Scaling AI with Storage Efficiency – with Leaders from Pure Storage, Generac, Lexmark, Comfort Systems USA, Danaher, Alcon, and More

This interview analysis is sponsored by Pure Storage and was written, edited, and published in alignment with our Emerj sponsored content guidelines. Learn more about our thought leadership and content creation services on our Emerj Media Services page.

Data fragmentation remains a critical challenge for organizations worldwide, significantly impeding usability and obstructing digital transformation initiatives. Research underscores that fragmented data systems not only complicate access and analysis but also substantially drain resources and innovation capacity.

A study from MIT highlights that data fragmentation threatens the effectiveness of data linking and analytics, thereby undermining the ability of organizations to generate comprehensive insights and actionable intelligence from disparate data sources.

A complementary UK public sector survey revealed that approximately 70% of organizations experience poorly coordinated or non-interoperable data, which limits their ability to maintain a comprehensive operational view and hinders their digital transformation efforts.

Emerj recently hosted a special series of the ‘AI in Business’ podcast with enterprise executives to explore how organizations are managing and scaling AI infrastructure, optimizing data storage, and improving storage efficiency to support advanced AI workloads.

Executives featured in the series include Shawn Rosemarin, Vice-President R&D in Customer Engineering at Pure Storage; Neil Bhandar, Chief Data Analytics Officer at Generac; Bryan Willett, Chief Information Security Officer at Lexmark; Amit Gupta, Chief Digital Officer at Danaher; Joe Lang, Vice President of Service Technology and Innovation at Comfort Systems; Norma Scagnoli, Chief Learning and Innovation Officer at the Illinois Institute of Technology; Greg Ratcliff, Chief Innovation Officer at Vertiv and Julian Tang, Chief Operations Officer for the Innovation Office at BlackRock.

During these conversations with Emerj Editorial Director Matthew DeMello, the leaders dived deep into the challenges of AI adoption, including data governance, infrastructure scaling, team collaboration, and ethical deployment.

This article examines several key insights from their conversations for leaders aiming to scale AI effectively, optimize data storage, and strengthen governance:

• Optimizing power usage to grow: Assessing and centralizing existing data while evaluating energy constraints to ensure AI initiatives remain feasible, efficient, and aligned with business value.
• Balancing flexibility with discipline: Ensuring cloud scalability delivers real ROI by actively managing storage, cleaning up unused data, and making cost–elasticity tradeoffs before investing in AI infrastructure.
• Evaluating and controlling data risk: Assessing your risk tolerance to determine whether sensitive data should remain on-device, on-premises, or in a hybrid setup.
• Building a layered, decoupled data foundation: Structuring data architecture around aggregation, integration, transformation, and harnessing while using a decoupled design to enable seamless integration.
• Aligning teams to scale systems: Bringing key stakeholders from academic affairs to finance, IT, and student services into the governance process to ensure systems, workflows, and data infrastructure can handle large-scale digital programs.
• Sequestering trusted data: Keeping AI models trained only on secure, internally generated data reduces misinformation risks, improves accuracy, and ensures more reliable outcomes.
• Deploying modular AI infrastructure: Building standardized, modular, and scalable data center units to quickly deploy local AI capabilities while leveraging the same infrastructure as hyperscale systems.
• Engaging stakeholders upfront: Bringing legal, compliance, and InfoSec into AI projects from the start prevents governance delays and ensures smoother implementation.

Optimizing Power Usage to Grow

Episode: Scaling AI with Storage Efficiency – with Shawn Rosemarin of Pure Storage

Guest: Shawn Rosemarin, Vice-President R&D in Customer Engineering at Pure Storage 

Expertise: Customer Engineering, Data Intelligence, Analytics

Brief Recognition: With over 25 years of industry experience, Shawn has held leadership positions at Hitachi Vantara, Dell, and IBM. In his current role at Pure Storage, he leads strategy efforts with engineering teams and customers.

Shawn explains that before jumping into AI, organizations need to take inventory of their existing data. Over the years, businesses have invested heavily in digitizing information. Still, much of it remains fragmented, context-poor, and complex for machines to interpret — for example, doctors’ notes that can make sense to humans but not to AI systems.

The first step, he says, is to understand what data exists, what’s usable, and where it lives — often spread across dozens or hundreds of systems. From there, companies must consolidate and centralize this data to improve accessibility and speed, ensuring it’s close to the systems that need it (a strategy he refers to as giving into “data gravity”).

Finally, AI infrastructure decisions must be grounded in business practicality — the cost of managing and using data should never outweigh the value it delivers to end users.

He warns that energy constraints are becoming the biggest limiter for AI and data growth. He explains that enterprises, countries, and even individuals may soon face power quotas as current energy consumption — especially from GPUs, which use 10 times more power than CPUs — risks compromising public well-being.

