MWC25 | Shahid Ahmed, NTT

Clips
More from MWC25

Shahid Ahmed

Global Head Edge Services

NTT

play_circle_outline Importance of smaller language models for edge AI and industrial manufacturing use cases.

play_circle_outline Benefits of using mathematical-based AI models for time series data.

play_circle_outline Navigating China-US AI Competition & Open vs Proprietary Models: Implications for Market Innovation

play_circle_outline Intersection of politics and tech in 2025

Info
Transcript

Shahid Ahmed, NTT

Shahid Ahmed

Global Head Edge Services NTT

Join us for an insightful discussion with Shahid Ahmed, Executive Vice President of New Ventures and Innovation at NTT, as he delves into the dynamic landscape of small language models, Edge computing, and the ever-growing role of AI at MWC25 Barcelona. Hosted by theCUBE analysts Savannah Peterson and Dave Vellante, the conversation navigates the exciting intersections of technology and regulation, providing a fresh perspective on industry trends and innovations. Follow theCUBE's wall-to-wall event coverage https://siliconangle.com/events/.

Ahmed empha... Read more

explore Keep Exploring

What are the reasons for the importance of building a small language model compared to a large one? add

What are the advantages of using a mathematical based AI framework over a language based framework? add

What two things did the DeepSeek model prove? add

What do you hope to be able to say when we're hanging out at Mobile World Congress in 2026 that you can't yet say today? add

bolt Powered by CUBE AI

Shahid Ahmed, NTT

search

Savannah Peterson

>> Good afternoon, nerd fam, and welcome back to Barcelona, Spain. We're here coming to the conclusion of day one of our four days of coverage on theCUBE. My name's Savannah Peterson, joined with Dave Vellante. Dave, what is the coolest thing you've learned?

>> Day one. I feel like the week is ending. Holy cow.

Dave Vellante

>> Well, we've had enough great content today-

Savannah Peterson

>> Yes, definitely. - ...

Savannah Peterson

>> that I feel like we could have gotten a whole scope

Dave Vellante

>> of the show in just the short time we've been here, although so many fun stories to unpack, including with our next guest. Shahid, thank you so much for taking the time to hang out with us.

Shahid Ahmed

>> Always a pleasure. Thank you for having me.

>> I mean, you were hosting Dave last night at Happy Hour.

Savannah Peterson

>> Now you're here with us. We're going to be talking small language models. We're going to be talking Edge. There's a whole lot going on.

>> That's a good tradition you guys do on the

Dave Vellante

>> Verbena rooftop.

Shahid Ahmed

>> Yeah, it's Sunday evening right before the show begins.

Dave Vellante

>> It's become a sort of a classic meeting spot now.

Shahid Ahmed

>> Thank you. - It was a good crowd last night.

Shahid Ahmed

>> And it didn't rain, so we like that.

Dave Vellante

>> It didn't rain. - I was thinking

Savannah Peterson

>> about that with your rooftop.

Dave Vellante

>> I appreciate, I respect the courage that it took to do that. What's the most exciting thing you've seen here on day one?

Shahid Ahmed

>> It does feel like day four, by the way, so I kind of agree with you. Well, first and foremost, it feels like we got more people than last year.

>> It feels super buzzy.

Shahid Ahmed

>> I don't know if you guys have noticed this,

Savannah Peterson

>> but hard to get through from hall five to six to four. So I have a sense that there's more people. We'll get the numbers later from John Hoffman, CEO of GSMA, but it feels like the buzz is here around AI, the buzz around Europe versus US, regulate versus innovate. That debate is boiling. I was in a panel where they're talking about the role of regulations and how to fair share for telcos. My message was, "Go innovate guys. No one's stopping you to building those over-the- top applications. " So, look, I think there's a lot to talk about. I'm very excited about the show.

Dave Vellante

>> So, by inference, where are you on regulation, regulatory capture? You're like on the innovation side of the equation, right? Is that right? Am I inferring correctly?

Shahid Ahmed

>> A hundred percent. I think ... look, as the adage goes, Europe regulates, US innovates. I think that continues to happen. I think, US has taken the lead on AI, clearly. Of course there's China there too. We've seen that story play out the last six weeks, so we cannot ignore that. But in the end, I think the whole model around innovation and creating new opportunities, startups, Silicon Valley, you guys are very familiar with it.

