How Lenovo wants to unlock AI for all at the edge | Kirk Skaugen interview

It makes a lot of sense to put AI as far out on the edge of the network as possible, as it can solve problems for people like first responders and reduce the amount of useless storage sent to the cloud.

Lenovo demonstrated a bunch of these solutions at the recent Mobile World Congress 24 show in Barcelona, Spain. The big company unveiled its “AI for All” vision with telco customer solutions that will accelerate global AI deployment.

Kirk Skaugen, EVP and president of infrastructure solutions group at Lenovo, spoke with me about the plans deploy public safety AI applications at the edge and reduce power consumption of cloud services by 25%. The company has partnered with the likes of Telefonica, Orange Business, Intel and Rakuten to propel innovation and savings across industries with integrated solutions for mass deployment of AI at the edge.

The goal is the help enterprises harness vast bodies of data at the far edge for transformative AI applications at scale while reducing energy consumption. The innovations are part of a comprehensive pocket-to-cloud portfolio of Lenovo Hybrid AI solutions designed to simplify the path to intelligent transformation for all industries and are attracting new customer collaborations with industry leaders, like Telefonica, that unlock the power of AI anywhere data.

As the telecom industry continues its dramatic evolution to enable the rollout of 5G and an AI-powered future, innovations in IoT networks, cloud infrastructure and edge computing are critical to connecting today’s digital economy. Edge computing allows businesses to analyze data in real-time, enabling faster actionable insights for more efficient operations and services.

Combined with AI-optimized servers its ecosystem of partners, Lenovo’s comprehensive range of far edge to cloud solutions are enabling key service providers to quickly deploy an entire network of high-power computing to drive revolutionary efficiency and intelligence for their own customers and beyond.

Here’s an edited transcript of our interview.

Kirk Skaugen: I just got back from Barcelona Supercomputing. We’re now running the largest Intel-only supercomputer in the world, non-GPU. We’re number 19 on the top 500, but just CPUs. All water-cooled.

VentureBeat: I remember that place because I read about it in a Dan Brown book, Origin.

Skaugen: That’s right! I got a signed copy from Mateo Valero. They were very excited that Dan was inspired when he went and took a tour. I’m still waiting for the next Tom Cruise movie where he comes down from the ceiling into the middle of the data center.

VentureBeat: What are the big topics on your mind?

Skaugen: I had a briefing last night from the people that run MWC. It’s almost back to the highest–we’ll see what the final count is, but it’s almost back to pre-COVID attendance. We have our pocket to cloud strategy. Edge, AI, and sustainability are our core messages here. We have tons of devices. We have software stacks validated for management orchestration. We have cool hardware. We’re leading beyond Intel now into AMD at the edge. We have good carrier relationships.

You may remember, we’re powering the smart city of Barcelona now in the street kiosks with our edge servers. We can do things like, if there’s an accident in the night, it can automatically call the police, detect if an ambulance is needed, call more first responders. Those are more traditional AI use cases at the edge.

We’re also doing things like helping the visually impaired walk the streets of Barcelona. Not just coming up to a street corner, but “Hey, there’s a bicyclist coming up behind you, be careful.” If you think about five years from now, you can get a professional shopper with AR glasses. Get on your notebook, send them to the market, and have them taste the olives. If he likes them as a professional olive taster, you can get them sent to your apartment.

What we announced this week, with Telefonica we’re doing the city of Madrid as well. We’re using Moto for push to talk on their 5G network. We’re using the city cameras, and even putting drones in the air, to detect fire and smoke for faster police response. I thought that was a little crazy at first. How many fires do you have here? But they’re running drones and using AI algorithms to detect smoke and fire. If there’s a big multi-building fire they can put the drone up and decide how to deploy the response more effectively. That’s a cool one.

We have a new AMD edge server. We’ve tested this thing with Orange. We’re getting 25% lower power in cloud. We also have something that’s 50% higher in CPU performance, three times the number of GPUs, half the power. We have this tested with some telcos. We get 46-52% lower power for the ORAN network and 50% for the acoustics. What we mean by that is, when you’re in someone’s retail edge locations, the limiter has been the fans. We won a large retailer in Australia because our competition didn’t meet the safety spec. The server was spinning with the GPUs. People were on conference calls and they couldn’t meet the acoustic spec. We created some patented technology around how we’re using the dual rotor fans and how the screws go through the chassis, so they don’t create turbulence. That product is incredible. Every big telco is talking about finally deploying their open radio access network. There’s such a focus over here on sustainability. When you say you’re 46-52% lower power, they’re definitely listening.