New data centers are already being restricted in some regions to prevent power shortages during heat waves. Organizations need to assess their “bridge to running out of power” (e.g., 12, 24, 36 months) and proactively plan whether to invest in alternative energy sources, such as nuclear, hydroelectric, or coal, or wait for innovations like nuclear fusion. Without this foresight, even highly profitable AI initiatives could stall due to insufficient energy.

“I’m confident that at the end, when we look at what’s happening, systems are getting more efficient. The challenge is that there’s a lot of legacy infrastructure being put in place today. There are a lot of inefficient systems. There’s a lot of legacy storage that has been deployed over decades, which is needlessly consuming power that could be better used elsewhere. I’ve been in the business as well. I know that sometimes kicking the rock down the road is a better option than actually doing a wholesale modernization, but if you don’t clean up the foundation of your house, then your bridge to running out of power is going to be 12 months.”

– Shawn Rosemarin, Vice-President R&D in Customer Engineering at Pure Storage

​

Balancing Flexibility with Discipline

Episode: Why Immutable Snapshots Matter for Compliance and AI – with Neil Bhandar of Generac

Guest: Neil Bhandar, Chief Data Analytics Officer, Generac

Expertise: Artificial Intelligence, Machine Learning, Cloud Computing,

Brief Recognition: In his current role at Generac, Neil leads the development of the company’s data strategy and oversees the buildout of analytics platforms and capabilities across the Generac franchise. Previously, he has held roles at Procter & Gamble, JPMorgan Chase, Campbell’s, and Evanta (a Gartner company), among others. Neil holds a master’s degree in Industrial and Systems Engineering from Lehigh University.

Neil explains that many executives face challenges making AI investment decisions because they lack hands-on experience with data, GPUs, and compute. He points out that AI infrastructure is in a deflationary cycle — what costs a certain amount today will be cheaper and more powerful in a few months — creating hesitation around when to invest. He also challenges the common belief that more data always leads to better outcomes.

“There are certain substitutable data elements, which are protected classes of data that could be a proxy. And so, by that definition of being a proxy, they became sensitive data elements. One real example of this is if you look at people’s country of birth, it’s highly correlated to their country of undergraduate education. But your country of undergraduate education is not a protected class variable. Your country of birth is now contingent. So now you’ve got to be sensitive when you think about how you use certain data just because of that proxy association.”

– Neil Bhandar, Chief Data Analytics Officer at Generac

He also explains that while the concept of storing and processing data externally isn’t new — credit agencies have been doing it since the 1960s — the scale of today’s cloud use is far greater. People and businesses now store everything from financial records to personal photos in the cloud, largely because storage costs have fallen and connectivity has improved.

When deciding between cloud, on-premises, or hybrid setups, he urges leaders to evaluate two key factors: cost and elasticity. Cloud platforms offer scalability and convenience, allowing organizations to quickly expand capacity during mergers or spikes in data. However, this same flexibility can become a hidden cost if unused data continues to sit in storage.

Neil’s takeaway from the priority framework he presents is that cloud adoption isn’t just about flexibility — it requires discipline in managing and cleaning up data to ensure that scalability doesn’t quietly turn into wasteful spending.

​

Evaluating and Controlling Data Risk

Episode: Building Storage Strategies That Scale with AI Workloads – with Bryan Willett of Lexmark

Guest: Bryan Willett, Chief Information Security Officer, Lexmark

Expertise: IT Security, Data Privacy, Internal Audit

Brief Recognition: Bryan worked with Lexmark for close to three decades. In his most recent role as CISO, he oversaw all global IT security, data privacy, internal audit, and physical security for 140+ sites worldwide. He has also built Lexmark’s first-ever enterprise-wide IT security and privacy risk program to drive transformational change across the business.

Bryan emphasizes that strong AI governance depends on close collaboration between security, privacy, and AI teams. He explains that when evaluating any new AI solution, organizations should first conduct an ethics review (his team uses the EU AI Ethics Framework as a benchmark), followed by a security review to assess data flow, protection, and access controls.

The goal is to ensure data confidentiality, integrity, and availability, while minimizing the exposure of sensitive information and clarifying who has access and is accountable for the data.

His key point: most companies only connect security and privacy, but excluding the AI team from governance is a mistake. All three must work together from the outset to ensure the ethical and secure deployment of AI.

Bryan segues to discuss how organizations need to be stringent about any data taken from IoT devices, especially biometrics, and that this data should reside in a secure enclave on the device and never leave. Individuals should decide whether they’re comfortable sharing sensitive data, while IoT vendors have the responsibility to be transparent and ensure data is appropriate for the service.

“We know in life sciences, there’s going to be other sensitive data. When you get into that more sensitive data, it still makes sense to do the capital investment on-prem to store your data. But you still can use a cloud service if you need to — that is a model. Like everything, it’s a risk. The organization has to make that decision on what the risk tolerance is, and then they can decide if it is something they’re going to do on-prem, or are they going to go for speed in the cloud.”