>> Indeed. - I think, that's definitely

Shahid Ahmed

>> helping these new waves of technologies that we're beginning

Dave Vellante

>> to see and happen every other week.

Dave Vellante

>> So as we talked about earlier, two years ago, theCUBE research put out the ... it's actually two and a half years now, theCUBE, Gen. AI power law, and there we said, "Okay, you're going to have a lot of big LLMs, big players, but there's going to be a huge opportunity for smaller language models that are sovereign, more secure, bringing AI to the data on prem, and at the edge. " You've been talking about this for about the same amount of time. We've seen it. We've predicted it. It's now happening, isn't it?

Shahid Ahmed

>> It's awesome that you guys have been talking well before DeepSeek showed up to the scene. It became a family word. I think, look, for me, anybody can build a large language model. You got deep pockets, you can get the GPUs, you can build it. It's not a big deal. Big deal is to build a small language model.

Savannah Peterson

>> Why? - Because three big reasons.

Shahid Ahmed

>> One, you don't want to be consumed with large resources or compute power, which means-

>> Hundred percent. - ...

Shahid Ahmed

>> not only it's not sustainable, it's power-hungry

Savannah Peterson

>> and it's not economical. Almost 80% of the use cases, as an example in industrial manufacturing plants, are based on edge use cases.

>> 80%. - 80% of AI use cases are on the edge.

Shahid Ahmed

>> And so, intuitively, think about it. How many people in ...

Savannah Peterson

>> blue-collar workers are going to actually use ChatGPT, Perplexity, Anthropic on their laptop? Few.

Savannah Peterson

>> No, it's AI beyond the browser.

Shahid Ahmed

>> Right. - Yeah.

Shahid Ahmed

>> Yeah. But if you want

Savannah Peterson

>> to automate all your conveyor belts, your thermostats, your robotics, that requires a very edge-type service. It requires smaller language models. I'm only talking about 20 tokens, 30, 40 tokens versus-

>> A whole different experience.

Shahid Ahmed

>> Yeah. - Yeah.

Shahid Ahmed

>> And, to me, that's

Savannah Peterson

>> where the breakthroughs are going to happen.

Dave Vellante

>> And I'm so glad that DeepSeek, one of the side benefits of DeepSeek, besides bringing the whole economics down, is to bring to this idea that you don't need large language models.

Savannah Peterson

>> Correct. - You can garner the benefits

Shahid Ahmed

>> of AI from small models that can go on your cell phone-

>> A hundred percent. - ...

Shahid Ahmed

>> a little Raspberry Pi CPU.

Dave Vellante

>> And we've known this for a while, you predicted it,

Savannah Peterson

>> we predicted it, and yet the market responded by selling off Nvidia. So my question to you is, "Okay, you don't necessarily need these large language firms, let the big guys spend all that money, but you still need GPUs."

Shahid Ahmed

>> Look, I think, if you look at the software cycle over the years, even pick Microsoft operating system, even personal experience, you're like, "I want a bigger, better laptop with larger CPUs so I can do more things. " And so hardware, you'll always need a bigger CPU. There's no doubt about it. GPUs will play a part, Nvidia will play a part. Yeah, their stock dipped, but, by the way, has regained a lot of it. So, I mean, look, and the earnings was great as well.

>> Awesome. Yeah. - So hardware will always catch up

Shahid Ahmed

>> to software

Dave Vellante

>> and there's no doubt that you'll need bigger compute power. But I think there is, what was forgotten in all of this talk, is the idea of building smaller language models or just smaller AI models. Agentic AI, as an example, doesn't require big, huge, monolithic models anymore. You could just do it with a small 20- token model and be done with it.

Dave Vellante

>> The other thing is, if any market is not a zero- sum game, it's this one. So it's not as though if ... pick your vendor. Whether it's you AMD or AWS or Google or Microsoft, just because those guys are selling chips doesn't mean Nvidia's not going to sell. They're all going to sell chips, because there's different use cases. What are you seeing at the edge for that type of deployment, and where do you guys fit?

