Hi, everyone, and welcome to The Six Five Summit: AI Unleashed 2026. We're continuing the conversation with another AI infrastructure spotlight. I'm Nick Patience from The Futurum Group, and today we're exploring the next evolution of AI infrastructure as organizations look to build, activate, and operate AI capacity at greater scale. Joining me today are Binoy Unnikrishnan, Vice President of Worldwide Cloud Sales, and Wil Wellington, Global Director of AI Data Center Services SSG at Lenovo. Binoy, Wil, welcome to The Six Five.
Thank you very much.
Thank you.
Binoy, I'll just start with you first of all. If compute is no longer the primary constraint on AI at scale, what has become the defining challenge for organizations building the next generation of AI infrastructure?
Yeah. It's hard to pinpoint as one particular item. We've seen the constraint sort of move around quite a bit from GPUs to memory to power, land, power, shell, all of the above. I call it a comprehensive execution, if I were to use a couple words to define it. It's how soon can organizations coordinate all of the above from land, power, shell, infrastructure, deployment, getting it all up and running.
It's not one thing. It's not like being able to pinpoint on that one thing anymore. It's a whole gamut of things.
Yeah. I think it's a gamut of all things coming together to get the infrastructure up and running and delivering tokens at scale.
Wil, a question for you. If execution has become the real differentiator in scaling AI infrastructure, where do organizations need to start thinking differently before deployment has actually begun?
One core belief that drives our strategy is that optimization shouldn't start after deployment. It has to start during the initial design phase. That's precisely why we engage so early with our customers. We help them evaluate site readiness, power availability, thermal footprint, rack density, and long-term operational targets right from the start. Because those up-front decisions directly impact how efficiently that AI factory performs over its entire lifetime. Our Lenovo Neptune liquid cooling chassis is a prime example of this in action. We aren't just helping customers select cooling hardware, we're working alongside them to address overall power efficiency, warm water cooling strategies, thermal management, and facility design. The environment is engineered for long-term economics. By the time the AI factory actually goes live, we already help optimize the design to support higher compute utilization, superior power efficiency, and far easier lifecycle management.
At that point, the conversation will naturally shift to continuous optimization, maintenance, you focusing on fine-tuning thermal performance, maximizing supports in the environment with our onsite team, and adapting as AI models, inference workloads, and business demands evolve. Overall, our mandate is to keep the environment operational post-deployment. It's very important to remember that. Keep the actual environment operational post-deployment. Ultimately, optimizing isn't just the final project phase for us, it's the guiding principle across every stage of the AI lifecycle, ensuring efficiency is built in from day one.
Binoy, back to you. With thinking about building AI capacity with confidence, how does that translate into getting customers onto that infrastructure and creating business outcomes faster?
Yeah, Nick. I maybe draw a parallel going back 10 - 12 years to the beginning of the cloud era, right? We saw hyperscalers were building massive infrastructures going back then, and it feels a lot similar to me. But it's not just a few hyperscalers. You've got a lot of neo clouds in this space as well, a lot of startups. Infrastructure back then was probably in the millions, tens of millions. Now you've got some of these startup companies investing billions, tens of billions of dollars, to generate tokens and results as soon as possible. In quite a few cases, the capacity is already sold out before they even start building their infrastructure. Any delay in any of the phase, right?
I've talked about constraints being not just one, but there's multiple constraints in acquiring the land, planning it, manufacturing it, deploying it, activating it, meeting your commitments and keeping it up and running. Any delay has a significant impact in terms of the investment going in and the ROI that businesses can get out of it. We've heard people use KPIs and metrics like tokens per dollar, tokens per dollar per hour, tokens per dollar per watt. Those are the key metrics that are being used to track how soon infrastructure can be up and running. Speed is definitely key in this space. From our perspective, we view our role as removing friction, removing some of these roadblocks, and we work with customers upfront in planning ahead, shifting left, and thinking ahead on some of these problems before you get into execution.
The more you can do before the execution actually starts, the sooner it is that you can deploy and deliver outcome for customers. I also want to talk a little bit about capital. It hit the billions or tens of billions of dollars, and lots of neo clouds in the space. Access to capital. Now flexible capital is key. Lenovo has also had to work with customers with our LGFS, our financial services organization, in helping them access capital that's needed to get the infrastructure up and running. So it's sort of a comprehensive view in terms of how we engage customers in not just selling hardware, but all the services that sort of go with it.
