Snowflake’s $6B AWS Pact: A New Phase of AI Infrastructure Spending
Snowflake’s USD 6 billion (approx. RM27.6 billion) commitment to Amazon Web Services is a long-term cloud infrastructure agreement in which the data platform reserves AWS capacity for AI accelerators and Graviton CPU deployment, signaling how enterprise AI workloads are reshaping software company spending and strategy. Announced as a multi‑year strategic collaboration, the deal is Snowflake’s largest cloud spend commitment to date and centers on AWS’s GPU‑accelerated EC2 instances and Arm‑based Graviton processors. Snowflake plans to run and train its generative AI models and services on this mix of cloud AI accelerators and CPUs, supporting products like Cortex AI for text‑to‑SQL, summarization, and sentiment analysis. The commitment also extends a go‑to‑market partnership where AWS Marketplace has already generated more than USD 7 billion (approx. RM32.2 billion) in lifetime sales for Snowflake, underlining how tightly infrastructure, distribution, and AI revenue ambitions are now linked.
From Warehouse to AI Platform: Why Snowflake Needs Dedicated Cloud AI Accelerators
Snowflake’s pivot from a cloud data warehouse to what CEO Sridhar Ramaswamy calls “the platform for the AI era” explains the scale of its Snowflake AWS infrastructure commitment. Cortex AI and Cortex Code depend on fast access to enterprise AI workloads that run close to governed data, so securing reliable GPU and CPU capacity becomes a strategic necessity rather than an optional optimization. According to AWS, Snowflake’s lifetime AWS Marketplace sales crossed USD 7 billion (approx. RM32.2 billion) and exceeded USD 2 billion (approx. RM9.2 billion) during the 2025 calendar year, suggesting a growing share of its business already flows through Amazon’s cloud. By locking in access to GPU‑accelerated instances, Snowflake aims to keep training and inference capacity available for agentic AI systems that plan, retrieve data, and call tools. The bet is clear: future revenue growth will come from AI‑driven applications, not only traditional warehousing.
Graviton CPUs and Custom Silicon: The New Battleground for Cloud AI
A defining feature of this deal is how heavily it leans on Graviton CPU deployment alongside cloud AI accelerators. Snowflake has shifted an increasing share of compute from Intel and AMD to AWS’s Arm‑based Graviton instances, which now include fifth‑generation chips with 192 Arm Neoverse V3 cores and high‑bandwidth memory. These CPUs handle the orchestration layer around AI models—SQL queries, Python scripts, and tool calls—while Nvidia GPUs and other accelerators run the models themselves. This combination shows how cloud providers are competing not only on GPU supply but also on custom silicon stacks tuned for AI‑heavy software. While AWS’s Trainium chips are not emphasized in this announcement, the focus on GPU‑accelerated EC2 and Graviton highlights a broader trend: CPUs are back in the spotlight as critical infrastructure for agentic AI workloads that must respond quickly and at scale.
AI Infrastructure on the Balance Sheet: New Economics for Software Companies
Snowflake’s AWS commitment illustrates how AI infrastructure spending is moving directly onto software balance sheets. In earlier cloud eras, software vendors tried to stay light on infrastructure commitments to preserve flexibility and high gross margins. AI has changed the equation. To sell serious AI products, companies now pre‑book vast amounts of compute, accepting more capital intensity in exchange for dependable capacity. Snowflake’s USD 6 billion (approx. RM27.6 billion) spend over five years means AI infrastructure costs become a major, long‑lived line item, tying future margins to how efficiently those resources are used. If Snowflake’s AI services drive usage and revenue, it can defend a premium position as a core AI platform. If demand lags, the fixed commitments weigh on profitability. Either way, the deal signals an industry shift where the boundary between software business models and cloud hardware procurement is increasingly blurred.