Operationalizing AI: How Inferencing Changes Enterprise AI Deployment at Scale - Six Five On The Road

Enterprise adoption is entering a more demanding phase, where intelligence is expected to perform consistently, not just demonstrate potential.

Patrick Moorhead and Daniel Newman speak with Scott Tease, Vice President, Product Group, ISG at Lenovo, about how inferencing connects infrastructure, architecture, and operations as enterprises scale intelligent systems beyond pilots.

Their discussion looks at what it takes to run inference across distributed environments such as factories, retail locations, healthcare settings, and remote sites. As organizations encounter the limits of centralized strategies, factors like latency, data gravity, and operational complexity increasingly shape outcomes.

Key Takeaways Include:

🔹 Inferencing marks the shift from experimentation to operations: Real-time execution, not model creation, determines whether enterprise initiatives deliver durable business value.

🔹 Centralized strategies are reaching practical limits: As data spreads across locations, latency, bandwidth, and coordination challenges expose weaknesses in cloud-only designs.

🔹 Operational complexity defines success at scale: Deployment discipline, monitoring, and lifecycle management are as critical as compute capability.

🔹 Distributed architectures are becoming foundational: Aligning inference closer to where data is generated improves responsiveness and reliability.

🔹 Execution separates leaders from laggards: Long-term advantage comes from embedding intelligence into daily operations, not from labeling projects as transformational.

Learn more at Lenovo.

Watch the full video at sixfivemedia.com, and be sure to subscribe to our YouTube channel so you never miss an episode.

Listen to the audio here:

Disclaimer: Six Five On The Road is for information and entertainment purposes only. Over the course of this webcast, we may talk about companies that are publicly traded, and we may even reference that fact and their equity share price, but please do not take anything that we say as a recommendation about what you should do with your investment dollars. We are not investment advisors, and we ask that you do not treat us as such.

Patrick Moorhead:
The Six Five is On The Road, and we are here at Lenovo Tech World at the iconic Sphere in Las Vegas. Daniel, we are live, baby. Look, this infrastructure here at the Sphere is unbelievable, and we are ahead of the big show with YY, with Jensen, with Lisa, with Lipu, with Cristiano, and a lot of other great guests.

Daniel Newman:

Yeah, we've got a star-studded lineup. Not as good as the lineup here at our desk.

Patrick Moorhead:

Oh, that's right. Just wait till you meet our next guest.

Daniel Newman:

We will have fun. I mean, look, Pat, it is an exciting week. And yes, this venue is iconic. But the AI transformation is just mind-blowing. The speed at which things are happening right now. I know there's some people out there that have their doubts, but what I'm seeing day in and day out is the pace of change and the opportunity that we have to drive productivity into this world with AI is just incredible. So I'm just loving listening to all these experts and having these conversations here.

Patrick Moorhead:

Yeah, it's crazy. Infrastructure used to be not sexy, but guess what? It's sexy. It's driving the markets today. It is back. And by the way, it's changing very rapidly. And one of the things that we saw in the early days of machine learning, about seven or eight years ago, was the transition from training to inference. And with generative AI and agentic AI, we're seeing the same thing. But architectures need to change to be able to meet that new agentic AI need. And I would like to welcome Scott Tease back to The Six Five to have this conversation about inference.

Scott Tease:

Good to be with you guys. Thanks for having me again, as always. I appreciate it. Absolutely. Is this like your third or fourth time? I always like seeing you guys, especially in places like this. Isn't this great? It's insane.

Daniel Newman:

Yeah, it is great. Yeah. Let's talk a little bit about doing AI, right? I mean, some of us are building these more frontier kind of organizations where we're looking at every process in our business and we're saying, where does that fit? Where is it? autonomous, where is it AI human plus machine, where is it still very human heavy with maybe a little support, machine learning, analytics, there's this whole continuum. But that's not what, I mean, most organizations I think are still, we're still early. We're still really early. I'm just kind of like, What are you seeing in terms of that gap between the ambition, stuff we're seeing at events, and the speakers we mentioned here, what they're saying, and the enterprises that companies like yours at Lenovo is dealing with every single day?

Scott Tease:

Yeah, we're talking to customers all over the world, as you guys know. And I think the big thing that we're seeing is a lot of companies that go into AI, they're not quite ready for the operational excellence that it's going to take to run this AI. What we're trying to do now is move out of those frontier models and take a llama or open AI, and actually turn it into something people are actually going to use. When we do that and we start running inference, that's where people are going to experience the benefits of AI. That's where a worker is going to get more efficient. That's where a factory is going to make decisions smarter about quality. Doctors are going to make better decisions. It's that inference workload that we're starting to see take off. But some customers are not ready for operationalizing that. They're still focused on building the model out. not putting into like an ongoing, perpetually evergreen state. And that's what we're helping them do right now. And that's the exciting part of this journey. And that's just getting started.

