Snowflake-AWS chip deal
Last reviewed
Jun 2, 2026
Sources
11 citations
Review status
Source-backed
Revision
v1 · 1,499 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Jun 2, 2026
Sources
11 citations
Review status
Source-backed
Revision
v1 · 1,499 words
Add missing citations, update stale details, or suggest a clearer explanation.
The Snowflake-AWS chip deal is a five-year, roughly $6 billion infrastructure commitment that the data-cloud company Snowflake made to Amazon Web Services (AWS), announced on May 27, 2026. Under the agreement, Snowflake will spend the funds on AWS compute, with the company's custom Arm-based Graviton processors at the center of the arrangement, alongside GPU-accelerated cloud instances used for AI model training and inference. It is Snowflake's largest infrastructure commitment to date, roughly five times the $1.2 billion it pledged at its 2020 initial public offering, and it became a prominent early example of an enterprise software vendor (rather than a foundation-model lab) anchoring a multibillion-dollar deal around custom CPUs rather than GPUs.[1][2][3]
The commitment was disclosed alongside Snowflake's first-quarter fiscal 2027 earnings, and the combination of an earnings beat and the AWS announcement sent Snowflake shares up about 36 percent the following trading day.[2][4]
Snowflake operates a cloud-based data platform that runs on top of the major public clouds, with AWS as its largest underlying provider. Most Snowflake customers already host their data on AWS, and the two companies have a long commercial relationship spanning data warehousing, the AWS Marketplace, and, more recently, AI services. The 2026 agreement deepens that relationship by tying a specific, large dollar commitment to AWS infrastructure over a defined five-year term.[1][3]
The deal's defining feature is its emphasis on AWS Graviton, Amazon's in-house line of Arm-based server CPUs designed as a lower-cost, more power-efficient alternative to x86 processors from Intel and AMD. Snowflake framed the expanded Graviton usage as price-performance optimization for the data-warehousing and AI workloads that run on its platform, while GPU-accelerated Amazon EC2 instances would handle the more compute-intensive work of AI model training and inference.[1][5]
Snowflake positioned the spending as support for what it calls the "agentic enterprise," in which AI systems act on a company's governed internal data rather than simply answering one-off questions. Its Cortex AI suite, which provides capabilities such as text-to-SQL, summarization, sentiment analysis, and entity extraction over governed data, was cited as the primary driver of accelerating customer demand on AWS.[1][2]
The publicly disclosed structure of the agreement is summarized below. Snowflake's chief executive, Sridhar Ramaswamy, indicated that the company would publish more technical detail at the AWS re:Invent conference later in 2026.[3]
| Element | Detail |
|---|---|
| Parties | Snowflake (customer) and Amazon Web Services (provider) |
| Announced | May 27, 2026 |
| Commitment | Approximately $6 billion |
| Duration | About five years |
| Primary silicon | AWS Graviton Arm-based CPUs |
| Secondary silicon | GPU-accelerated Amazon EC2 instances (AI training and inference) |
| Stated workloads | Data warehousing plus Cortex AI and agentic AI over governed data |
| Type | Multi-year strategic collaboration / compute commitment |
The $6 billion figure represents a contractual spending commitment for AWS compute and AI infrastructure over the term, not an equity investment or a one-time purchase. AWS chief executive Matt Garman said Snowflake's deepened commitment to run on Graviton would deliver "the world-class performance, flexibility, and cost savings customers need to run data warehousing and AI workloads at scale."[1]
The 2026 commitment continues a pattern of escalating Snowflake-AWS agreements. The progression of Snowflake's disclosed multi-year AWS spending commitments is shown below.[2][3]
| Commitment date | Amount | Approx. multiple of prior |
|---|---|---|
| 2020 (IPO) | $1.2 billion | baseline |
| 2023 (renewal) | $2.5 billion | ~2.1x |
| 2026 | ~$6 billion | ~2.4x (and ~5x the 2020 figure) |
Separately, Snowflake reported that customer spending transacted through the AWS Marketplace had reached roughly $7 billion in lifetime sales since the company's founding in 2012, with marketplace transactions exceeding $2 billion in calendar year 2025, more than doubling year over year.[1][2]