Wil, another question for you. If we think about deployment as only being the beginning, once an AI factory is operational, what does it take to keep improving its performance and efficiency as workloads and custom demands evolve?
Well, one of the big lessons we've learned along the way is to always design for where you're going, not just where you are today. A fundamental mistake that a lot of people make. Even if you aren't building a gigawatt scale AI Cloud Gigafactory right now, these exact same execution principles still apply. You have to think holistically about power, cooling, networking, operations, and lifecycle management from day one. It's also critical that we don't treat deployment as the finish line. An AI factory is a live production environment, one that will continuously evolve as workloads, modules, and customer demand shift over time. That's why building repeatable into the design is non-negotiable. Whether a customer is deploying a single cluster or scaling across multiple global sites, a repeatable architecture makes operations smoother and continuously improve vastly easier.
At the same time, we have to recognize that supply chain is no longer just procurement. Manufacturing, deployment, maintenance, [firmware-based spare parts], provisioning, and full lifecycle planning are all integral parts of the long-term operating model. Finally, don't underestimate the power of a good partnership. The most successful organizations aren't just buying hardware. They're looking for partners who engage early to reduce execution risk, optimize their footprint, and stay by their side throughout the entire operational cycle.
Looking ahead, Binoy, I'll come to you first and then we'll get Wil's comments. What capabilities will distinguish the organizations that successfully scale towards AI Cloud Gigafactories? What lessons can enterprises begin applying today, regardless of their size?
Yeah. If you look at organizations that'll lead the next generation of AI, it's not just building more infrastructure, it's building the confidence and time. I go back to the analogy of back in the early days of cloud. If you look at, cloud was initially a destination, then it was a deployment model and architecture. But when you actually deploy something at scale, you're solving a different kind of problem, right? That sort of becomes the default. Similar to that, I think only a small number of organizations will truly be building gigawatt scale factory. But the deployment principles, the problems that are solved at that scale, will sort of percolate all the way down to smaller enterprises as well. Liquid cooling is an example that drives a lot of density.
Once your infrastructure is liquid cooling ready, then there's no going back. That's how infrastructure is designed, and that's how you can get a competitive advantage, and that sort of builds on it. So, it's not about, you asked what the constraints were. I said it's not just one constraint, it's kind of many constraints and solving it at that scale and that speed. I see these problems being solved at the gigafactory level and then the same sort of architectural innovation and solutions will be available to enterprise customers right off the bat. The technology is obviously critical, but how you operate the environment over time is also equally important. I'll probably hand it over to Wil to talk a little bit more about our services capability and how we do that.
One core belief that drives our strategy is that optimization shouldn't start after deployment. It has to start during the initial design phase. That's precisely why we engage so early with our customers. We help them evaluate site readiness, power availability, thermal footprint, rack density, and long-term operational targets right from the start, because those upfront decisions directly impact how efficiently that AI factory performs over its entire lifespan. Our next-gen Lenovo Neptune liquid cooling expertise is a prime example of this in action. We aren't just helping customers select cooling hardware. We're working alongside them to address overall power efficiency, warm water cooling strategy, thermal management, and facility design, so the environment is engineered for long-term economics. By the time the AI factory actually goes live, we've already helped optimize the design to support higher compute utilization, superior power efficiency, and a far easier lifecycle management.
At that point, the conversation naturally shifts to continuous optimization and maintenance. A few focusing on the fine-tuning thermal performances. I say a few are focusing on the fine-tune thermal performances, but they are maximizing it and supporting the environment with our on-site team and adapting AI models, inference workloads, and business demands evolve. Overall, our mandate is to keep the environment operational post-deployment. Ultimately, optimization isn't just a final product or project phase for us. It's a guiding principle across every stage of the AI factory lifecycle, ensuring efficiency is built in from day one.
That's great. Wil, Binoy Unnikrishnan, thanks for joining us for this AI infrastructure spotlight here at The Six Five Summit: AI Unleashed 2026.
Thanks.