We created this thing called Letopia, a virtual city. We have 7 million urban subscribers, 3 million rural subscribers. We have all the connectivity. We use the best assumptions from the telcos. You look at Letopia now and you can see how you can run your ORAN at 50% lower power. That was a takeaway for pretty much every major telco. We also validated–Intel has the new Intel Edge Compute Platform. All the folks that came over from VMware created this edge orchestration. We won T Systems global data centers in Barcelona. Now you have one server in 10,000 locations instead of 10,000 servers in one location. You can go into a fast food chain, authenticate with your Motorola phone, and it’ll tell you if there’s a drive-through or a kiosk. Maybe you’re in an airport, so you don’t need the drive-through automation software. It’ll download everything from firmware, BIOS, all the way up to the containers. Then we hand it to the Intel Edge Compute Platform and they do all the application optimization. We did it with both Intel and with Rakuten. I think they’re the largest ORAN software company in the world.

They were giving me an overview. Close to 50% of the people that come to MWC are not in tech anymore. They’re verticals, business managers. If you went back to 1999 most of the audience would have been the carriers, the chip makers, the software providers, the systems manufacturers. More and more people are going because it’s becoming so multifaceted. It’s becoming a kind of combination of CES and the old 3GPP and Davos. Everything cooked together.

I can be so productive here, it’s unbelievable. You can meet Intel, AMD, Nvidia, VMware. I ran into the CEO of Deutsche Telekom. He met us, and then he was off to the Intel booth. The chip goes to the system, the system goes to Deutsche Telekom, and then you have the end user, who might be a car company, and they’re here too. You have all the layers of the ecosystem here, from chips to software to systems to telcos to end users. We’re all just rotating around each other’s booths.

VentureBeat: I wanted to drill in a bit on the smart cities. How widespread is this now? You have Barcelona and you’re moving to Madrid. Do you see that coming on strong?

Skaugen: Definitely. We’ve done Bogota, Colombia, for example. They were having a lot of car thefts. If they do a very simple algorithm that matches a license plate with a color and type of car, they can detect if someone’s swapped license plates. That cut down dramatically on car thefts. We’re doing every Kroger in the U.S. now, which isn’t smart city, but–for a long time these things were proofs of concept. You might do it in a couple stores. But now we’re in every single Kroger self-checkout. We took the same solution to every K-Mart in Australia. Their ROI was a month and a half. Theft went down 75% in the self-checkouts. These ROIs are compelling enough now that you’re starting to see widespread adoption.

We did this AI innovator program. Three years ago we invested $1.2 billion in AI, and we just invested another $1 billion. We had four global AI innovation centers. The problem these people have, you’d think they have massive IT departments, but really they might have five people to work on AI. There are 16,000 AI startups out there. We found 50 or so, the best ISVs. Sometimes we’d invested in the company. Then we’ve created 165 solutions. With Kroger and K-Mart it’s Everseen.

VentureBeat: For those grocery applications, self-checkout, what kind of knowledge do they get back?

Skaugen: They can’t really go to the cloud, because within 15 seconds the person will be outside the store. Of course, 85% of it is just mis-scans. That’s pretty easy. The camera that’s sitting above the self-checkout–the scales are 20 years old sometimes. People are getting sophisticated at stealing. They know exactly how much a can of Kool-Aid weighs. They’ll peel that label off and put it on a $55 cut of meat that weighs the same. The scale says it’s right, the Kool-Aid says it’s right, but the camera sees it’s meat. They can immediately detect that.

Then they have what they call “friending.” You’re scanning and scanning, and then your friend shows up and something “accidentally” doesn’t run over the scanner. Detecting the arm movement to see if something’s really going over the scanner. Detecting the latency in this kind of thing. They can check and see if there are things that weren’t scanned because they were on the bottom of the cart.

VentureBeat: Are there other benefits beyond theft control?

Skaugen: There’s the classic benefit to the consumer, giving them real time coupons when they’re doing the right thing. But now they’re moving it into the aisle. Apparently the number one thing stolen at Kroger is Tide pods. You can take these very large containers, split them up, and sell them at flea markets. They can take it out into the parking lot and watch for shopping carts running into cars. Who’s really at fault?

VentureBeat: There’s this invisible AI presence. People don’t realize it’s there.

Skaugen: Stores have always had security guards, but now–with CTRO they can do analytics, but it’s just using a stick figure. There’s a European spec to make sure there’s no identity recorded. You can see the person at the gas pump. You can see how many people go into the store, where they spend their time, how much money they spend. We were doing a pilot around how many EV chargers you should put in front of the store. The CFO needed to get the ROI. This person charged his car, walked in the store, and spent an average of X% more. You can determine the ROI on letting someone charge their car. How many stations were occupied, for how much time, and how much revenue did that lead to? How many people walk into a convenience store and go straight to the restroom without buying anything?

VentureBeat: I wanted to look at the division between the cloud and the edge when it comes to things like these AI PCs that everyone is talking about now. The benefit is supposed to be having your own private LLM run by your PC’s AI processing. Then your private data doesn’t have to leave your home. Are there things that need to go out into the infrastructure, though? Where is the dividing line?

