Snowflake · 2026-05-27 · major
Snowflake Commits $6B to AWS Over Five Years — Graviton ARM CPUs and GPU EC2 Power Cortex AI as 'Agentic Enterprise' Lifts Snowflake's AWS Spend From $1.2B at IPO to $6B Today
Snowflake more than doubled its AWS commitment to $6B over five years, leaning on Amazon's Graviton ARM CPUs for agentic workloads and GPU EC2 for model training. The deal lifts Snowflake's annual AWS spend trajectory from $1.2B at IPO to $2.5B in 2023 to $6B now.
Snowflake bets $6B on AWS Graviton ARM CPUs to power the round-the-clock workloads its Cortex AI agents create.
Key specs
| Deal size | $6B |
|---|---|
| Duration | 5 Years |
| Snowflake aws marketplace lifetime sales | $7B+ |
| 2025 customer aws spend on snowflake | $2B |
| Snowflake aws commitment progression | $1.2B (2020 IPO) → $2.5B (2023) → $6B (2026) |
What is it?
This is a multi-year Strategic Collaboration Agreement between Snowflake and AWS announced May 27. Snowflake will direct $6B of infrastructure spend to AWS over five years, primarily for Graviton ARM-based CPU compute and GPU-accelerated EC2 instances. The agreement lands the same week Snowflake reported a blowout Q1 and announced its $6B Natoma acquisition, sending the stock up 36% on the day.
How does it work?
Cortex AI, Snowflake's agentic data layer, runs natural-language queries, semantic agents, and Cortex Analyst directly on customer warehouses. Those agents create steady background CPU load that's traditionally been the wrong shape for GPUs, so Snowflake is leaning on AWS Graviton — Amazon's home-grown ARM CPUs — for the persistent inference and orchestration tier. GPU EC2 stays for model training. The deal also gives Snowflake earlier access to new Graviton silicon as Amazon races Nvidia on price-performance.
Why does it matter?
It's the largest infrastructure commitment in Snowflake's history and the latest in a string of multi-billion-dollar AWS commitments from Anthropic, OpenAI, Meta, and now Snowflake. Together they signal that the 'agentic era' is a CPU story as much as a GPU story — agents don't only train, they run forever. For AWS, locking in Snowflake on Graviton is the most concrete proof yet that home-grown silicon can win flagship AI workloads against Nvidia.
Who is it for?
Data infrastructure teams, agentic-AI architects, cloud strategists
Try it
https://www.snowflake.com/cortex