Snowflake and AWS Reach $6 Billion Agreement for AI Infrastructure

What happened

Snowflake has signed a five-year, $6 billion deal with Amazon Web Services that places custom Graviton CPUs at the center of its next phase of AI infrastructure growth. The agreement is striking not only because of its size, but because of what it says about where enterprise AI spending is moving. Rather than focusing only on giant GPU clusters for training frontier models, Snowflake is betting heavily on the infrastructure needed for inference, data workloads, and AI agents that must run constantly and economically in production environments.

That distinction matters because the economics of AI are changing. Training remains prestigious and expensive, but the day-to-day cost of serving models, automating workflows, and scaling enterprise AI products may increasingly depend on the price-performance of non-GPU compute.

What's new in this update

The most important detail is the emphasis on AWS Graviton chips. These ARM-based CPUs are not merely generic cloud resources in this context. They are being positioned as a strategic alternative for workloads where Nvidia-class GPU dominance is not the only answer. Snowflake's Cortex AI and agent-driven services require scalable inference and repeated operational execution, and AWS wants Graviton to be the architecture that makes those workloads cheaper and more defensible over time.

The size of the deal also signals Snowflake's confidence that enterprise customers will keep expanding AI usage on top of its platform. If customer demand for generative AI, automation, and data-centric AI products were not expected to rise substantially, a five-year $6 billion commitment would be much harder to justify.

Key details

Snowflake has long been one of the strongest data-platform companies in the cloud era, and its strategic importance grows when enterprises want AI systems that can sit close to governed data rather than operate as isolated chat tools. This AWS agreement appears designed to ensure that Snowflake can serve that role at scale without relying exclusively on the highest-cost chip tiers.

Several themes emerge from the partnership:

• Snowflake is nearly matching its entire prior historical AWS spend with one forward commitment.
• AWS is using Graviton as a flagship answer to the inference and agent-compute economy.
• Enterprises are spending more inside Snowflake as generative AI products move toward production.
• The AI chip market is broadening beyond the headline race around Nvidia training hardware.

This is why the deal has strategic significance for both companies. Snowflake gets a large-scale infrastructure path for AI growth, and AWS gets a major proof point that custom silicon can win serious enterprise commitments.

Background and context

The AI infrastructure market is entering a phase where training and inference are being distinguished more sharply. Nvidia remains dominant in the training narrative, but cloud providers increasingly want customers to believe that the long-term revenue pool lies in repeated, everyday AI execution across business workflows. That is where CPUs, custom accelerators, and vertically integrated cloud economics start to matter much more.

Snowflake's role is especially interesting because it sits between data warehousing, application logic, and enterprise AI services. If companies want AI features embedded in data products and governed business systems, Snowflake is well placed to capture that demand. The AWS partnership suggests that both firms see that opportunity as large enough to justify unusually deep infrastructure alignment.

What to watch next

The next thing to watch is whether Snowflake's AI products drive enough real customer usage to validate this level of long-term infrastructure commitment. Signed capacity is one thing. Durable monetization from Cortex AI, automated agents, and related services is another. If enterprise adoption stays strong, the agreement could become a template for how cloud and data-platform players co-invest around AI economics.

It will also be worth watching the competitive response. Google, Microsoft, and Nvidia all have reasons to challenge the idea that CPU-centric or custom-cloud-chip strategies can capture the most valuable parts of the AI runtime market.

Why this matters

This matters because Snowflake, AWS, Graviton, cloud computing, AI chips, Cortex AI, and the broader shift toward inference-heavy enterprise AI are all wrapped into the same strategic bet. The $6 billion deal suggests the next major infrastructure prize may not be only who trains the biggest models, but who can run AI cheaply, repeatedly, and at enterprise scale. That is where custom cloud silicon could become a decisive competitive layer.

Reader context

This story belongs to Northstar Herald's Generative AI and AI Infrastructure coverage, with related entities including Snowflake, Amazon Web Services, AWS, Graviton. The report is based on TechCrunch AI source material.

Related coverage

• More ai coverage
• Generative AI updates
• Earlier coverage: Payroll Startup Remote Reports 50% Revenue-Per-Employee Growth via AI Adoption