Skaugen: We believe there will be three different kinds of generative AI. There will be public LLMs. There will be enterprise LLMs for data sovereignty and regulatory reasons. Some people have already committed to the cloud. Some people absolutely will not commit to the cloud. Then there will be private LLMs on your PC or phone.

The interesting thing is, when you ask a question, how does it become contextually aware? Are you asking a public question? Maybe you want to know the top 10 places people go on vacation. Maybe you want to know a place you should go on vacation this weekend. But then it needs to know your schedule. It needs to know where you are. It needs to know that you don’t like the beach around there because you’ve been there before and you hated it. It needs to know that you have to be in New York on Monday. Maybe you want to know a combination of the public response and something catered to your personal preferences and schedule.

Obviously we have to navigate a Microsoft ecosystem and a Google ecosystem. If you have personal AI in your phone and your PC, you don’t want to have two digital twins of yourself, one on each device. Somehow we have to ensure that gets done correctly. At the same time, we don’t want to overhype these things. With Centrino, at least you either had one or you didn’t. You knew that if you had Centrino and you walked into a Hilton or a Marriott, it would connect. The logo was in the lobby. The difference for me, now you have to start with, “What is an AI PC?” We should continue to evolve, but we’re not trying to say it’s all here today. There are multiple phases coming. They will get smarter and smarter.

Ultimately the digital twin makes you smarter, more relevant–my son is applying for colleges. He wanted me to go and find all of the pictures of him volunteering from the last five years. It took me 45 minutes in the taxi today trying to find these pictures. I had some basic knowledge of places to look. But in the future, an AI PC could do that for me and just give me the answer with a high degree of accuracy and index all of it. The usage models will be interesting.

We’re showing a couple of concepts, more old-fashioned innovation concepts here. We have a transparent display notebook. You can put items behind the screen and the camera will detect them. They’re imagining creators that might sit by the side of a river, put their notebook down, and sketch a bridge over the river on the transparent display, using a pen on the screen. We have a reverse foldable. I think we’re number one in foldables in the U.S. now. The reverse fold lets you wear it as a bracelet.

VentureBeat: The sovereign concept–if we have secure computer communications, do we really care where the data is? Things like sovereign enterprise data or sovereign personal data, keeping that data close to you versus keeping it in some kind of cloud. Why do you think the architecture should still be focused on keeping your data as local as possible?

Skaugen: Some of that’s going to be determined by governments regardless of what the technical reality is. You just acknowledged that. Where is the connectivity? Could the connectivity be removed? We’re very clear at Lenovo, given that we have 35 global factories–we need our factories to live regardless of whether the WAN goes down between any two countries. We need to be able to take orders and supply chain and all that sort of thing through resiliency. You’ve got that. Then you have the contextual awareness of what’s public, what’s enterprise, and what’s private. Also we think you should have transparency into the data sources that gave you the answer. Is your answer 80% public, 20% private? Is it 100% private or public? How do you create that algorithm? I don’t think we have it figured out yet, but certainly research is working on that.

VentureBeat: Is it a given that the AI processing belongs with the data?

Skaugen: We’re going to double the data in the world over the last three years. That’s more data than created in the history of the world. We’re throwing away 98% of it. We want to bring AI to the data instead of bringing data to the cloud.

VentureBeat: The cloud can’t store it all.

Skaugen: If you’re monitoring a door, whether the door has been open, because it’s supposed to be a secure door, you can throw away 99% of the data. Nothing’s happening. But when someone walks through it, then you’re going to want to process that. You’re going to want an immediate answer, a very specific answer. You may want to go to the cloud for that, or to your corporate database. Was that someone who was authorized to go in there? Same thing with worker safety in a warehouse. Are they in places they’re not supposed to be? I don’t think you want to run all that data to the cloud to get analyzed. You only want to drive it up there when there’s something interesting. Those are some of the characteristics.

We’re a bit unique versus our top two competitors in infrastructure, because we sell to eight of the top 10 clouds. If you’re our two major competitors, they don’t participate in the top 10 clouds in the world. If you look at my earnings announcements, I’m usually talking about 40-50% public cloud, 50% on-premises. We’re trying to be quite neutral in the hybrid world. We’re not saying everything has to be a service on-prem. If you want to go to a big cloud, we have several billion-dollar cloud customers. That’s okay with Lenovo, because we sell servers and storage to the cloud. We’re trying to think about it that way. To your point, I think it will be hybrid. But most important, you should have complete control of what data is going where, transparency on that. Also, we have an ethical AI committee to make sure there’s no bias in the products and the answers. You hear a lot about that. Sustainability and ethical AI.

