Snowflake just signed a $6 billion, five-year deal with Amazon Web Services, the two companies announced Wednesday. According to TechCrunch AI, this isn’t a routine cloud renewal. The number tells the story: Snowflake has sold roughly $7 billion worth of its services through the AWS Marketplace since it launched in 2012. So this single contract is close to everything Snowflake has ever earned through that channel, packed into the next five years.
What stands out here is what’s driving it. AI, and more specifically, chips.
The numbers behind the deal
Snowflake has always run on AWS, even though it’s now also available on Microsoft Azure and Google Cloud. But customer spending on AWS is accelerating fast. Snowflake says that spend doubled in 2025 to $2 billion for the calendar year alone.
The engine is Cortex AI, Snowflake’s AI building tool that’s been around for a couple of years. It makes sense given where Snowflake sits. Most of a company’s data already lives there, so the tool can offer things like a plain-language interface for database queries, automatic summary reports, and similar features. Ask a question in normal English, get an answer from your own data.
Why the chip detail matters
Here’s the part worth slowing down on. Snowflake is signing this contract for more access to Graviton, AWS’s home-grown ARM-based CPU chip.
Not a GPU. A CPU. And that distinction is the whole point.
- GPUs handle the heavy lifting: training models and reasoning.
- CPUs handle most of everything else, especially the work that AI agents generate.
As AI shifts from training to daily use to automated agents running tasks on their own, CPU demand skyrockets. Amazon CEO Andy Jassy claimed last month that Amazon’s own AI chips deliver “better price-performance” than Nvidia’s, per TechCrunch AI. AWS still uses Nvidia chips in its cloud, but its own silicon is cheaper to deploy. Amazon, true to form, says it passes those savings to customers. And those savings are now pulling in multi-billion-dollar contracts.
This is a pattern, not a one-off
The Snowflake deal lands right after another big one. Last month, AWS agreed to supply Meta with millions of Graviton chips for its AI compute needs. That win stung a little extra, because Meta had signed a $10 billion deal with Google Cloud just months earlier.
The message to Nvidia is getting hard to ignore. The cloud giants are building their own CPUs and going after Nvidia’s territory:
- Google has made its own AI chips for years.
- Microsoft launched its Maia AI chip in January.
- AWS keeps stacking Graviton deals.
Nvidia isn’t flinching. CEO Jensen Huang said last week he’s ready to defend and grow his turf. His new AI-specific CPU, Vera, represents what he called a “brand new” $200 billion market, and he claims he’s already sold $20 billion worth. That came right after another record quarter.
Why this matters
This is significant because it shows where the AI money is actually flowing. Everyone watches GPU headlines and Nvidia’s quarterly records. But the next leg of AI spending, the agent era, runs heavily on CPUs. That’s a market the cloud providers think they can win with their own silicon, and deals like Snowflake’s and Meta’s are the proof points.
For practitioners, a few things to keep an eye on:
- Cheaper compute. If AWS keeps passing Graviton savings along, running AI workloads on ARM-based CPUs gets more attractive, especially for agent-heavy and inference-heavy apps.
- Architecture choices. Most AI tools are still built specifically for Nvidia chips. Expect more pressure to make workloads portable across ARM-based options.
- Vendor leverage. As cloud providers build their own chips, the Nvidia-or-nothing assumption loosens. That changes pricing conversations.
Nvidia isn’t surrendering market share anytime soon. But these multi-billion-dollar cloud contracts make one thing clear: whoever wins the AI race, the cloud providers are collecting their cut along the way. Watch the CPU side of this story closely. It’s where the next round of big deals is coming from.
Full details are available at the original report from TechCrunch AI.