Defining Snowflake’s $6B AWS Bet
Snowflake’s USD 6 billion (approx. RM27.6 billion) cloud infrastructure investment with Amazon Web Services is a multi‑year agreement through which the data platform commits to buy large amounts of Graviton CPU compute and GPU‑accelerated AI capacity, signalling how artificial intelligence demand is pushing software vendors into long‑term, capital‑intensive partnerships with hyperscale cloud providers. This Snowflake AWS partnership centers on AWS’s Arm‑based Graviton processors for general cloud infrastructure and GPU‑accelerated EC2 instances for AI model training and inference. AWS said the agreement is Snowflake’s largest cloud spend commitment to date, and that Snowflake has surpassed USD 7 billion (approx. RM32.2 billion) in lifetime AWS Marketplace sales. The cloud infrastructure investment supports Cortex AI, Snowflake’s suite for text‑to‑SQL, summarization, sentiment analysis, and other AI workloads that run close to governed customer data, making AI compute capacity a core part of its go‑to‑market story.
From Asset-Light Software to Capital-Heavy Cloud Infrastructure
Snowflake’s move highlights a wider shift in software vendor spending: AI infrastructure is moving onto software balance sheets. In the older model, cloud providers carried most hardware risk while software firms kept their operations asset‑light. Now, securing AI compute capacity requires long commitments that look more like capital projects than pay‑as‑you‑go experiments. According to Startup Fortune, Snowflake’s USD 6 billion (approx. RM27.6 billion) AWS agreement shows how far data platforms will go to lock in compute for “agentic” AI systems that plan tasks, retrieve data, and call tools across applications. Those systems consume both accelerator and CPU resources at large scale. The economics change as gross margin, operating leverage, and usage risk become tightly linked to cloud infrastructure investment. If AI demand grows as expected, Snowflake protects its position; if it slows, the firm must absorb sizable procurement obligations.
Competing on Custom Chips, Performance and Cost
A key dimension of the Snowflake AWS partnership is the emphasis on custom silicon and workload‑specific infrastructure. Snowflake is shifting more compute to AWS Graviton CPUs, which in their latest generation pack 192 Arm Neoverse V3 cores with high‑bandwidth memory channels, and pairing them with GPU‑accelerated instances for AI. The strategy is to run traditional data warehousing and control-plane tasks on cost‑efficient Graviton instances, freeing budget and capacity for expensive AI training and inference workloads. This pattern reflects a broader cloud infrastructure investment trend in which software vendors differentiate on performance and cost efficiency, not only on features. The models run on GPUs, but the orchestration layer—SQL queries, Python functions, and agent logic—leans heavily on CPUs, making CPU density and price‑performance critical factors. Vendors that optimize this mix can offer lower costs and faster AI experiences without owning physical data centers.
AI Workloads, Cloud Alliances and Lock-In
Snowflake’s AI strategy, under CEO Sridhar Ramaswamy, is to recast the company as “the platform for the AI era,” centered on Cortex AI and agentic enterprise workloads. To deliver that at scale, Snowflake is expanding its AWS footprint into more regions and deepening its joint go‑to‑market through AWS Marketplace, where it exceeded USD 2 billion (approx. RM9.2 billion) in sales in a single calendar year. These steps point to tighter cloud alliances in which the cloud provider is both infrastructure backbone and distribution channel. At the same time, they increase vendor lock‑in: Snowflake gains priority access to AWS AI compute capacity, while AWS embeds itself further into Snowflake’s revenue engine. Combined with Snowflake’s ongoing acquisition streak since June 2025, the commitment suggests a consolidation strategy aimed at owning more of the AI data stack while tying its future closely to a single hyperscale partner.