Summary: Snowflake and AWS picked May 27, 2026 to unveil a five-year, $6 billion infrastructure agreement centered on wider deployment of AWS’s in-house, ARM-based Graviton processors and its AI infrastructure. Customer spending through AWS doubled in 2025 to $2 billion, with AI as the engine. The same day carried Snowflake’s first-quarter earnings: a raised full-year outlook and a 36 percent surge in extended trading.
Snowflake and AWS picked May 27, 2026 to unveil a five-year, $6 billion infrastructure agreement centered on wider deployment of AWS’s in-house, ARM-based Graviton processors and its AI infrastructure. The same day carried Snowflake’s first-quarter earnings: a raised full-year outlook and a 36 percent surge in extended trading. The detail deserving a second look is which chip takes the headline — not a GPU. As AI work moves out of model training and into everyday inference and agent automation, the CPUs inside cloud platforms have quietly turned into a contested battleground of their own.
Breaking Down the $6 Billion
TechCrunch’s account splits the agreement into three components: product integrations reaching deeper into generative and agentic AI, a wider joint-selling motion through the AWS Marketplace, and workload migration support to carry businesses “from experimenting with AI projects to using them routinely.”
How far the relationship already extends shows up in the arithmetic. In the fourteen years since Snowflake’s 2012 founding, it has moved roughly $7 billion in services through the AWS Marketplace — a single contract now comes close to that entire total. Customer spending through AWS doubled in 2025 to $2 billion, with AI as the engine. The company’s Cortex AI stack (natural-language database queries, summary reports) has been available for a couple of years, and on the earnings call management named Cortex Code and Snowpark as the quarter’s growth engines.
Why the CPU, Not the GPU, Takes Top Billing
Training and large-scale reasoning stay in GPU territory. The moment AI becomes daily software, however — agents juggling API calls, data preparation, scheduling, state management — those jobs fall to CPUs, and the load climbs with the number of agents in flight. Last month, Amazon CEO Andy Jassy argued that his company’s own chips beat Nvidia’s on price-performance, and in the same stretch AWS cut a separate agreement to supply millions of Graviton chips to Meta — striking, given that Meta had sealed a $10 billion Google Cloud deal only months earlier. Microsoft, meanwhile, brought out its own Maia AI chip in January.
Nvidia CEO Jensen Huang answered last week, framing Vera — the company’s new AI-focused CPU — as a “brand new” $200 billion market where a record quarter has already moved $20 billion in sales (we analyzed Vera in depth). Line the moves up and the pattern is unusual: cloud giants press homegrown CPUs to claim a bigger share of the AI infrastructure stack, while the incumbent GPU leader answers by striking at those vendors’ CPU turf. Seldom does the CPU sit at the center of a collision like this.
Earnings, Not the Deal, Did the Heavy Lifting
The agreement set the table; the financials fed the market. First-quarter revenue arrived at $1.39 billion, ahead of the $1.32 billion analysts had modeled. The FY2027 product revenue outlook moved from $5.66 billion to $5.84 billion, and second-quarter product revenue was guided to $1.415–1.42 billion, clear of the LSEG consensus at $1.37 billion. “Based on a combination of strength in our core data platform business and meaningful uplift from AI capabilities,” offered CEO Sridhar Ramaswamy by way of explanation. Gil Luria of D.A. Davidson held that the AWS pact “adds another element to the growth path” for Snowflake and sets it up to play “an even bigger role in their customers’ transition to AI,” while drawing it closer to its largest partner.
The Takeaway for Developers and Platform Teams
Three things matter in practice. One: homegrown cloud CPUs are now a default tier for AI-adjacent workloads, so cross-architecture compatibility (amd64/arm64) belongs in your CI pipeline today — before the first migration incident, not after it. Two: the major data platforms are tying customers in with multi-year infrastructure commitments. Build on Snowflake and the next five years will pull you toward AWS’s AI services, with exit costs climbing the entire way — put that in your vendor strategy now. Three: when you cost out agent workloads, GPUs are only part of the bill. Whether features like Cortex Code pencil out depends largely on CPU price-performance, and that is precisely what this deal wagers on. AI’s compute story runs broader than the GPU shortage narrative, and the least glamorous workloads go to whichever capacity is cheapest.