What happened
Snowflake has signed a new five-year, $6 billion agreement with Amazon Web Services. The deal, reported by TechCrunch, will give the data cloud company increased access to AWS's custom ARM-based Graviton CPU chips. Snowflake says its customers are accelerating their spending on AWS, driven by AI tools like its Cortex AI offering. This single contract is nearly equivalent to all the revenue Snowflake has ever generated via the AWS Marketplace since its founding.
How the room's reading it
This deal is being read as another major signal of compute diversification. While Nvidia dominates the GPU market for AI training, the move highlights the growing importance of CPUs for inference and agentic workloads. Infra teams on X note that this is a big win for AWS's custom silicon strategy, following a similar Graviton deal with Meta. It's seen as a direct challenge to Nvidia's dominance, proving that cloud providers can lure huge contracts with their own, more affordable chips — a trend also seen with Google's TPUs and Microsoft's Maia.
Sailfish's take
We see this as more than just a cost-saving play. It's a bet on the future shape of AI workloads. Everyone is chasing scarce GPUs, but the real, unglamorous work of running AI agents at scale — orchestration, data handling, simple tool use — is heavily CPU-bound. Snowflake is preparing for a world where its customers' AI bills are driven by inference and agent activity, not just model training. We've shipped enough production AI to know that compute costs rarely scale the way you expect. This suggests the smart money is now optimising for the entire stack, not just the training cluster.