Snowflake plans to invest $6 billion over the next five years in Amazon’s custom Graviton CPUs and AI accelerators. The partnership is designed to streamline the connection between Snowflake’s customer data and the expanding portfolio of AI services hosted on AWS.
“We are making it easier for enterprises to bring AI directly to governed data, so they can move faster, operate with greater density and create measurable impact at scale,” said Snowflake CEO Sridhar Ramaswamy.
Snowflake has been an AWS customer since 2011, gradually shifting its compute workloads from Intel and AMD processors to Amazon’s Arm‑based Graviton instances. The latest generation of Graviton processors features 192 Arm Neoverse V3 cores and supports up to 12 memory channels with speeds of 8,800 MT/s.
While GPUs continue to power model training, many AI‑related tasks—such as SQL queries or Python scripts—still rely on CPUs. This renewed focus on CPU performance is driving demand for high‑core‑count processors. Under the new agreement, Snowflake will train and run its generative AI models using a mix of AWS GPUs and Graviton CPUs. Its Cortex AI platform, for example, can translate natural language into SQL queries, summarize datasets, and perform sentiment analysis.
Amazon reports that Snowflake’s cumulative sales on the AWS Marketplace have surpassed $7 billion, with more than $2 billion generated in 2025 alone. The company appears confident that AI‑driven tools will generate sufficient revenue to justify the $1.2 billion annual spend on additional infrastructure. Wall Street shares the optimism, as Snowflake’s shares jumped over 30 % in after‑hours trading following the announcement.
Snowflake is not alone in expanding its use of Graviton. Earlier this year, Meta announced a multi‑year plan to deploy tens of millions of Graviton 5 CPU cores, positioning itself as one of AWS’s largest consumers of the home‑grown silicon. While Meta’s investment is also aimed at supporting AI workloads, the company may later transition to Arm’s next‑generation AI‑optimized processors.
