As organisations increasingly adopt AI and machine learning, the demand for storage will only continue to grow. However, with rising power consumption and finite energy resources, not all enterprises can afford the costs or access the capacity they need.
With cloud costs surging and data centres becoming more power-intensive, how can organisations strike a balance? Robert Lee, Chief Technology Officer of Pure Storage, a unified enterprise data platform provider, shares his insights on these challenges. He discusses AI’s growing impact on storage, global trends in cloud repatriation, and the evolving role of data infrastructure in addressing these issues.
What are your recent observations in terms of how businesses are leveraging Pure Storage?
There are several things, but if I were to pick two, certainly AI is dominating many of the conversations and shaping how businesses approach IT choices. The second is a heightened focus on sustainability, particularly energy efficiency, which mirrors what we’re seeing globally.
About five years ago, we began to see this focus emerge in Europe and the United States. In Europe, it was partly driven by sustainability and ESG reporting requirements, as well as energy security concerns tied to geopolitical events. Now, with AI making its way into data centres across various regions, the emphasis on energy efficiency has intensified. GPUs, for example, are incredibly power-hungry, so organisations are looking for efficiencies in other areas. These factors are influencing customer conversations and IT decision-making in ways we haven’t seen in a long time.
Do you also see geopolitical implications around energy efficiency here in Asia?
Until recently, Singapore had a moratorium on data centre construction due to concerns about energy scarcity. This highlights how energy efficiency isn’t just a technology or data centre issue — it’s also a regulatory one. It’s a top priority for everyone as they work to meet growing demand. There are a few ways to approach this: increasing energy supply, improving efficiency, and conserving energy where possible.
From our perspective, we see significant opportunities to help customers in the industry save energy. This is our biggest driver, for sure. A secondary driver is the speed of our storage solutions, which can make applications faster. Faster applications require less compute power, which, in turn, saves energy.
But pound for pound, the main driver is this: for the storage that needs to be deployed, we can generally deliver that at five times the efficiency, or sometimes 10 times the efficiency, of the systems we’re replacing or competitive systems. This means we can reduce power utilisation by 80%.
One of the trends we’re seeing in APAC, and globally as well, is cloud repatriation. Is that something you’ve observed across the board, as well?
We’re definitely seeing a greater focus on cloud cost management across the board. Cloud repatriation, like any major shift, begins with admitting you have a problem, understanding the issue, and then deciding on a solution. On one end of the spectrum, repatriation is a clear choice; on the other end, some organisations may simply request more money to cover rising costs. In the middle are those who want to stay in the cloud but are looking to optimise their operations and save money.
I don’t think asking for more money to throw at the problem is a great strategy, so we’re working with customers on the other two approaches: either repatriating workloads or optimising their cloud usage to run more efficiently. These trends are becoming more pronounced globally, and I see them as an artefact of the post-COVID global economic slowdown. That’s when I began having more conversations with customers, as CFOs started taking a closer look at their cloud bills.
Over the last couple of years, customers have become much more sophisticated in understanding their cloud spending, which has driven many of these repatriation discussions. Another factor, though secondary to cost, is growing sensitivity around data control. This stems from uncertainties over how large language models (LLMs) were trained. However, cost remains the primary driver behind cloud repatriation.
What does the AI revolution in the enterprise mean for storage in the future?
I think we’re entering a period of change, and the technology is still very nascent. Things are evolving week by week, month by month, and quarter by quarter, making it difficult to predict what the landscape will look like in five years. Just consider this: ChatGPT didn’t even exist five years ago, and there’s always going to be something new. The pace of change is happening so quickly.
That said, I do have some beliefs about where things are headed. I think the most successful customers and technologists in this environment will be those who position themselves for maximum flexibility and optionality to navigate these changes. Since we can’t predict the outcomes, I’d argue that the best strategy is to stay as flexible as possible. This enables organisations to react quickly to changes, rather than relying on the static IT infrastructure strategies that have been in place for decades. While those strategies worked well for large, monolithic ISVs (independent software vendors) like Oracle, I don’t think they’re suited for AI.
I think the biggest change will be that the most successful customers position themselves for maximum flexibility. The broad interchange of data will also be crucial. For AI to be deployed in the enterprise across multiple use cases, it must integrate with a variety of data sources — like databases, customer interactions, and documents. Systems and infrastructure that enable this fluid interchange and connection of data will become increasingly important. Looking ahead, flexibility and the ability to integrate data sources are the two biggest priorities I see.
Speaking of Oracle, you spent 10 years there, and now you’ve been with Pure Storage for 11 years. Before that, you worked on compilers during your time at MIT. What was the transition like?
I’ve always gravitated toward tackling hard systems problems, particularly in areas with significant untapped potential or at points of major change. When I was in graduate school studying compilers, the semiconductor industry was undergoing massive consolidation. Just five years earlier, there were probably two dozen different chipsets — not just x86 but also IA64, PA-RISC, and many others whose names I can’t even remember now.
As the industry consolidated behind x86, it brought greater efficiency. Software providers no longer had to build and port applications for 35 different platforms — they could focus on just one and start adding more value. However, to make that happen, you had to get really good at making that chipset work, and compilers played a key role. That’s what drew my interest.
Moving to Oracle felt like the next wave of that change. Java as a programming language was the next stage in the evolution of compute virtualisation, especially as the internet gained traction. The ability to run applications in a browser required excellent compiler support, which Oracle was heavily involved in. So I joined them. During my time there, I worked on various projects, but always at the forefront of where I saw technology evolving.
That’s also what took me to Pure Storage. I joined Pure Storage in 2013. At the time, the technology had a lot of potential, but there was still a lot of engineering work needed to fully realise it. That challenge appealed to me — it was a tough problem with a big market and huge potential.
I think we’re only now reaching the finish line of that initial journey. Where we are today allows us not just to finish that chapter but to open new ones. I’ve always chosen to work on hard problems that I think will deliver meaningful value, and the most valuable opportunities lie at the intersection of technology change and market change. That creates opportunity.
What’s the most interesting thing happening in the Pure Storage labs right now?
There’s still a long way to go with our core flash technology, but we’re by far the industry leaders — not just in understanding how flash works at the fundamental physics level, but more importantly, in connecting that to enterprise systems. Think of flash as a raw material with immense potential. You need to harness, distil, and refine it. We’ve perfected our technology to the point where we can now take the benefits of flash beyond the enterprise and into some of the world’s largest hyperscale environments, replacing the still excessive reliance on disk deployments. At the same time, we can tune and configure that same technology slightly differently to address higher-performance use cases. There’s still a lot of room for innovation in what we’re doing with flash, and we have many exciting projects in the lab.
To give you an idea, this ties back to the efficiency discussion from earlier. The rate of improvement we’ve been demonstrating with our flash technology has essentially been doubling every 12 to 18 months. It’s like Moore’s Law and not sustainable forever, but we’re still on a very rapid improvement curve.
We’re also investing a lot in and working on really helping on-premises traditional data centre customers get more of what we call cloud operating model. We are seeing a lot of people repatriating from the cloud, which makes you wonder why people went to the cloud in the first place. They didn’t go because it costs less, but because when you go to the cloud, there’s less you have to manage. You just describe what you want and how you want it done. There’s a bunch of operating model changes that happen, and I think we have a very unique opportunity to bring that operating model to our customers on their own premises.