Shahid Ahmed

>> So we launched Edge AI, as a product and service, back in August, and we were telling the market-

>> We chatted back then.

Shahid Ahmed

>> Right. Right. And we said, "Look, we could do a lot

Dave Vellante

>> of things on the edge. You don't need a big, massive GPU infrastructure or data center to power these AI models. You could do that in a small little edge box that sits right next to an x-ray machine or a conveyor belt or a airport locomotive, AGVs, warehouses. It needs to be there , because it needs to take action right away. " So what Edge AI is basically looking at time series data as data is coming in, it does something with it, puts it in a nice model and takes action. That's the benefit of an Edge AI model. And I'm oversimplifying it, I know my engineers are going to kill me.

Savannah Peterson

>> You're not necessarily oversimplifying it, though. Just because you're not using a lot of acronyms and words and abandoning half of the audience who doesn't understand our lexicon, I think you're actually hitting the nail on the head. And I think what I like about what you're saying is you're essentially saying that this experience at the edge is going to be the way that most people interact with AI in a longer- term value creation way, which I think is fascinating. You mentioned the supply chain example. I'm very curious what else you're seeing in terms of edge, where the benefit is as obvious as it is with the 80% of the use cases there.

Shahid Ahmed

>> I'll give you a use case. And I think, David, we talked about this some time ago and, since then, we've matured on this use case. Basically, take a big large warehouse and typically you'll find hundreds, if not thousands, of thermostats regulating the temperature of this huge space. Sometimes these-

>> I bet they're doing it in here.

Shahid Ahmed

>> Yeah, think about it, right?

Savannah Peterson

>> Literally. It's great. Yeah. That's why I'm sweating.

Shahid Ahmed

>> And each section should have their thermostats, so

Savannah Peterson

>> that you can be comfortable. But in typical, in these large warehouses, factories, there's literally a person with a laptop, much like your mixer there with, it looks like a DJ, he's calibrating all the thermostats to get an ambient temperature of 70 degrees. Making that up. To do that, you have to regulate all these thermostats across the board. And literally this person is going from thermostat to thermostat, connecting to it, and being able to get a ambient temperature of 70.

>> Crazy. - Now, if you put that in a AI model,

Shahid Ahmed

>> it takes all the data from these thermostats,

Dave Vellante

>> understands it, and then regulates it. And you just say to your little UI model, just say, "Hey, put the temperature to 70 degrees. " It does the magic.

Savannah Peterson

>> And it probably does it more efficiently.

Shahid Ahmed

>> More efficiently. You don't need a person to do this. But it does that effectively and immediately, which is key, versus somebody going to hundreds of them and trying to manage it individually.

Savannah Peterson

>> Well, and you could probably ... sorry, I know you got something to say, but-

>> That's good. Go .

Savannah Peterson

>> And you could probably ... I mean,

Dave Vellante

>> I'm just thinking about this from a sustainability perspective. If you sense, you could know that there aren't people in the room, for example, so it doesn't need to be at 70. It can go a little bit up and down, depending on the machines or whatever's in there.

Shahid Ahmed

>> Yeah. In this case, by the way, one of the goals was for lower carbon footprint, energy footprint, sustainable goals. That's why they wanted to have a consistent standard temperature.

Savannah Peterson

>> Yeah. - It's funny you was joking

Dave Vellante

>> before, your engineers wouldn't like the way you described it, but you were actually talking tech, time series. That's streaming, that's data. You talked about action taking action. That's agentic. Everybody's talking about agentic. You're actually applying it in the real world. So I'm interested, because there's a lot of agent washing going on, a lot of AI washing. You're talking about real use cases in the field. This is real revenue that you're generating for you guys and your customers. So where are we? What's the journey look like? How real is this today? How meaningful is it?

Shahid Ahmed

>> So at Edge AI, in our view, is blue-collar, meat and potatoes use case. Things that people want and need in real time. I would love to have Google search replaced by my Perplexity browser. Great. But you know what? I don't even know if I need it. Maybe, I do. Who knows? I can build my itinerary when I go back to Europe in summer. Great. But these are use cases that could be used today, in today's world. We do believe, we call them blue collar use cases, but they're basically industrial manufacturing use cases that there are critical, mission-critical, applications. So look, for me, it's important to solve today's problem using today's tools, and that's what AI and Edge AI does.