– Bryan Willett, Chief Information Security Officer at Lexmark

Building a Layered, Decoupled Data Foundation

Episode: Storage Strategies That Keep GenAI on Budget – with Amit Gupta of Danaher

Guest: Amit Gupta, Chief Digital Officer, Danaher 

Expertise: Digital Transformation, IT Strategy, AI

Brief Recognition: As Chief Digital and Information Officer at Danaher Life Sciences, Amit led digital integration during Danaher’s acquisition of Abcam, delivered $60M+ in AI-driven funnel growth, and built global IT and digital platforms across multiple operating companies. He has over 25 years of experience driving IT, AI, and digital transformation across the Life Sciences, Biotech, CPG, Pharmaceuticals, Medical Equipment, and Industrial Manufacturing sectors. He holds an MBA from the University of California, Berkeley’s Haas School, Wharton, and Nanyang Business School.

Amit explains that data is the fuel powering AI, and to make it effective, organizations need a structured data architecture built on four layers.

He outlines these as:

1. Data aggregation: Collecting data from all sources.
2. Data integration: Connecting systems like CRM and ERP.
3. Data transformation: Cleaning, synthesizing, and preparing data for AI.
4. Data harnessing: Where AI applies insights to drive real business outcomes.

He adds that tools like Salesforce Data Cloud and MDM platforms (like Tamr) help establish a single source of truth by organizing master data (e.g., customer information) before layering transactional data (e.g., sales history) on top.

Amit then explains that during acquisition integrations, data shouldn’t be the primary focus — companies must first align with the overall transition strategy, including basics such as single sign-on, email systems, and cultural integration:

“This is where that aggregation layer that I talked about earlier and the decoupled architecture help. Because as you acquire companies, having that isolated or decoupled aggregation layer helps you integrate those acquisitions to the right data sources. And you can’t boil the ocean. You have to have a prioritized plan for what data types and what data sources you want to tap and integrate into. Again, keep your end goal of the use case and the business case impact in mind. So with any such initiative, it boils down to four things, you know, we all want, cheaper, better, faster, safer, which is cost, quality time, and compliance.”

– Amit Gupta, Chief Digital Officer at Danaher

​

Sequestering Trusted Data

Episode: How Data Ownership Drives Trustworthy AI Models – with Joe Lang of Comfort Systems

Guest: Joe Lang, Vice President of Service Technology and Innovation, Comfort Systems 

Expertise: Leadership, Innovation, Sales

Brief Recognition: Joe has been with Comfort Systems for nearly two decades. He has provided the company with service leadership to develop and grow the organization while creating long-term, strategic goals and expectations for the corporation. He is also an advisory board member for Field Service USA, The Service Council, and Aquant.

Joe warns organizations against overestimating what AI platforms can do for them. He says many teams make two key mistakes: first, over-cleaning or limiting their data based on storage costs, which can restrict insights; and second, they assume that cloud or tech giants like AWS or Google will automatically “fix” their data and deliver ready-made results.

He stresses that while AI is powerful, it doesn’t remove the organization’s responsibility to manage, understand, and apply its own data effectively. Success still depends on human oversight and thoughtful data preparation because no AI model today can fully replace that responsibility.

Lang continues, advising leaders to treat AI implementation like an R&D project, not a quick-return investment. Organizations shouldn’t expect immediate ROI — the real value comes after the data is organized, refined, and usable. The early stages require significant investment, iteration, and fine-tuning before reaching a point where AI can have a meaningful impact on business outcomes.

Joe also explains that his organization has taken an all-cloud approach, but designed its security framework so the infrastructure choice doesn’t affect safety or performance.

“In the grand scheme of things, sequester the data that you know and trust that you’ve generated as an organization. It may not be 100%, but you’ll eliminate 30% that can actually pollute your results. So I think sequestering your data and having it all in one place where it’s easily accessible is a good approach, and it has worked well for us.”

– Joe Lang, Vice President of Service Technology and Innovation at Comfort Systems

​

Aligning Teams to Scale Systems

Episode: Data Pipelines that Support Globalized Education and Training Programs – with Norma Scagnoli of the Illinois Institute of Technology

Guest: Norma Scagnoli, Chief Learning and Innovation Officer, the Illinois Institute of Technology

Expertise: Instructional Design, E-learning Development

Brief Recognition:  In past roles, Scagnoli has been Assistant Vice Chancellor of Enterprise Learning Innovation at Northeastern University and Research Associate Professor at the University of Illinois Urbana-Champaign. She holds a PhD in Human Resource Education from the University of Illinois Urbana-Champaign.

