Snowflake locked in a five-year, $6 billion agreement with Amazon Web Services to purchase custom AI chips, marking a major vote of confidence in AWS's silicon strategy and a significant blow to Nvidia's dominance in enterprise AI infrastructure.
The deal positions Trainium chips, AWS's custom-built processors designed for AI training workloads, as a core component of Snowflake's infrastructure. Snowflake, which generates roughly $3.1 billion in annual revenue and commands a major share of the cloud data warehouse market, will deploy these chips to power its AI and machine learning capabilities across its platform.
This agreement reflects a broader industry trend. Hyperscalers including Amazon, Google, and Microsoft are aggressively developing proprietary silicon to reduce dependency on Nvidia GPUs and cut costs. AWS launched Trainium in 2021, but adoption has been measured. Snowflake's commitment signals the chips have matured enough for production workloads at scale.
The timing matters. Nvidia maintains near-monopoly pricing power in AI chips, with its H100 and newer Blackwell GPUs commanding premium prices that squeeze enterprise margins. AWS Trainium chips offer cost advantages and integration benefits for customers already operating on AWS infrastructure. Snowflake, which runs on AWS, benefits from tighter optimization and lower opex.
For Snowflake, the deal locks in pricing predictability and supply security for a critical resource. AI features have become table stakes for data platforms. Snowflake launched Cortex, its AI assistant layer, in 2024 to compete with rivals embedding generative capabilities into their products. Custom chips funded by a major partner reduce the capital burden of scaling these features.
AWS gains a marquee customer validating its chip roadmap. Snowflake's scale and public visibility make this deal a confidence signal to other enterprises evaluating