Sridhar Ramaswamy, CEO of Snowflake, poses in entrance of the corporate signage outdoors of the New York Inventory Trade on Sept. 30, 2025.
NYSE
Amazon mentioned Wednesday that its cloud division has landed a $6 billion spending dedication from Snowflake, which incorporates using the corporate’s customized silicon and chips for synthetic intelligence.
Snowflake’s buy of companies and expertise from Amazon Net Providers will happen over 5 years, in response to a press launch concerning the settlement. Snowflake intends to increase its use of Amazon’s Graviton general-purpose chips, in addition to cloud-based graphics processing models for AI.
It is the most recent signal that AWS is gaining momentum in AI as purchasers flip to the market-leading cloud for extra superior applied sciences. In April, Claude creator Anthropic mentioned it goals to spend over $100 billion on AWS over a decade. Amazon additionally has a take care of OpenAI.
Each of its agreements with the AI mannequin corporations embrace an fairness funding, whereas the Snowflake deal doesn’t. Snowflake, which went public in 2020, has a market cap of simply over $60 billion, and has lengthy relied on AWS.
Snowflake rose as a lot as 33% in prolonged buying and selling after it introduced robust outcomes for its fiscal first quarter, which ended on April 30. The corporate reported 39 cents in adjusted earnings per share on $1.39 billion in income, up 33% 12 months over 12 months. Analysts polled by LSEG had anticipated 32 cents per share and $1.32 billion in income.
Steerage was additionally robust, as Snowflake known as for a 12.5% fiscal second-quarter adjusted working margin on $1.415 billion to $1.420 billion in product income. Analysts surveyed by StreetAccount had been searching for a 11.9% margin, with $1.37 billion in product income.
Snowflake additionally mentioned it was buying AI startup Natoma for an undisclosed sum.
On the time of Snowflake’s IPO, it disclosed an amended take care of an unnamed cloud supplier for $1.2 billion in spending over 5 years, with $350 million coming within the remaining 12 months. The supplier was Amazon, a Snowflake spokesperson instructed CNBC. In 2023, the settlement climbed to $2.5 billion.
The brand new $6 billion association implies a median annual spend of $1.2 billion.
AWS introduced its first Arm-based Graviton chip in 2018, and it stays the corporate’s most profitable customized chip so far. Snowflake first talked about adopting Graviton in 2022.
Snowflake additionally has shut ties to Nvidia, following a partnership introduced in 2023. In November, Snowflake touted updates to simplify the method of working AI workloads on Nvidia GPUs.
The deal marks yet one more giant tech firm selecting customized Arm-based processors over these primarily based on conventional x86 structure.
For many years, server chips have been constructed on the x86 instruction units pioneered by Intel within the Nineteen Seventies and Superior Micro Gadgets a few a long time later. Arm’s different power-efficient structure went mainstream when Apple adopted it for the primary iPhone in 2007. However it was Amazon that introduced Arm chips into knowledge facilities with Graviton. Cloud rivals Google and Microsoft adopted Amazon in bringing out customized Arm chips.
Quick-forward to 2026, and central processing models like Graviton are seeing a renewed flurry of demand as mass AI adoption shifts from call-and-answer chatbots to task-oriented agentic apps.
Whereas GPUs like Nvidia‘s excel at coaching AI fashions as a result of they’ve 1000’s of tiny cores narrowly targeted on performing many operations concurrently, CPUs have a smaller variety of highly effective cores working sequential general-purpose duties. Agentic AI requires quite a lot of common compute energy to maneuver giant quantities of knowledge round for AI workflows, orchestrating throughout a number of brokers.
In April, Meta mentioned it could draw on a whole lot of 1000’s of Graviton chips.
“Graviton is our industry-leading CPU chip, which permits Meta to run the CPU-intensive workloads behind Agentic AI with the efficiency and effectivity they want at their scale,” Amazon CEO Andy Jassy mentioned on the corporate’s April earnings name.
WATCH: How Amazon chips are powering Anthropic with out Nvidia