Norma highlights that effective data governance and scaling of educational programs hinge on culture, collaboration, and operational readiness. She emphasizes that challenges in higher education are often cultural rather than technical; longstanding systems, accreditation metrics, and ranking pressures shape institutions’ ways of thinking, making them hesitant to adopt scalable, digital approaches. Overcoming these cultural barriers is the first step before addressing infrastructure or content creation.

Scaling data and programs requires bringing all key stakeholders to the table, including finance, legal, student affairs, academic affairs, research, and faculty, as well as representatives of learners. These units form the “central nervous system” of data governance, ensuring that operational, regulatory, and learner-centric considerations are incorporated.

For example, expanding programs globally necessitates systems that can handle diverse tuition methods, transfer approvals, and learner demographics, while maintaining accuracy and compliance.

Norma also underscores that instructional systems have evolved. Faculties no longer work in isolation but are supported by teams that modularize content, adapt videos for different learner types, and coach instructors on clarity and presentation. The modular approach she describes allows programs to scale efficiently, repurposing evergreen content while enabling personalization for degree-seeking students, corporate learners, or non-credit audiences.

​

Deploying Modular AI Infrastructure

Episode: Scaling GenAI Without Melting the Data Center – with Greg Ratcliff of Vertiv

Guest: Gregory Ratcliff, Chief Innovation Officer, Vertiv 

Expertise: Data Science, IoT, Cloud Computing

Brief Recognition: Gregory has over 30 years of experience in leading and managing technology teams, developing and launching new products and services, and creating and executing data and innovation strategies. He has been a doctoral candidate at Liberty University, where he researched Agile Project Management of IoT and Big Data projects.

In his conversation, Gregory draws an analogy between data infrastructure and food distribution to explain trends in AI and data storage. He says that just as large, centralized food warehouses serve significant metro areas efficiently, today’s enterprise AI relies on massive centralized data centers and services for efficiency.

However, he points out that there’s a growing need for smaller, local “markets” — modular, regional data stores that are highly connected and offer low-latency access.

These local data hubs, provided by colocation or cloud providers, mirror large data center services but serve regional needs with low latency and proximity benefits, such as local disaster recovery, signaling a shift toward distributed, locally optimized AI infrastructure.

He then emphasizes a hybrid AI approach, keeping sensitive data and specific AI capabilities in-house while leveraging external platforms for additional capabilities.

He notes that specific sensitive data, such as hospital MRIs, will never be handled by large external AI providers due to privacy and regulatory concerns, so some AI must remain local.

He explains that emerging industry standards allow efficient, lower-cost production and foster competition. These building blocks, already used in large hyperscale data centers, can be scaled down into smaller, modular systems, such as a one-megawatt, container-sized data center funded through the Department of Energy’s ARPA-E program.

The concept is akin to a “playset” for data centers: you can quickly deploy modular units wherever needed — at a factory, for high-security workloads, or for local AI processing — while utilizing the same standardized components as large-scale data centers. Essentially, he predicts a future where flexible, modular, and scalable data center building blocks support both hyperscale and local AI needs.

Engaging Stakeholders Upfront

Episode: Turning AI Ambition into Infrastructure Reality – with Julian Tang of BlackRock

Guest: Julian Tang, Chief Operations Officer for the Innovation Office, BlackRock 

Expertise: IT Strategy, IT Operations, Leadership

Brief Recognition: Julian oversees technology scouting, accelerates AI and digital solutions, and develops patent strategy at BlackRock. He has a Master’s Degree in Business Administration from USC Marshall School of Business.

Julian emphasizes the importance of involving all key stakeholders early in the implementation of AI, particularly legal, compliance, and Information Security (InfoSec) teams.

He explains that many organizations focus heavily on AI’s risks and often wait too long to include these groups, which leads to governance issues and delays. When these teams are brought in only at the end of a POC, it becomes messy — with teams scrambling to meet requirements retroactively or shoehorning it through the door, as Julian puts it.

He says most established companies already have experience with due diligence through “Know Your Third Party” (KY3P) processes — frameworks for assessing risk, legal, and compliance when working with vendors. He suggests that they leverage the same mindset for AI initiatives by bringing in legal, risk, and security teams early and clearly defining what the organization is trying to achieve.

He emphasizes that an AI strategy doesn’t need to be overly complex — it should fit on a single page.

“Start to pull in all of these AI projects that are siloed. Get it into a chart, a one-pager. If it’s taking more than one page, you’re probably looking at too many things. As you review those projects, begin to align them. Are they core to our company’s strategic goals? If not, then maybe some of those POCs should go on a pause and get down to a handful of the ones that they can really focus on. As we start to think about the roadblocks that might be there, assign some owners and start removing those roadblocks. Get them to be really clear about what they’re going to deliver in terms of their business value.”

– Julian Tang, Chief Operations Officer for the Innovation Office at BlackRock