Snowflake’s Q1 earnings exceeded expectations, sending shares up 35% in after-hours trading. But the headline numbers—$1.39B in revenue (up 33% YoY) and $0.39 adjusted EPS—were only part of the story. The real signal came from a simultaneous $6 billion, five-year spending commitment with Amazon Web Services, covering Graviton CPUs and cloud-based GPUs for AI.

This isn’t just a procurement deal. It’s an architectural bet. Snowflake has been running on AWS since its 2020 IPO, when it disclosed a $1.2B commitment. That grew to $2.5B by 2023. The new $6B figure implies average annual spend of $1.2B—a significant escalation. But the technical detail that matters is the pivot to Graviton, Amazon’s Arm-based custom silicon.

For data engineers, this marks a concrete shift from x86 dominance. Graviton’s power efficiency makes it attractive for general-purpose compute, but Snowflake’s growing AI workload portfolio—including agentic applications that orchestrate across multiple models—demands a different CPU profile. Agentic AI tasks are heavy on sequential, data-movement operations, not just parallel matrix math. That’s where CPUs like Graviton, with fewer but more powerful cores, outperform GPUs for orchestration.

Snowflake’s partnership with Nvidia remains intact, but the AWS deal suggests a layered compute strategy: Graviton for pipeline orchestration and data shuffling, Nvidia GPUs for model training and inference. The acquisition of AI startup Natoma (undisclosed terms) reinforces this dual-path approach.

The broader trend is clear: custom Arm chips are moving from edge cases to core infrastructure. Meta already runs hundreds of thousands of Graviton chips for agentic AI workloads. Snowflake’s commitment validates that pattern for enterprise data platforms. For teams designing data pipelines, the takeaway is that CPU architecture decisions now directly impact AI workload performance and cost—not just in training, but in the increasingly expensive orchestration layer.

Source: CNBC