.svg)
Executive Summary
A leading North American automobile manufacturer is driving the future of autonomous vehicles, leveraging AI to continuously improve driving capabilities across a global fleet. The challenge: how to store, manage, and deliver exabyte-scale training data to GPU clusters at the speed they demand, while ingesting petabytes of new data every day from hundreds of thousands of vehicles worldwide.
A core element of their strategy is continuous AI model improvement. Vehicles collect sensor, telemetry, and video data that feeds directly into an on-premises data lake, powering model training, simulation validation, factory vision inspection, and advanced analytics. This requires an AI data infrastructure capable of handling exabytes of capacity today while scaling to meet aggressive growth trajectories.
Environment: AI at automotive scale
The manufacturer operates one of the largest private cloud AI data infrastructure deployments in the world, spanning multiple data centers across the United States, Europe, Australia, and China. Over 100,000 vehicles upload autopilot data directly into the storage infrastructure, generating more than 6 petabytes of new data every day. This data powers AI model training for autonomous driving, vehicle simulation and replay validation, factory vision inspection using computer vision, and analytics workloads. The infrastructure supports sustained throughput exceeding 2 TiB/sec and handles over 600,000 S3 requests per second in individual clusters.
"By bringing our AI data on-premises with AIStor, we eliminated the unpredictable cloud costs that were making our program unsustainable. Now we have predictable economics at exabyte scale, on hardware we control, across every region we operate in."
Challenge: Outgrowing existing infrastructure at exabyte scale
The manufacturer's existing storage infrastructure was fragmented across multiple clusters from a prior vendor, none of which could individually scale to the throughput and capacity demands of the AI workloads. As data volumes grew to petabytes per day, the limitations became acute. GPU clusters were being starved for data, idling expensive compute resources. Adding capacity meant standing up yet another isolated cluster, increasing operational complexity without solving the fundamental scalability challenge.
Simultaneously, public cloud costs for a portion of the data estate were economically unsustainable, with storage, API, and egress fees compounding at the volumes being generated. The manufacturer needed a single, unified storage platform that could deliver both throughput and IOPS for GPU-intensive AI workloads, run on commodity hardware to keep costs viable at exabyte scale, and scale linearly from hundreds to thousands of nodes without re-architecture.
"We chose MinIO over alternatives because of its scalability and performance. When your AI workloads demand exabyte-scale throughput on commodity hardware, there is no substitute."
Outcomes: Exascale AI infrastructure on a unified platform
AIStor now powers over 1 exbibyte of usable capacity across thousands of nodes in multiple global data centers. This includes single clusters exceeding 1,000 nodes, making it one of the largest private cloud AI data infrastructure deployments in the world. The platform delivers 2.2 TiB/sec of overall throughput, ensuring GPU clusters are never waiting on data.
The manufacturer ingests more than 6 petabytes of new data every day from its global vehicle fleet, with AIStor handling over 600,000 S3 requests per second in individual clusters. AIStor's single-binary deployment model enabled 300+ node clusters to be stood up in a single weekend, with the platform scaling to 1,000+ node mega-clusters as the data footprint grew.
By repatriating data from public cloud to on-premises AIStor infrastructure, the manufacturer achieved millions of dollars in annual savings, eliminating unpredictable storage, API, and egress fees. The manufacturer is also consolidating from over a dozen fragmented clusters from a prior vendor into a unified AIStor platform, reducing operational complexity while gaining consistent S3 compatibility across all workloads.
The platform supports multiple concurrent workloads, including autonomous driving model training, vehicle simulation and replay validation, factory vision inspection, and ClickHouse-backed analytics, all on a single, globally consistent infrastructure. Operations span the United States, Europe, Australia, and China, with a 20%+ annual data growth trajectory fully supported.
Solution: AIStor-powered private cloud AI data lake
The manufacturer selected MinIO AIStor as the high-performance, S3-compatible object storage foundation for its AI data infrastructure. AIStor was chosen over alternatives because of its superior scalability and performance on commodity hardware, including configurations using enterprise NVMe drives, high-core-count AMD processors, and 100Gb networking.
AIStor is software-defined, enabling compute and storage to be decoupled for independent scaling. This was a key factor in the rapid deployment cadence, with AIStor's single-binary architecture enabling 300+ node clusters to go from zero to production in a single weekend. The manufacturer runs AIStor on hardware from multiple vendors, including HPE, Dell, and Supermicro, with consistent performance and operations across all platforms.
Vehicles upload data directly into AIStor via pre-signed URLs, enabling a streamlined car-to-storage pipeline without intermediate staging. The architecture integrates with secure traffic management to route fleet data from worldwide collection points into centralized data lakes, with trained AI models delivered back to vehicle systems. This end-to-end pipeline operates entirely on-premises, providing full data sovereignty over the manufacturer's most sensitive intellectual property.
"AIStor gave us the ability to go from zero to hundreds of nodes in a single weekend. That speed of deployment was critical, and we have since expanded to clusters exceeding a thousand nodes on the same platform."
.svg)