Daniel Newman:

So is it just them building the model or what about like these enterprise, like how about the adoption side of these things? Like, are you still seeing like Doctors Resist or, you know, the lead foreman on the shop floor? I mean, are they buying into it too? Or is that another point of contention?

Scott Tease:

You know, when we first started doing testing with AI, it was data scientists or technical people were doing testing, and we were doing a lot of showmanship. We were teaching a machine to do something that people only thought a human could do. Now we've moved past that. Now the lines of businesses are coming to the table, and they're asking for, hey, how do we deploy this AI? How do we use it? How do I get closer to my customer? How do I get more efficient? So it feels like people are embracing it a lot more than they did in the past, which I think is step one in getting it done. So we're seeing real viable AI being deployed right now today.

Patrick Moorhead:

Yeah, there's a lot of enterprises. I did a recent enterprise advisory board in Europe for a customer I may not name, but had a bunch of CIOs there and they were talking about AI and how they're deploying. A lot of them were using AI, but it was through basically the web or through a SaaS. And then on-prem and on the edge, they had done a lot of experiments. They had done some POCs, and some of those POCs deliver value, and they're scaling it. And that's where we bring in inference. And can you talk to me why, above the obvious, which is why inference is important, because you actually have to run the applications. Is it any deeper than that? the value of inference in rolling this out?

Scott Tease:

You know, you said something pretty neat, Pat, that the early ones you saw were a lot of SaaS type things. I think what was nice about SaaS is it made it really easy to make AI part of the workflow. You know, if you're doing AI for AI's sake and you're just doing like these one-off projects, experimentations, It's not likely going to end in something you can actually deploy and do in real time. If you're building it in as part of your workflow, much better chance of success. And I think that's what we're seeing a lot more, whether it's done in the cloud, whether it's done in the data center, or if it's done on the edge. And we're seeing all of that. That's part of that hybrid AI vision that we've talked to you guys about in the past. AI is going to have value chain, bits of the value chain, all along that, from the cloud all the way out to the edge, even the desktop and in your pocket.

Patrick Moorhead:

That's right. Hybrid cloud I knew was going to be a success because it's what customers want. And I always like to say that a company is one acquisition away from having another cloud vendor as well.

Daniel Newman:

Remember that silly time when clouds thought they'd be The cloud.

Patrick Moorhead:

Right, it was going to be everything. And here we are with AI, Daniel, and, you know, every cloud, public cloud is building all of this software just for their customers. And they're really not talking to other clouds, which I find fascinating.

Daniel Newman:

Yeah, well, I mean, what most organizations look like, you know, they're very distributed.

Patrick Moorhead:

Yeah.

Daniel Newman:

You know, so a lot of things we're talking about here is like, you know, you might think of a company where a traditional chatbot could do a lot of things for that company. Like, think about a retail, you know, a large retailer. You know, Amazon's one that's come to mind, but think about a physical retailer, like Walmart. Well, you know, the customer experience might be very much like a centralized AI thing could work in terms of creating a customer experience. Cloud, you don't need, latency's not a huge issue, it could take a few seconds. But then you think about the whole supply chain. Think about distribution centers. You've got trucks. You've got packaging. You've got boxing. You've got financials. And all of a sudden, you've got this really crazy distributed compute requirement. And that's really hard. So I'd love to get your take on this. We go across factories. You've got stores. You've got hospitals. You've got remote workers. You've got all these things. centralized AI can work for some of them, like I just said. Can't work for all of them, so how are you helping companies deal with that fragmentation, that complexity?

Scott Tease:

Yeah, again, I think a big part of it is accepting that AI is going to be hybrid, that centralized model, where everything goes in either into my data center or in a cloud data center, it's not going to make it. Data has way too much gravity. And we talk about data gravity a long, long time. This AI is proving how strong that gravity pull is. Customers have two choices. They can bring their data to AI somewhere out in a cloud, or they can bring AI to their data. And a lot of companies are going to find that the latency the data sovereignty issues, security, just the networking, you know, bandwidth and bulk to move that much data is going to mean you want to bring the AI to the data rather than the other way around. So, being able to run properly sized AI at the edge, in the data center, and out in the public cloud is the right way to do, and that's what we're guiding our customers to do as they get started.

