Snowflake has signed a five-year, $6 billion contract with Amazon Web Services โ one of the largest corporate cloud agreements in the history of the sector. Announced on May 27, 2026, the deal centers on access to Graviton, Amazon's own ARM-based CPU chips, to power Snowflake's AI infrastructure.
Key takeaways
- The contract totals $6B over 5 years โ nearly equal to Snowflake's cumulative AWS Marketplace revenue since 2012 ($7B total)
- Snowflake customer spending on AWS doubled in 2025 to $2B for the calendar year
- Key component: Graviton (Amazon's ARM-based CPU) chips for AI agent workloads
- Snowflake is building out Cortex AI โ an enterprise AI tool operating directly on cloud-resident data
- Within a single month, AWS signed comparable deals with Meta (millions of Graviton chips) and now Snowflake
Snowflake and AWS: the scale of the deal
The five-year, $6 billion contract is comparable in size to the total value of all Snowflake transactions through the AWS Marketplace since the company was founded in 2012 โ AWS puts that cumulative figure at $7 billion. In practical terms, the new agreement is set to nearly match fourteen years of accumulated business in just five years. The driving force is AI: Snowflake customer spending on AWS infrastructure doubled in 2025 alone, surpassing $2 billion for the calendar year.
Snowflake built its business on AWS from the start, with Microsoft Azure and Google Cloud added as deployment options later. This makes the contract a logical extension of an existing relationship โ but the scale of growth suggests more than inertia: enterprise companies are rapidly scaling their data processing and AI spend in the cloud.
Graviton: CPU over GPU
The detail that distinguishes this contract from a typical cloud deal is its emphasis on Graviton processors โ Amazon's own ARM-designed CPUs. Not GPUs, not FPGAs, but conventional (though highly optimized) CPUs. Why?
Large language models require GPUs for training and inference. But AI agents โ which became the dominant deployment pattern in 2025 and 2026 โ generate a very different compute profile: tool-call orchestration, context memory management, session state handling, routing logic. These are CPU-intensive tasks. Snowflake Cortex AI, the core of the company's AI offering, serves exactly these operational layers.
Amazon has consistently positioned Graviton as a cheaper, more cost-efficient alternative to Nvidia for this category of work, claiming it passes those savings along to customers. The deals with Meta in April 2026 (millions of Graviton chips) and now Snowflake suggest this narrative is landing effectively.
Cortex AI: enterprise AI on your own data
Snowflake Cortex AI is a toolkit for building AI applications directly on data stored in the Snowflake warehouse โ without copying it to external systems. Capabilities include: a natural-language interface for SQL queries, automated summary reports, semantic search across documents, and agentic workflows integrating data from multiple sources.
For enterprise customers, this is a meaningful value proposition: data never leaves the Snowflake environment, simplifying regulatory compliance (GDPR, HIPAA) and reducing latency. The alternative โ exporting data to external AI platforms โ incurs additional transfer costs, security complexity, and synchronization overhead.
The chip context: Amazon versus Nvidia
The Snowflake deal fits into a broader AWS campaign to reduce dependence on Nvidia โ or at least to offer enterprise customers a credible alternative when GPU costs become prohibitive. Jensen Huang announced in May 2026, following a record quarter, that Nvidia's new CPU chip โ Vera โ opens a "brand new $200 billion market" for the company, and that he had already sold $20 billion worth. That was a direct response to growing pressure from cloud-native chips.
The situation is more nuanced than a simple rivalry, however. AWS still deploys Nvidia chips at scale in its cloud โ Graviton does not replace GPUs, it fills a different niche: cheaper compute for tasks that do not require GPU throughput. In practice, Snowflake will likely use both Graviton (for Cortex and agents) and Nvidia GPUs (for model training and heavy inference). The contract is not a statement of vendor change โ it is a bulk purchase for a specific workload profile.
Why this matters
The $6 billion Snowflake-AWS contract is another data point confirming a fundamental shift in enterprise IT spending: AI is moving from experiment to core operational infrastructure. The fact that Snowflake customer spending on AWS doubled in a single year โ to $2 billion โ indicates that companies are running production AI systems, not just pilots. For AWS, the agreement is a negotiating chip with other large customers: "Snowflake chose Graviton at $6 billion scale โ you can too." For Nvidia, it signals that dominance in the AI chip market is being challenged not by one competitor but by an entire class of cloud-native chips โ Google TPU, Microsoft Maia, Amazon Graviton and Trainium โ which are beginning to capture specific workload segments.
What's next?
- Snowflake plans to expand Cortex AI capabilities toward autonomous enterprise agents โ further product announcements are expected at Snowflake Summit 2026, scheduled for June.
- AWS is competing with Google Cloud for contracts with other major AI customers โ Meta signed $10B with Google Cloud, then a comparable chip deal with AWS weeks later, suggesting enterprises are diversifying rather than picking a single provider.
- Nvidia's Vera CPU chip is expected to reach broad availability in H2 2026 โ its pricing and supply will determine whether Amazon Graviton can sustainably capture the CPU-for-AI segment.
Sources
- TechCrunch โ In more good news for Amazon, Snowflake signs $6B deal with AWS for AI CPU chips
- Snowflake โ Snowflake and AWS partnership announcement
- TechCrunch โ Meta signs deal for millions of Amazon AI CPUs
