TLDR: ATHENS—Snowflake agreed to a $6 billion, five year AWS contract for AI CPU access, including AWS Graviton, as AI drives higher compute demand. The deal signals accelerating AWS spend and rising CPU competition around enterprise workloads.
Key Takeaways:
- Snowflake runs primarily on AWS but also supports Azure and Google Cloud, with rapid AWS Marketplace growth.
- Snowflake signed for AI CPU access on AWS Graviton in a five year $6B agreement, fueled by Cortex AI usage.
- As agent style workloads shift beyond training, CPUs become a battleground where Amazon pushes price performance against Nvidia.
- The contract echoes AWS wins like Graviton deliveries to Meta, after Meta previously committed $10B to Google Cloud.
Amazon is turning AI CPU demand into leverage, and Snowflake is helping it sharpen the pitch. In this race, savings and access matter as much as raw horsepower.
Amazon is turning AI CPU demand into leverage, and Snowflake is helping it sharpen the pitch. In this race, savings and access matter as much as raw horsepower.
Q&A
Why do CPUs suddenly matter more for AI than they did during the training era?
As AI moves into everyday features and agent workflows, inference, orchestration, and background tasks multiply, pulling more work onto CPU systems.
What does Snowflake gain beyond cheaper compute when it commits to more AWS chip access?
Deeper hardware alignment can improve performance consistency for Cortex AI features, making it easier to sell upgrades to enterprises already standardized on Snowflake and AWS.
How could Nvidia respond if more cloud customers choose ARM based CPUs for agent workloads?
Nvidia may push broader software stacks, offer pricing or bundling incentives, and accelerate platform partnerships to keep GPUs central for the most demanding segments.
If AWS savings flow to customers, who will feel it first, developers or enterprise finance teams?
Developers can benefit from faster experimentation and scaling, but CFOs usually feel the impact first through lower per workload costs and more predictable bills.
What happens next if multi billion chip deals become common between cloud giants and data platforms?
Hardware access could become a competitive baseline, shifting differentiation toward tooling, reliability, and integration speed rather than just chip performance.
No comments yet. Be the first to share your thoughts!