Patrick Moorhead:

Yeah, that's right. If we look at history, typically compute starts in a centralized place. We saw it with mainframes, we saw it with minicomputers. Actually, minicomputers were the disaggregated version of the mainframe. And then client-server disaggregated that. The industry goes through this aggregation and disaggregation, and what I'm pleasantly surprised at is the tools to manage between the different environments are getting better. You want to manage the data, you need to manage the security. between those, and there will be agentic orchestrators as well that go between these different environments. But as AI becomes embedded in everyday life, what is going to separate I don't want to call it the winners and the losers. There will be. I'll call it a lasting competitive advantage. There you go. Is that better? That's nicer. A very politically correct.

Scott Tease:

That's a polished version of it.

Patrick Moorhead:

Durable. A durable lead. Well, look what we saw in the age of the web, in the age of e-commerce, in social, local, mobile. Those companies who didn't embrace it and just crush it, they went out of business, or they went from the number one market share company to the number five market share company.

Scott Tease:

Yeah, I think a big part of it is how you approach AI. If you're going into it thinking you're going to build the killer model for some specific function, you're likely not going to be one of the long-term winners. If you go into it thinking this is all about operationalizing for the long-term, keeping it reliable, keeping it evergreen, keeping it secure, those are the winners there. Those are the people that are treating AI, again, as part of that workflow that they're already doing. That's who's going to ultimately win. People that have good data practices today already in place, they're going to benefit because they're going to be able to adapt to that AI better, embrace it more fully.

Patrick Moorhead:

Also, we try to give guidance to enterprises on really find out what is strategic to your company. And you need to do that yourself. Don't outsource it. It doesn't mean you do all the tech yourself. It's just that is a proprietary value add that your company is. And then quite frankly, Commoditize the rest, because they don't differentiate you, right? Maybe HR is not necessarily something that is differentiable in your company, but hey, if you have a million employees, it probably needs to be. If you're making cars.

Daniel Newman:

Look how many agents you have.

Patrick Moorhead:

Yeah, exactly. And then if you're making something, it's manufacturing, maybe not a different area. Are you getting that same kind of feedback?

Scott Tease:

Yeah, I mean, again, I think people are looking at where can they build an AI? What's already built? What do I have to build myself? Again, their data, I think most companies come to realize their data is the key to unlocking the biggest amount of value. But there's some low-hanging fruit that you can easily implement. A lot of that's being done in the cloud, which is, again, part of that whole value chain. Choose it to go in the cloud for those types of things. Other bits where it's your data and you're going to unlock valuable insight from that, you want to own that. And again, if you're in another country, you may not want that data leaving even the country border. You're in the EU, you don't want to leave in the EU. You're in Japan, you don't want to leave in Japan. You're going to want to be able to do a lot of that kind of stuff on your own. Maybe it's outsourced to a GPO as a service company, but somewhere locally. And again, I think a lot of it's going to come back to where that data resides. You're going to want to bring the AI to it. And not all AI has got to be super power intensive, super exotic. A lot of it can be on devices like this, bound it to a wall, an edge device, or even just regular data center stuff. Exactly.

Daniel Newman:

It's really interesting, too, because I think a lot of people have sort of misconstrued the size of the opportunity for AI. They really look at it, you know, as chatbots and LLMs. And I think it was a really good point you made, though, is companies start to realize their data. Companies that have something unique will be the valuable companies. And so what AI is doing, though, is people that have sort of become brokers of widely publicly available information, that's going to collapse. Because it's too easy to organize, sort, distribute. But if you like an aggregator like someone was a job aggregator and there's all those companies I'd be like oh we aggregate all the job postings from everywhere like hey I could do that you can build that up and we can stay together and no problem. But like if you have really truly unique understanding of. you know, say like a Tesla that has billions of driven miles, that can truly build a world model for what driving looks like. There's so much value in that. So I think companies figuring out how they can take their data is going to be one of the big projects over the next handful of years for them, because they're going to have to reposition.

Scott Tease:

And what they do. Think about it, Dan. So, like, today, if you want to interface with an SAP environment or some big software database or whatever, you've got to be a pretty deep expert to be able to do something powerful in that software. With AI, we're going to interface with it in human language. I'm going to speak to it in English. So, I mean, it's like the playing field's been leveled for so many people to take advantage of this super powerful technology, and we're just at the beginning of touching what it's going to be able to do.

Daniel Newman:

Well, Scott, I think your playoff music just came on. I know, that's my walk-off music. Walk-off music, but in all seriousness, great to have this conversation. Thanks for joining us. Always good to see you. Pleasure, mate. Thanks, guys. Thanks, Scott. And thank you, everybody, for being part of this Lenovo Tech World pregame coverage. Patrick Moorhead and myself here having a number of conversations. We're going to take a quick break for now. Stick with us. We'll be back soon.

‍