Everpure Redefines AI Infrastructure with GPU-Based SLAs for FlashBlade//Exa
San Francisco, CA – March 16, 2026 – Everpure, formerly known as Pure Storage, is significantly altering the landscape of artificial intelligence (AI) and high-performance computing (HPC) infrastructure with the introduction of Evergreen One for AI. This new consumption model extends performance-backed guarantees to its highest-performing storage platform, FlashBlade//Exa, tying service-level agreements (SLAs) directly to the number of graphics processing units (GPUs) a customer deploys. The move signals a shift towards ensuring storage environments can consistently deliver the throughput required to maximize GPU utilization, a critical bottleneck in many AI workflows. Alongside this announcement, Everpure has released Datastream, a beta version of an automated AI pipeline appliance designed to streamline data preparation for AI models.
For organizations grappling with the escalating demands of AI and HPC, maintaining optimal performance across the entire infrastructure stack is paramount. GPUs are often the most expensive component, and their efficiency is directly impacted by the speed and reliability of data access. Evergreen One for AI addresses this challenge by guaranteeing a specific level of performance based on GPU count, effectively shifting the financial and operational risk away from the customer. Kaycee Lai, Everpure’s vice president for AI, explained that the offering sets performance levels based on GPU numbers, providing an SLA-backed performance guarantee. This represents a departure from previous flexible capacity offerings, where FlashBlade//Exa was not included in the Evergreen One consumption model.
FlashBlade//Exa: Designed for the Demands of Modern AI
FlashBlade//Exa was initially launched to tackle the unique storage requirements of next-generation, GPU-intensive AI and HPC workloads. Its architecture, introduced as a departure from previous Pure Storage designs, disaggregates metadata and bulk storage, utilizing different hardware and protocols for each. This allows for unparalleled scalability and performance, crucial for handling the massive datasets common in AI training and inference. According to Everpure, FlashBlade//Exa delivers read performance exceeding 10+ TB/s, minimizing GPU idle time and maximizing productivity. Write speeds are as well substantial, scaling up to 50% of read performance within a single namespace.
The platform’s capabilities are particularly relevant for organizations operating at scale, supporting thousands – even tens of thousands – of GPUs. Everpure emphasizes the platform’s ability to power “AI factories” and exabyte-scale workloads with efficiency. The company’s focus on simplifying data access is also key, aiming to eliminate manual tuning and complex client setups, providing “instant, frictionless, unlimited access to data” for modern AI and HPC environments. This is a critical need as AI models grow in complexity and data volumes continue to explode.
Streamlining the AI Pipeline with Datastream
Recognizing that data preparation is a significant hurdle for AI teams, Everpure has also unveiled Datastream, currently in beta. Lai highlighted that data teams often spend approximately 80% of their time preparing unstructured data for utilize in AI models – a statistic frequently cited within the industry. Datastream is designed to address this “data readiness” challenge by automating the retrieval-augmented generation (RAG) pipeline. This includes the critical steps of data ingest, curation, and vectorization, transforming raw data into a format suitable for AI consumption.
The appliance operates as a “single SKU” solution, integrating Nvidia GPUs with Everpure storage. This integrated approach aims to provide a simplified, “easy button” experience for enterprises building chatbots or autonomous agents. Datastream’s software capabilities were developed in-house, but the system is designed to connect to a variety of third-party data sources, including environments from Dell, HP, and NetApp, as well as cloud-based data repositories. This flexibility allows organizations to centralize their AI readiness efforts regardless of where their data resides. The system automates tasks like chunking, embedding, and indexing, ensuring data accuracy and relevance for AI agents.
Benchmarking Performance: Everpure Claims Leadership in AI Workloads
To underscore the performance capabilities of its hardware, Everpure has released new benchmark results. In MLPerf 2.0 testing, the company asserts it achieved the top spot for checkpointing – a vital function for saving the state of a model during lengthy training runs – with results up to two times better than competitors such as Huawei and Vast. Everpure also cited Spec Storage AI image benchmarks, where it reportedly outperformed NetApp’s AFX platform by approximately 20%. These benchmarks aim to validate the hardware’s ability to handle the demanding workloads associated with AI and HPC.
The introduction of Evergreen One for AI and Datastream reflects a broader trend in the storage industry towards offering more flexible and consumption-based models. As AI adoption accelerates, organizations are increasingly seeking solutions that can scale efficiently and deliver predictable performance. Everpure’s approach, with its focus on GPU-based SLAs, positions the company as a key player in enabling the next generation of AI innovation. The company’s transition from Pure Storage to Everpure, completed recently, signals a broader expansion of its business focus to encompass comprehensive data management solutions.
Key Takeaways
- GPU-Based SLAs: Everpure’s Evergreen One for AI ties storage performance directly to GPU count, guaranteeing throughput and minimizing bottlenecks.
- FlashBlade//Exa Expansion: The high-performance FlashBlade//Exa platform is now included in the Evergreen One consumption model, offering greater flexibility for AI and HPC workloads.
- Automated Data Pipeline: Datastream, a new appliance in beta, automates the RAG pipeline, reducing data preparation time for AI teams.
- Performance Benchmarks: Everpure claims leadership in key AI benchmarks, including MLPerf 2.0 checkpointing and Spec Storage AI image processing.
Everpure is expected to provide further updates on Datastream’s availability and performance metrics in the coming months. The company’s continued focus on innovation in AI infrastructure will be crucial as organizations strive to unlock the full potential of artificial intelligence. Readers interested in learning more about Everpure’s offerings are encouraged to visit the company’s website at pure.ai.
What are your thoughts on the evolving landscape of AI infrastructure? Share your comments below and let us know how these developments might impact your organization.
Related reading