Why flash storage may be taking off in the enterprise

Flash storage is what has been used for years in mobile devices like smartphones and laptops. Now, it could be at an inflexion point also in enterprise data storage, where hard disk drives (HDDs) still dominate.
In November, flash technology pioneer Pure Storage announced that one of the hyperscalers has decided to integrate Pure’s flash technology across their storage portfolio. Their plan, says Pure Storage CTO Rob Lee, is to over time replace all their hard drives, and even existing flash drives, and standardise on Pure’s tech.
HDDs, which first emerged in the 1970s, have spinning disks onto which data is written, and a mechanical arm moves across it to read and write data. Movable mechanical parts are prone to failure. Flash has no movable parts, making it robust, and much faster. But per GB of storage, flash is fourfive times more expensive than HDDs, even though flash’s total cost of ownership, over the lifetime of the product, is lower. Since the upfront cost is high, hyperscalers like AWS, Microsoft Azure and Google Cloud have been slow to move to it. But that looks to be changing.
Meta uses Pure to train its large AI systems, again demonstrating how the speed and reliability of flash is critical in these expensive environments.
Pure’s innovations have been in software, and increasingly, a lot of that is being done in the company’s India R&D centre. Lee says since operating systems were written to speak to hard disks, the early flash players created what is known as a solid state drive (SSD) that could translate between flash and what a hard disk expects. “There’s very complex logic in the SSDs, and all that complex work takes energy and resources, and it becomes very challenging to scale. In our case, we built our software specifically for flash. So it requires fewer resources, and allows us to scale our drives to be much, much larger while still delivering the performance and reliability needed,” he says. Today, in the same footprint and the same power that others use for 8 or 15 TB of storage, Pure ships 150 TB storage, and expects it to be 300 TB by early next year.
That level of power efficiency is a big advantage at a time when AI systems demand massive compute. “All the hyperscalers are running out of power. That’s a huge area of continued innovation and advancement that we’re driving. And a lot of that work is being done in India,” Lee says.
GenAI for security
Aakash Bist, who leads genAI R&D at Pure Storage, also sits out of the Bengaluru centre. He says AI is built into their systems by default now. His team has built a platform that collects data from all installations of Pure worldwide. That allows customers to ask, in natural language, about the organisation’s security posture, how that is relative to others in their industry. Lee says since they manage data storage, they can analyse configurations, identify how the storage is being used, assess which systems are interacting with it, and based on that analysis, provide recommendations.