Google Cloud Axion for Snowflake Gen2 Warehouses: Next-Generation Price Performance and Memory Bandwidth

Building on a year of expanding collaboration, Snowflake and Google Cloud have deepened our technical integration to incorporate Google Cloud’s new C4A virtual hardware (VMs) — powered by custom Google Axion processors — into Snowflake Generation 2 (Gen2) and Adaptive Warehouses (coming soon).

Google Axion is purpose built to accelerate modern, data-intensive workloads core to Snowflake’s customers, including AI inferencing, high-throughput data analytics and containerized applications. By leveraging Axion’s Arm-based efficiency and Titanium offload system, we anticipate a step change in performance across these critical tasks, reducing latency and saving customers money.

By combining Snowflake’s governed data foundation with Google’s custom silicon, we are delivering a significant leap in performance and cost efficiency for the AI era. Axion processors represent the latest in custom Google silicon, joining a lineage that includes Tensor Processing Units (TPUs) and Titanium offload processors, now available for the compute-intensive demands of the Snowflake AI Data Cloud.

Why the AI era demands a dynamic architecture

Enterprise workloads are evolving. As organizations move toward agentic AI and complex data engineering, the demand for raw compute power has shifted.

According to a recent analysis of over 600 million production queries — the majority of Snowflake queries are now compute-bound. This shift is driven by extreme data pruning (averaging 99.4% of data skipped) and high-performance cloud networking, which reduces the time spent waiting for disk.

However, a single query’s behavior often shifts during its execution — transitioning from a network-bound "scanning" phase to a CPU-bound "aggregating" phase. By leveraging Google Axion processors, Snowflake enables the improved performance to handle CPU-intensive demands while dynamically monitoring resource saturation to adapt on the fly.

Delivering up to 50% better performance and memory bandwidth

Snowflake’s architecture is designed to keep CPUs busy by asynchronously pre-fetching data into memory before it’s needed. As CPUs get faster, memory bandwidth and I/O throughput become the critical scaling factors.

“The collaboration between Snowflake and Google Cloud allows BlackLine to significantly enhance the responsiveness of our data platform. Transitioning our workloads to Snowflake Gen2 warehouses powered by Google Cloud Axion yielded a 30% improvement in query response times for BlackLine. Axion-powered Snowflake warehouses provide the high-performance infrastructure essential for our platform to stay ahead of the most demanding compute-bound workloads. This speed is critical for orchestrating complex end-to-end processes, from Invoice-to-Cash to Record-to-Report, and for powering the governed, auditable agentic AI that automates intricate financial processes for the Office of the CFO.” —Manishankar Rajagopal, Senior Director of Data and Analytics, BlackLine

When evaluating Axion for our customers, we focused on raw CPU speed, vector performance for data scanning and memory latency. Internal benchmarking, which measured query latency compared to internal benchmarks, and real-world customer replays reveal that the migration to Axion enables:

• Up to 50% performance improvement: Significant gains are made across full-stack benchmarks such as TPC-DS and TPC-H.

• Enhanced memory bandwidth: The combination of Axion’s hardware architecture and DDR5 memory provides up to a 50% improvement in memory bandwidth, directly translating to lower query latency for hash tables and Bloom filters.

• Optimized throughput: Improved local SSD performance and increased network bandwidth accelerate data transfer between nodes during distributed hash joins and ephemeral storage access.

For Gen2 warehouses running on Axion, this means core analytics, such as large table scans and complex joins, can complete in significantly less time, allowing customers to process more data per credit spent.

Performance meets sustainability

The partnership between Snowflake and Google Cloud isn’t just about speed; it’s about efficiency. Google Axion-powered instances support Snowflake customers’ efforts to reduce their environmental impact without sacrificing performance. These new processors consume less energy than comparable x86-based instances, helping organizations scale their AI initiatives while supporting their ESG goals.

Simple, secure and available globally

Because Google’s Axion infrastructure is integrated with the Snowflake platform, there is no complex architecture to rebuild. Your existing security rules, access policies and universal governance controls carry over automatically to every task you run with Snowflake on Google Cloud. Axion-powered instances are available globally where Snowflake operates in Google Cloud.

Instead of moving your data to the AI, we are bringing the industry’s advanced, custom-cooled infrastructure directly to your data.

Get started today

• Try for free: To experience these performance gains for your own organization, set up a free trial for Snowflake on Google Cloud today.

• Learn more: Read the Workload Insights from the Snowflake Data Cloud: What Do Production Analytic Queries Really Look Like? paper for a technical deep dive and Shaved Ice: Optimal Compute Resource Commitments for Dynamic Multi-Cloud Workloads about cloud hardware evolution.

• Join us at Google Cloud Next: Author Murray Stokely will be speaking at Google Cloud Next on April 23 at 8 a.m. PT, where he will share more insights during the Architecting High-Performance SaaS: Lessons from Top ISVs breakout session.

_{Forward-looking statements}

_{This article contains forward-looking statements, including about our future product offerings, and are not commitments to deliver any product offerings. Actual results and offerings may differ and are subject to known and unknown risk and uncertainties. See our latest 10-Q for more information.}