VentureBeat: What about the sourcing of the data, whether it’s legitimately owned or licensed?

Skaugen: Exactly. What are the sources for the answers you’re getting? I asked a generative AI in a competitor’s booth what was the largest storage company in the world, and it said Lenovo. Okay, we’re not there yet! Very convincing answer, but we’re number three, not number one.

VentureBeat: With someone like Kroger, I guess it’s a no-brainer that AI and cameras go together, and there’s obvious ROI there. What are some things that are more on the frontier, where there’s promise, but you’re not sure the ROI is there yet?

Skaugen: We’re working with more than 50% of the fast food chains in the U.S. For example, if you cook a chicken sandwich and you’re waiting for the Uber Eats driver to pick it up and it’s been there long enough to become a food safety hazard–it’s almost like what you would have done with an unattended suitcase at an airport. AI at the edge is getting to a point where you can get this deployed in a container on a server you already own, you could address issues like that. We’re working with some Chinese chains to make sure they replenish their buffet. You don’t just get a worker to make a round every 30 minutes to replace the broccoli. They can have a camera to see if a football team came in and ate everything in five minutes.

We can do license plate detection. If you opt in, very critical, to be a frequent customer, they can now tell how many customers in line at the drive-through are going to order french fries. They can have your order on the screen before you arrive, because it’s Tuesday after your kid’s soccer game and you always order a Happy Meal, and your kid has a very particular customization because of a food allergy. You don’t have to talk through that every single time. I’ve probably spent 30 minutes in the Panera drive-through because I’m so particular about the sandwich I want. If I could just walk up there, it knows who I am, and my order pops up exactly as I want it every time? That seems so easy. Now we’re doing multiple languages in the drive-through, speech to text. That’s also an interesting use case.

You’ve probably seen some of the more far-out cases. We’re working with Grupo Pinsa, one of the larger Pacific fishing fleets. When the fish go into the net, they can look at the size. They have a better supply-demand picture, because they know exactly how many fish are in the tank. They know if the fish are too small and they can kick them off the boat right away. They can make sure they don’t have endangered species in the net, because they get fined. It’s almost like Procter and Gamble, talking about an HPC example on the Pringles line. If you had a black, burnt potato chip, it would shoot it off the assembly line. Now you have these things on fishing boats. With no connectivity out there it has to be an edge server. It might be three days before they come back to port. Then the marketing department already knows the supply-demand picture and can price according to what’s on the boat. That’s a super cool fringe example.

It took us more than a year, year and a half to get an oil rig safety spec. It took us more than a year to get the grease spec for some of the fast food chains, because these things are mounted on the wall above the fryer. They’ll kick up the fans. These edge servers–people were plugging their phones into the USB ports of the server in the store. Not exactly great for security. They have lockable bezels on them now. Our new edge server has a GPS on it. You can say, “Find my server.” If it moves at all, or if you remove the cover, it’ll take the keys off the drive. You can protect the data, even if you’re not in a data center with fences and security guards and biometrics. It’s just sitting on a factory floor or in the back of a kitchen. Those are some unique things happening on the hardware side.

VentureBeat: I’m curious about the whole industrial metaverse and digital twin initiatives. Things like Nvidia’s OmniVerse. They’re suggesting that we’re going to have some enormous ROI coming out of that.

Skaugen: I don’t know if you know this, but Satya and Jensen announced that Lenovo is the exclusive provider to the Nvidia OmniVerse. When you’re an end user, an Nvidia customer, you run on Lenovo hosted in the Azure cloud. As people are building these new EV factories and digital twinning the entire thing–whether you believe in 6G or 5G+ or just 5G finally hitting the mainstream, you see the adoption curves between now and 2028. The crossover is finally hitting between 4G and 5G. They’re doing to digital twin their entire tower network. We were on some of the largest wind turbine farms in the world with HPC, determining the turbulence going through one blade, where to place the next blade. Now you’re talking about how to get roads through 200, 300 wind turbines up on a mountaintop. These are major construction costs.

I was talking to one of the largest mining companies in India. I thought I had a hard job. They’re like, “There’s a mountain. We have 1,000 people living on it. We have to move them. We have to be environmentally friendly. We have to do erosion control. We have to dig a hole a couple of kilometers deep and get everything in there out.” Every one of these things is being digital twinned. I’m a big proponent right now. The good news is we’re leading in this one.

The coolest thing I’ve seen here walking around, which has nothing to do with Lenovo–there are so many robo-dogs. I’ve never seen so many. They’re unbelievable, the personality. They stop, look up at you, shake their heads. They can do backflips. They’ve started having a personality now. I don’t know if that’s good or bad. But it’s the first time I’ve seen robo-dogs with personality. People are walking them on leashes around the MWC show floor. I don’t know about the usefulness. I’ve seen them in Japan for years. But now they’re finally interesting.