spectrum networking gear and Groq chips signals that Nvidia sees the AI economy extending far beyond GPU processing, encompassing entire system architectures. AWS's willingness to integrate Nvidia's networking solutions into its data centers—a traditionally guarded area—reflects the intense competition for AI infrastructure superiority and suggests that performance gains may justify architectural compromises.
