What Snowflake’s $6B AWS Commitment Really Means
Snowflake’s USD 6 billion (approx. RM27.6 billion) AWS infrastructure commitment is a multi‑year cloud software spending bet that shows how AI is pushing software vendors to secure dedicated compute, including custom CPUs and accelerators, so they can control performance, cost, and differentiation at the infrastructure level rather than only in application code. The agreement, spread over five years, centers on Snowflake AWS infrastructure built on Graviton CPU deployment and GPU‑accelerated EC2 instances used for AI training and inference. AWS described it as Snowflake’s largest cloud spend commitment to date and tied it directly to the company’s AI ambitions. Under CEO Sridhar Ramaswamy, Snowflake is repositioning itself from a data warehouse into a “platform for the AI era,” with its Cortex AI and agentic tools depending on reliable access to large pools of compute.
Graviton CPUs, GPUs and the New AI Compute Stack
At the heart of the deal is a deliberate mix of Arm‑based Graviton CPUs and GPU‑accelerated instances. Snowflake has already shifted more compute from Intel and AMD chips to Amazon’s Graviton line, which in its latest generation packs 192 Arm Neoverse V3 cores and 12 memory channels. These CPUs will support Snowflake’s traditional warehousing and query workloads, making room in its AI compute investment budget for higher‑cost training and inference on GPUs. While AWS also offers Trainium accelerators, the announcement focuses on “GPU‑accelerated” EC2 instances, likely reflecting Snowflake’s desire to avoid vendor‑locked silicon for model training. The architecture lines up with how AI systems run in practice: GPUs handle model execution, while the agents, SQL translation, Python tools, and orchestration logic that wrap those models still rely heavily on CPU throughput.
Software Companies Are Quietly Becoming Infrastructure Players
Snowflake’s AWS move is part of a wider pattern: AI is pulling software companies toward the hardware layer. In the past, vendors stayed asset‑light, avoiding deep infrastructure bets because they added capital intensity and operational risk. Now, dependable AI performance demands firm commitments to underlying compute. According to Startup Fortune, the USD 6 billion (approx. RM27.6 billion) agreement shows “how far data platform companies are willing to go to secure compute for the next phase of their AI businesses.” These commitments no longer sit only with hyperscalers; they show up on software balance sheets as forward cloud spend, affecting gross margins and operating leverage. The payoff is potential AI differentiation and guaranteed capacity. The risk is clear too: if AI usage underperforms, those long‑dated infrastructure obligations can drag on profitability and future platform flexibility.
AI Workloads Are Reshaping Cloud Alliances and Economics
The Snowflake AWS infrastructure pact also tightens their commercial and go‑to‑market ties. Snowflake has surpassed USD 7 billion (approx. RM32.2 billion) in lifetime AWS Marketplace sales and more than USD 2 billion (approx. RM9.2 billion) in sales during a single calendar year, giving AWS a strong role not only as infrastructure provider but as a distribution channel. That dual role blurs traditional lines between software vendor and cloud platform. AI workloads such as Cortex AI’s text‑to‑SQL, summarization, and sentiment analysis demand predictable capacity, which favors large, pre‑negotiated compute blocks over ad‑hoc consumption. AWS, for its part, strengthens its position as a default home for enterprise AI, with revenue from the unit rising 28% in a recent quarter, while software partners trade some independence for access to scale, security, and silicon roadmaps tailored to agentic AI.
Control of Compute Becomes a Core Competitive Advantage
Snowflake’s pivot toward AI agents and governed data makes control over compute more than an efficiency play; it becomes strategic. The company wants to power what Ramaswamy calls “the era of the agentic enterprise,” where AI agents plan tasks, call tools, and act across business systems. For that to work at scale, Snowflake needs predictable performance on both CPUs and accelerators, tight integration with storage, and pricing it can forecast years ahead. That is why software‑only differentiation is no longer enough. Leading AI vendors now compete on the shape of their infrastructure deals and their alignment with chip roadmaps as much as on model features. In this sense, Snowflake’s USD 6 billion (approx. RM27.6 billion) AI compute investment is not an outlier but an early signal of how software companies are becoming infrastructure players to defend their AI edge.