Dave Vellante

>> We did a study, it was over 10 years ago now ... actually, GE asked us to take a look at it. Remember when they were trying to sort of become a software company? And I remember our analyst who quantified this, he came to me, he was like, "This thing is so big. " He goes, "It starts with a T, with a trillion. " I mean the market is massive, but, of course, it's very fragmented. So how do you deal with that sort of fragmentation from a business model standpoint?

>> It's an awesome question,

Shahid Ahmed

>> because there's so many AI models out there today. It's kind of crazy. You remember the old browser wars?

Dave Vellante

>> Yeah. - Oh, yeah.

Shahid Ahmed

>> Well, it feels like we're seeing this,

Savannah Peterson

>> right now, play out. Look, for us, it was very important to pick the right AI framework, and the one we looked at was mathematical based. Not to get too technical, but a mathematical based model is different than a language based model in that it can pivot very quickly to a variety of different use cases through mathematical permutations, like linear programming models, not to get too-

Dave Vellante

>> No, it's good. Linear programming, you use a matrix math.

>> Exactly. Thank you.

Shahid Ahmed

>> You can get nerdy with us. Don't worry.

Shahid Ahmed

>> Absolutely. And using those

Savannah Peterson

>> constructs are easier to not only program, but also apply in these industrial use cases that require time series data. So back in the linear program classes that you probably took, it was all time series.

Dave Vellante

>> And they require very high degrees of accuracy.

Shahid Ahmed

>> A hundred percent. - They can't hallucinate, right?

Shahid Ahmed

>> And mathematical models do not hallucinate.

Dave Vellante

>> Exactly. They're not- - They're very discreet.

Shahid Ahmed

>> It's discreet mathematics there.

Savannah Peterson

>> Yeah, you're either correct or you're not correct.

Dave Vellante

>> That's what we learned, back then. Right? Okay, so DeepSeek, good news, right?

Shahid Ahmed

>> I think so. - Okay. How about China-US?

Dave Vellante

>> You hear a lot of talk. You guys are a global company. I hear people say, "US must win the race for AI. " Not even quite sure what that means anymore. I mean, China clearly is a player here, but you know the US innovation model, very well. How do you think it's going to play out?

Shahid Ahmed

>> It's a loaded question, Dave. Always a good one from you. Look, I think the DeepSeek model proved two things. One, the economies have scale. I think Satya said it very eloquently, which is, look, it doesn't mean that you can ... just because you have something, economics go down, doesn't mean the demand is not there.

Dave Vellante

>> Jevons paradox. - Exactly.

Dave Vellante

>> And so, I think, that's one. Outcome.

Shahid Ahmed

>> The second one is that it's open source, which means you don't necessarily have to build a proprietary model, you can open source it. And, by the way, yes, DeepSeek has all the data. If you have that app on your iPhone, which I do, all that data goes to China. We know the sensitivities there. But, look, it's open source too. You can bring that data and make it your own, so it doesn't go to China. So there's a lot of nuances. People have to really understand the differences. It's not just necessarily ... just because it's made in China doesn't mean it has all the political attachments. So I think there is going to be a lot of China versus US. I think, to me, what is more important is to discuss open versus proprietary models. And I think that's where ... it's interesting that China would come up with an open source model. Isn't it? DeepSeek gets very high marks for that. I mean, IBM Granite, very open, didn't have quite the performance advantages. Although, very competitive performance, it just didn't get as much noise, because it wasn't a Chinese company.

Shahid Ahmed

>> Correct. - I'll give you my take.

Dave Vellante

>> I wonder if you could comment on it. You're not going to stop China from innovating. No way. Copying, innovating, whatever you want to call it, both. But the AI diffusion regulations, I don't know if you saw them.

Shahid Ahmed

>> Yeah. - They basically put countries into three buckets.

