Snowflake’s $6 Billion AI Bet on AWS, Defined
Snowflake’s USD 6 billion (approx. RM27.6 billion) AWS infrastructure commitment is a five-year pact to run its AI data warehouse and analytics services on Amazon’s custom Graviton CPUs and cloud AI accelerators, signalling a strategic shift toward AI-heavy workloads and tighter cloud alliances across the enterprise software market. The agreement, announced by AWS, is Snowflake’s largest cloud spend so far and centers on ARM-based Graviton compute and GPU‑accelerated EC2 instances for model training and inference. AWS did not highlight Trainium specifically, suggesting Snowflake will rely mainly on Nvidia-powered instances alongside Graviton CPUs. This dual focus links cost‑efficient CPU capacity for classic data warehousing with premium AI compute for generative and agentic workloads. For a company that runs across multiple clouds, concentrating such a large commitment on Snowflake AWS infrastructure shows how critical dependable compute has become to its growth story in the AI era.
From Cloud Data Warehouse to AI Data Platform
Under CEO Sridhar Ramaswamy, Snowflake is repositioning itself from a cloud data warehouse to “the platform for the AI era,” and the AWS deal is the infrastructure backbone for that pitch. Cortex AI, Snowflake’s suite of AI services, runs text‑to‑SQL, summarization, sentiment analysis, and entity extraction directly on governed data within its platform, turning the warehouse into an AI application layer. According to The New Stack, Snowflake will use AWS’s cost‑efficient Graviton instances to support traditional warehousing while reserving GPU‑accelerated capacity for intensive AI training and inference. That balance matters: CPUs handle the surrounding SQL and Python orchestration, while GPUs power the core models. As agentic systems grow more complex, each AI agent’s responsiveness depends on both fast CPU servicing and scalable accelerators, tightening the link between Snowflake’s product roadmap and AWS’s evolving chip portfolio.
Why Graviton CPUs Matter for Enterprise AI Economics
At first glance, Graviton CPUs might seem secondary to cloud AI accelerators, but they are central to how Snowflake manages cost and performance. As The Register notes, Snowflake has shifted an increasing amount of compute from Intel and AMD chips to AWS’s ARM‑based Graviton line, now in its fifth generation with up to 192 Arm Neoverse V3 cores. In AI data warehouse environments, GPUs run the large models, yet surrounding tasks — SQL queries, Python code, data preparation — still run on CPUs. For agentic workloads that call tools, coordinate workflows, and hit databases repeatedly, CPU bottlenecks quickly drag down perceived AI performance. By standardizing on high‑density Graviton CPUs for these layers, Snowflake aims to contain infrastructure costs while keeping latency low, so premium GPU time is reserved for the workloads where it adds the most value.
AI Infrastructure Moves Onto Software Balance Sheets
The Snowflake AWS infrastructure commitment also reveals a broader financial shift: AI is pulling software companies closer to the hardware layer. Startup Fortune notes that Snowflake’s USD 6 billion (approx. RM27.6 billion) agreement is a spending commitment, not just a partnership slide, which effectively moves more AI infrastructure risk onto Snowflake’s balance sheet. In the previous SaaS model, vendors tried to avoid heavy infrastructure obligations; now, dependable access to compute becomes a prerequisite for selling serious AI products. That raises the stakes for gross margins and operating leverage. If AI adoption inside Snowflake’s platform matches expectations, those fixed commitments support a premium position and make it harder for smaller rivals to compete. If usage lags, however, pre‑agreed spend on cloud AI accelerators and Graviton CPUs could squeeze profitability and limit Snowflake’s room to pivot.
Cloud Alliances Tighten as Enterprise AI Scales
For AWS, Snowflake’s decision strengthens its role as the default home for enterprise AI data platforms. AWS reported 28% revenue growth in the first quarter of 2026, and Snowflake has surpassed USD 7 billion (approx. RM32.2 billion) in lifetime AWS Marketplace sales, with more than USD 2 billion (approx. RM9.2 billion) in a single calendar year. Those figures show a flywheel where AWS is both infrastructure provider and distribution channel for Snowflake’s AI services. As data platforms chase enterprise AI workloads, these kinds of multi‑year, custom‑chip‑aligned deals are likely to become the norm. Vendors that once prided themselves on cloud neutrality will find it harder to stay loosely coupled when AI capacity planning demands firm, long‑dated commitments. Snowflake’s move signals that the next stage of enterprise AI will be shaped as much by chip roadmaps and cloud alliances as by software features.