Pure Storage has established itself as a leader in the enterprise storage market, partly due to its flash-only approach to storage, which emphasizes a simple deployment and management experience. Now the company has entered the modern AI storage market with the launch of FlashBlade//EXA, a data storage platform designed to address the unique needs of AI and large-scale high-performance computing deployments.
With the AI market so hot and specialized storage players driving a lot of hype, it is worth looking at how Pure’s approach to solving the needs of “AI factories” across large enterprise IT organizations can compete, and how the company can be successful in what is undoubtedly a big leap up the storage stack.
Moving From Enterprise To AI Is A Different World
Pure has done a stellar job of delivering enterprise-class storage solutions to the market. Its approach to delivering flash-only was not some unique gimmick or marketing ploy; it enabled better performance and reliability at a compelling TCO. And the market responded. Pure’s FlashArray and FlashBlade products now have more than 13,000 customers, and the company continues to see double-digit growth quarter after quarter.
However, the value proposition is bigger than simply selling flash into the enterprise. Anybody who has spent time in an IT organization knows how messy and complex managing a large-scale storage environment can be (which drives up costs). So, in addition to building out highly performant products, Pure has also emphasized removing the complexity from the storage management equation. Customers can consume its products however they want and manage them universally with the help of a copilot.
So, enterprise storage needs to be performant, flexible, cost-effective and simple. Check, check, check, check.
As Pure focused on establishing its presence in enterprise storage, the AI market (really HPC and AI) has become white-hot, and companies like VAST Data have leaned far more heavily into the complex data management that is required to make an underlying storage architecture perform at scale. In AI, there is much more to driving performant storage than simply adding more servers with more flash. In AI, many different data types are required to feed a training model and keep very expensive GPUs processing efficiently. The only way to do that is to effectively categorize, contextualize, cleanse and tag through metadata. However, when managing metadata, this time-consuming process strains traditional storage environments, preventing them from adequately feeding data to GPU clusters that are performing training.
Perhaps the most notable company to address this — certainly the first — was VAST with its Data Management Platform, a disaggregated, share everything architecture that separates compute from storage and that has what amounts to an AI operating system to manage data more effectively and drive the absolute best performance. VAST rightly recognized the importance of data management as the performance bottleneck in AI. Actually, it’s fair to say that all storage vendors recognized this, but VAST was early in delivering. Today, most storage vendors have solutions that compete in the AI data market — all of them making varying claims about performance, TCO and the like.
This brings us to Pure Storage and the FlashBlade//EXA. This product represents Pure’s expansion from enterprise storage into AI data storage.
FlashBlade//EXA — Metadata Management For Hyperscalers And Enterprise
There are two elements to FlashBlade//EXA — hardware and software. From a hardware perspective, the AI environment is disaggregated, separating metadata from data. In the Metadata Core (metadata cluster), servers are tasked with managing the unique requirements of this function, where non-linear, non-sequential access to data is the norm for functions such as metadata indexing, data lineage tracking, semantic discovery and query optimization — which are critical to performance for AI.
Metadata Core servers connect directly with Data Nodes, where enterprise data is stored. This data can be files, documents, images, audio, video and the like. There are also a range of confidentiality, privacy and relevancy classifications. While the servers in these Data Nodes clusters will store and serve a lot of data to be trained and inferenced, it is, in many cases, just data without any real context.
The above graphic shows that the data and metadata paths connect to compute clusters over separate, dedicated data paths. Powering the Metadata Core is Pure’s Purity//FB software, with pNFS (i.e., parallel NFS) to enable parallelized access to a distributed transactional database where metadata is stored and quickly accessed.
Interestingly, the Data Nodes in Pure’s solution are commodity, third-party servers certified by Pure. This is interesting because it marks the first time Pure has supported third-party infrastructure for its storage solution. While the company says it plans to incorporate its DirectFlash Module technology in the future, customers today will have to work with a third party to support it. This move somewhat undercuts the value the company places on its DFM technology, as it seems to view data nodes as fully commoditized in the AI equation. Quite frankly, perhaps it is commoditized in the AI equation. Or perhaps the delivery of metadata management in a disaggregated way is such a performance uplift that DFM at the data level is more of an incremental benefit in the bigger picture. I’ll be curious to see performance comparisons of clusters of Pure-based data nodes versus clusters built on third-party data nodes.
How Does Pure Stack Up In The AI Data Market?
Taking a step back from Pure and AI for a moment, let’s talk about technology adoption curves. In the early stages of any technology adoption curve, solutions tend to be more complicated in terms of deployment and configuration. Because of this, capabilities and features tend to appear to be deeper.
As technology evolves and matures, however, much of the “knob turning” required to deploy and optimize tends to be abstracted from the user experience. Pure Storage delivers this with FlashBlade//EXA — an AI data storage platform rich with capabilities and a big step forward in simplicity.
Companies like VAST have done a good job going after the largest-scale AI (and HPC) markets and have several hyperscale and large labs and data environments for their innovation. Now the market is starting to see enterprise AI — not just hyperscaler AI — enter the conversation. In fact, HPE CEO Antonio Neri talked about the uptick in enterprise AI his company is seeing in his most recent earnings call. Likewise, Dell and Lenovo are seeing AI backlogs that are coming down-segment from hyperscalers to include the enterprise.
This is where Pure comes into play — and why its work in developing such a strong enterprise presence is important. Pure understands enterprise storage administrators and the overall enterprise data landscape. It has built a product, solution and selling motion that has been very successful in this market. Most importantly, it has been a leader in the notion of stripping away complexity without sacrificing performance or capability.
If Pure can do for enterprise AI what it has done for enterprise storage, it will be well positioned. The key is to first go after existing customers with an eye on expanding the Pure footprint. Win some deals, learn some deployment and integration lessons and look to expand from there.
The enterprise AI market is incredibly nascent and wide open for AI solutions — and for the storage that supports those solutions. While there are some frontrunners, Pure should be able to leverage its standing in enterprise IT to build a presence for these enterprise AI projects that are beginning to kick off.
Of course, this is a long game. I’ll be watching for early signs of Pure’s progress with FlashBlade//EXA, such as wins and partnerships, and I’ll be sure to share my findings.
Moor Insights & Strategy provides or has provided paid services to technology companies, like all tech industry research and analyst firms. These services include research, analysis, advising, consulting, benchmarking, acquisition matchmaking and video and speaking sponsorships. Of the companies mentioned in this article, Moor Insights & Strategy currently has (or has had) a paid business relationship with Pure Storage and VAST Data.