Dave Vellante

>> You're with us, you're against us, or you're in the middle. And it's that in the middle, it's like India, Israel, Switzerland, and it's so convoluted that those guys aren't going to be able to get access. And, to me, that's a mistake. Speaking as an American, the United States should be welcoming those countries in as opposed to pushing them away. They could push them to China. Because my premise is you need an ecosystem to win in AI, and if you don't have those countries as part of your ecosystem and China does, that's advantage for China. Your thoughts?

>> I mean, yeah, no, absolutely.

Shahid Ahmed

>> I think the whole idea of sovereign AI kind of sucks, to be honest with you. I'm like, "No, please don't bring politics into tech. Let's leave that alone. " I mean, DeepSeek proved it right on point, that just because you're Chinese, it proved the whole economic transformation on AI. Reset our thinking. And so, hey, what's the next ... it's not to say maybe somebody in France who are, by the way, the best physicists and mathematicians university-wise, they might come up with something very different.

Dave Vellante

>> I saw some benchmarks. DeepSeek, obviously, competitive. You had Llama in there, you had Anthropic in there, you had Google in there, et cetera, but the one company that is consistently innovating is OpenAI. We're talking large language models, and I know you're a big fan of small language models, because your space and your serve market, but I want to ask you something that we often ask informed guests is do you think that market gets commoditized, that LLM market? And some people would say it already is. But then there's OpenAI. They keep pushing the envelope, which is kind of interesting. John Furrier, whom you know, smart guy, says that the key's going to be innovation as to whether it gets commoditized. I know a lot of VCs say, "We've seen this model before. It's a hundred percent going to get commoditized." What do you think?

Shahid Ahmed

>> I agree, I think, with Furrier. I think, look, it will have to consolidate at some point. Masa Son from SoftBank, he is investing, I want to use the expletive here, but a lot of money. And is there room for another model like that? I don't know, frankly speaking. I think what we have is pretty good. I think there is some consolidation that will have to happen. I like the way OpenAI is also focusing on business-to- business use cases. That's a good advancement, much like Microsoft is. That's going to drive a lot of innovation, a lot of different types of use cases, and ultimately we'll figure out whether there's a room for another model out there.

>> And it's great for you because it just trickles out.

Dave Vellante

>> What Grok is doing, what Elon's doing with XAI, yeah, bring it on. If you want to spend all that money creating systems, open systems, that we can then learn from, that's what's going to happen. One of our earlier guests today called it ... he called it Doge for AI, which I thought was brilliant.

Savannah Peterson

>> No, yeah. That was great, in general. All right, Shahid, we could talk to you all day, but I have one final question for you. Since you're a CUBE alum, VIP here, what do you hope to be able to say when we're hanging out at Mobile World Congress in 2026 that you can't yet say today?

Shahid Ahmed

>> Awesome question. Well, to me, I think 2025 is going to be a pivotal year. Many of us know the political landscape is changing almost daily, if not weekly. Look, I think the tech cycle is also in full correlation with the political cycle that we're witnessing. The two things are almost identically revolving, so I think it'd be interesting to see the intersection of politics and tech that I don't think we've seen before.

Savannah Peterson

>> Agree. Well, we've certainly never seen an inauguration with the executives of the largest social networks on the planet standing next to the President of the United States. .

>> Right. And the cabinet members are way back.

Savannah Peterson

>> Yeah, exactly. In the nosebleeds.

Dave Vellante

>> And there's an asymmetry now in the media.

Shahid Ahmed

>> Where it was, everything was sustainability and DEI, that's been a big backlash, and so we hope to fill that void.

Savannah Peterson

>> Yeah, I was going to say we're still going to be-

>> We'll balance that assymetry. - ...

Savannah Peterson

>> sustainable and care about DEI. Shahid,

Dave Vellante

>> thank you so much for this segment. This has been awesome.

Shahid Ahmed

>> My pleasure. Thank you for having me.

Dave Vellante

>> Shahid, great to see you, man. Thank you.

>> It's great to see you too.

Shahid Ahmed

>> And thank you, Dave. Great questions on that one.

Savannah Peterson

>> That was awesome. And thank all of you for tuning in wherever you might be. We're here in Barcelona, Spain, at MWC, coming to the end of day one. My name's Savannah Peterson. You're watching theCUBE, the leading source for enterprise tech news.