Snowflake commits $6B to AWS Graviton chips as cloud giants challenge Nvidia's dominance
Snowflake's multi-billion dollar investment in Amazon's custom processors signals accelerating enterprise demand for alternatives to Nvidia in AI infrastructure.
Last verified:
Snowflake doubles down on AWS infrastructure
Snowflake and Amazon Web Services announced a long-term procurement pact on May 27 totaling $6 billion over five years, with a strategic focus on AWS’s proprietary Graviton processor family. According to TechCrunch AI, this commitment is noteworthy because Snowflake’s cumulative AWS Marketplace revenue since 2012 totals approximately $7 billion—meaning the new contract alone nearly matches the company’s entire historical spending with the cloud provider.
The timing reflects accelerating cloud adoption within Snowflake’s customer base. Enterprise clients’ AWS spending through Snowflake doubled during the 2025 calendar year to reach $2 billion in annual run rate, signaling that the data platform has become a central hub for AI workload orchestration rather than just a storage layer.
CPU demand surge reshapes chip procurement
The shift toward custom silicon hinges on a fundamental architectural reality: GPU chips excel at training and inference, but production AI systems—particularly autonomous agents—depend heavily on general-purpose processors for orchestration, data marshaling, and inference serving.
According to TechCrunch AI, Amazon CEO Andy Jassy publicly asserted last month that Amazon’s internally designed chips deliver superior cost-efficiency relative to Nvidia’s solutions, though AWS continues to operate Nvidia GPUs across its portfolio. By standardizing on Graviton for CPU-bound workloads, Snowflake gains access to lower unit costs that AWS passes along to customers, effectively subsidizing Snowflake’s own margin expansion or competitive pricing.
Graviton gains traction against entrenched incumbents
Snowflake’s commitment joins a pattern of large-scale commitments to non-Nvidia silicon. TechCrunch AI reports that AWS closed a separate deal to supply millions of Graviton units to Meta—a notable acquisition given Meta’s earlier $10 billion contract with Google Cloud, indicating that cloud providers view custom silicon as a lever for customer retention.
Nvidia has not ceded ground passively. CEO Jensen Huang announced the company’s new Vera AI-specific CPU, framing the addressable market at $200 billion and claiming $20 billion in pre-sales—suggesting Nvidia is extending its franchise beyond GPU-centric workloads into CPU competition. Google operates its own AI accelerators (TPUs), and Microsoft launched its Maia processor in January, widening the competitive landscape.
Why This Matters
This deal signals that cloud-native AI infrastructure is fragmenting away from monolithic GPU dependence. For Snowflake customers building production AI systems—especially those using Cortex AI for real-time analytics and agentic automation—the cost differential between Nvidia and Graviton-based deployments directly affects ROI on AI projects.
Enterprise IT teams now face a genuine architectural choice: lock into Nvidia’s ecosystem for unified GPU/CPU performance, or distribute workloads across cloud providers’ custom silicon to capture cost arbitrage. Snowflake’s vote of confidence in Graviton legitimizes this multi-vendor strategy and pressures Nvidia to defend pricing or risk losing the “peripheral compute” segment to lower-cost alternatives. If this pattern holds across major platforms, Nvidia’s dominance may consolidate around training and complex inference, while cloud providers capture the higher-volume production and serving tiers.
Frequently Asked Questions
Why is Snowflake committing so much money to AWS chips instead of just using Nvidia?
Nvidia dominates GPU training, but as AI shifts to inference and autonomous agents, CPU workloads explode. AWS's Graviton chips offer lower costs for these tasks and integrate directly with Snowflake's data infrastructure.
Is this deal a signal that Nvidia is losing market share?
Not necessarily — Nvidia's new Vera AI-specific CPU and $20B in pre-sales suggest aggressive competition is intensifying rather than shifting control. Cloud providers are building alternatives, but Nvidia still dominates GPU compute.
How does this affect Snowflake's business model?
Snowflake's Cortex AI tool—which runs SQL queries, generates summaries, and powers autonomous agents—benefits from cheaper CPU cycles. Lower infrastructure costs could improve margins or allow more aggressive pricing to customers.