Snowflake's pivot toward AI has centered on letting customers run AI applications directly against data already stored and governed inside its platform, avoiding the need to move sensitive information into separate systems. Cortex AI is the centerpiece of that strategy, and Snowflake attributed much of its accelerating AWS consumption to Cortex adoption.[1][2]
The company's reported financial results reinforced the demand narrative. For the first quarter of fiscal 2027, Snowflake posted product revenue of about $1.33 billion, up roughly 34 percent year over year, and raised its full-year product revenue guidance to about $5.84 billion. Management presented the AWS commitment as infrastructure to support sustained, production-scale AI usage rather than experimental pilots.[4]
AWS Graviton is Amazon's family of custom CPUs based on the Arm instruction set, built to offer better price-performance and energy efficiency than comparable x86 chips. While much of the public attention on AI silicon has focused on GPUs for model training and on Amazon's own AI accelerators such as AWS Trainium and AWS Inferentia, CPUs remain essential for the broad set of supporting tasks around AI systems.[1][5]
That distinction is central to why the Snowflake deal foregrounds CPUs. As AI moves from training toward continuous, day-to-day use and toward automation through agentic AI, the volume of general-purpose CPU work tends to rise sharply: orchestration, data preparation, retrieval, and the routine background processing that agents perform between model calls. GPUs handle training and the heaviest reasoning, while CPUs such as Graviton handle much of everything else, often at lower cost.[2][5]
The Snowflake agreement was one of several large 2026 commitments organized around Amazon's custom silicon. In April 2026, Meta signed a deal for tens of millions of AWS Graviton CPU cores to power its own AI and agentic workloads, a volume that Amazon said made Meta one of the largest Graviton customers worldwide.[6][7] Around the same period, Anthropic expanded its strategic collaboration with Amazon into a commitment described as more than $100 billion in AWS technologies over the following decade, spanning Graviton along with successive generations of Trainium accelerators.[8][9]
Against those agreements, the Snowflake deal stood out as one of the largest commitments from a company that does not itself build frontier foundation models, signaling that demand for long-duration AI infrastructure was spreading from model developers to enterprise software vendors and their customers.[3]
Commentators treated the deal as a marker of enterprise AI maturing from short-lived experiments into persistent, foundational infrastructure with multi-year budgets attached. The size and length of the commitment suggested that Snowflake expected agentic and Cortex-driven workloads to generate durable, predictable demand rather than spiky, optional usage.[3][10]
The arrangement also reinforced the strategic value of custom CPUs in the AI buildout. By anchoring a marquee commitment to Graviton rather than to GPUs alone, Snowflake and AWS underscored that the economics of running AI at scale depend heavily on efficient general-purpose compute for the surrounding workload, not only on scarce accelerators. For Amazon, a string of large Graviton commitments from Snowflake, Meta, and others validated years of investment in its own chip designs and bolstered AWS's position against rival clouds.[5][6][10]
Investors responded strongly. Snowflake shares rose as much as roughly 40 percent intraday and closed up about 36 percent on May 28, 2026, the day after the announcement, lifted by both the earnings beat and the AWS deal.[4] Coverage from outlets including CNBC, TechCrunch, GeekWire, and The Information characterized the commitment as Snowflake's largest infrastructure pledge yet and as further evidence of Amazon's momentum in supplying compute for AI.[2][3][11]
Executives from both companies framed the agreement around the shift to autonomous AI. Ramaswamy said the industry was "moving into the era of the agentic enterprise, where AI systems don't just answer questions, but help organizations reason over trusted data," while Garman emphasized the cost and performance benefits of running those workloads on Graviton.[1]