The Six Billion Dollar Power Play: Why Snowflake Just Went All In on Amazon’s Custom Chips

The artificial intelligence race is getting wildly expensive, and data companies are writing massive checks to keep up. Just this week, cloud data storage giant Snowflake signed a staggering five-year, $6 billion contract with Amazon Web Services. This is not just a routine vendor renewal. To understand how massive this agreement is, you only have to look at the history. Since Snowflake launched in 2012, it has generated about $7 billion in total sales through the AWS Marketplace. Now, they are committing almost that exact same amount in a single hardware deal.

Snowflake’s customers are spending money at a breakneck pace. In calendar year 2025 alone, their clients doubled their AWS spending to reach $2 billion. The driving force behind this massive surge is simple: artificial intelligence. Snowflake recently rolled out an AI building tool called Cortex AI. Because Snowflake acts as the primary data vault for thousands of major enterprises, building an AI tool directly into that vault makes perfect sense. Cortex allows everyday workers to query massive databases using regular conversational language instead of writing complex SQL code. They can generate summary reports and extract deep insights in seconds.

But powering those natural language queries takes serious hardware. Here is where the specific details of the Amazon deal get highly interesting. Snowflake is not just buying generic server space. They are specifically securing access to Graviton, Amazon’s custom-built, ARM-based CPU chip.

If you follow tech news, you hear a lot about graphics processing units, or GPUs, usually made by Nvidia. Those heavy-duty GPUs handle the intense work of training new models and doing complex reasoning. However, as the industry shifts from basic chat interfaces to autonomous agents that perform daily tasks, standard central processing units, or CPUs, take over the heavy lifting. Agent automation requires massive CPU capacity.

Amazon recognized this shift years ago and started building its own silicon. Last month, Amazon CEO Andy Jassy pointed out that their homegrown chips offer significantly better price-to-performance ratios than the standard Nvidia hardware. While AWS still buys and offers Nvidia chips because developers want them, the demand for affordable processing power is pushing Amazon to deploy its own custom chips as fast as they can manufacture them. Amazon saves money by cutting out the middleman, and they pass those lower operating costs directly to price-conscious clients.

This strategy is working perfectly, and it is pulling in multi-billion-dollar commitments. For example, just last month, Amazon signed a separate agreement to supply Meta with millions of Graviton chips for their own computing needs. That was a major victory for Amazon, especially since Meta had just signed a massive $10 billion contract with Google Cloud a few months prior.

These custom silicon deals serve as a giant warning siren for Nvidia. The massive cloud providers are actively trying to break free from their reliance on a single chipmaker. Google has spent years developing its own processors, and Microsoft recently launched its custom Maia AI chip.

Nvidia CEO Jensen Huang is well aware of this threat and plans to defend his territory. Last week, Huang announced that Nvidia is launching a brand new AI-specific CPU named Vera. He estimates this new CPU opens up a huge new market for his company, and he already booked $20 billion in early sales.

Nvidia will fight hard to keep its market share, but the underlying truth of the tech boom is becoming clear. Whether a company buys custom chips or sticks with Nvidia, the massive cloud providers like Amazon always win. As businesses integrate automated agents into every daily workflow, the cloud platforms supplying the physical hardware will continue pulling in historic windfalls.