Dell Technologies World 2025 | Ihab Tarazi, Dell Technologies & Autumn Moulder, Cohere

Clips
More from Dell Technologies World 2025

Ihab Tarazi

CTO & Senior Vice President of AI, Compute, Networking and Security

Dell Technologies

play_circle_outline Cohere partnership and enterprise deployment

play_circle_outline Exploring Cohere's Advanced AI Solutions: A Comparison with OpenAI's Strategy

play_circle_outline Use cases: sales assistant, support scenarios, and DNA sequencing

Info
Transcript

Ihab Tarazi, Dell Technologies & Autumn Moulder, Cohere

Ihab Tarazi

CTO & Senior Vice President of AI, Compute, Networking and Security Dell Technologies

Ihab Tarazi, senior vice president and chief technology officer of AI, compute and networking at Dell Technologies Inc., and Autumn Moulder, vice president of engineering at Cohere Inc., join theCUBE’s Savannah Peterson and Dave Vellante at Dell Technologies World 2025 to discuss the accelerating path from AI experimentation to full-scale enterprise deployment. Their conversation explores how Dell and Cohere are building secure, efficient AI solutions that meet the demands of modern businesses.

Moulder shares how Cohere is working to democratize AI acc... Read more

explore Keep Exploring

What updates can you provide on everything you've been up to for the last 12 months? add

What is the difference between an open AI strategy and a Cohere strategy? add

What are some of the exciting use cases that get you personally excited? add

bolt Powered by CUBE AI

Ihab Tarazi, Dell Technologies & Autumn Moulder, Cohere

search

>> Good afternoon, Dell fans and welcome back to fabulous Las Vegas, Nevada. We're here coming to the conclusion of our coverage of Dell Tech World. My name's Savannah Peterson here with Dave Vellante. We've got a CUBE on this one.

Dave Vellante

>> Yes, and this is, to me, one of the most interesting topics because it's all about enterprise AI and it's about leveraging your own data.

>> Yeah.

Dave Vellante

>> Right? That's why we wrote that piece, Why Jamie Dimon is Sam Altman's Biggest Competitor 'cause Jamie Dimon is never going to put his data, that's proprietary to him, up in the internet for LLMs to train.

>> Yeah, absolutely.

Dave Vellante

>> And that's where Cohere comes in.

>> It's interesting. On that note, welcome to the show, Ihab and Autumn. So nice to have you both.

Dave Vellante

>> Great to be here. Thank you.

>> It's awesome to be back.

Dave Vellante

>> Ihab, I feel like the entire world has changed since the last time you and I were on this stage a year ago. So much is happening. Give me a little bit of an update of everything you've been up to for the last 12 months.

>> I said the same thing today to a few people that the moves from last DTW till now are big deal, huge.

>> Nuts.

>> Having our biggest enterprise customers on stage, Lowe's, JPMC, and then all the announcements, getting to the point to announce this partnership with Cohere that means that we're going from talking and POCs to full-time deployment for hundreds of enterprise customers.

>> That's exciting. It genuinely feels like a whole new dawn. I heard an analogy yesterday that was awesome that one day in this AI land is the equivalent of one year in the before times and it feels very much like that. Autumn, give us a little bit of an overview of what you're working on with Cohere.

>> Yeah, at Cohere, we are really excited about this space and this time right now 'cause as you said, we're moving from a lot of POCs, a lot of experimentation into let's actually get this stuff into production. So at Cohere, we've been doing enterprise AI since day one, and we want to help move the needle from 3% to 5% of other enterprises adopting AI and getting into product to a 100%. We want to help them get there and we're right in the middle of doing that with secure efficient solutions that you could just reduce your time-to-value and get out there.

Dave Vellante

>> So let's stay on that for a moment because I mean I love foundation models, I love when people talk about AGI, but that's not Cohere. You're not trying to chase AGI. You're not running after that, let others do that. You said enterprise AI from day one. So explain the difference between an open AI strategy and a Cohere strategy.

>> Yeah, a Cohere strategy is hyper-focused on understanding what are the unique things about an enterprise, the controls, the governance, the security that you need to feel comfortable in deploying something into production. There's a host of problems when you move from just a consumer chatbot to I want to actually deploy an agent, I want to have something that has access to all of my data in production and can help my business users really get value out of that system. And that's where Cohere has been focused from day one as saying, "We take our foundation models, we bring them to you. You don't have to bring them to our system. You don't have to worry about bringing another company into your control and trust boundary."

Dave Vellante

>> Are the economics, I think they are, but I'll ask you, are the economics dramatically different? Jeff Clark talked yesterday morning that the cost per token is going down four orders of magnitude in four years.

Dave Vellante

>> That's a good fit.

Dave Vellante

>> So down 10,000 times, but the cost of training the foundation model is going up for 10,000 times in that period from what, 2019 to 2025, whatever that four or five-year period. Are the economics different in your world?

>> No. Yeah, the economics are very much focused on how do you build something that you can run efficiently, and that's one of the areas that we differentiate is because we're so focused on how do you build a model that you can run efficiently in production, but token cost keeps going down and as a user, you can maintain that control while still running GPUs that maybe you've purchased at a much cheaper cost. And having a model and an entire application stack that just really helps you run this highly efficient AI stack. Ihab, I see you giggling over there.

>> Yeah. Yeah. I would say also the economics are completely different for the enterprise customer. What we announced with Cohere, the entry level, you're able to get a model from Cohere with all the tools of 110 billion parameter model on one server and two H100s. And that is enough for a whole office and hundreds of employees, most likely. That's what Jeff was talking about. You don't need thousands of GPUs. You don't need a whole transformation where you're taking all your data and moving it to the cloud to say, "Let's get started." It's that simple. Wherever your data is and your tools, you can bring all the power of Cohere and the model and the tools and the way Dell does business with enterprise, with all this distributed compute, you can do it .

Dave Vellante

>> Do you see a difference in the market? Is it bifurcated? I'm sure it is, but I wonder if you could comment. Take that on the Global, I don't know, 100 or 500, they generally, the ones I talk to, big financial services organizations, they're building their own on-prem. It's like JPMC building their own on-prem AI stack. They want the latest and greatest Blackwells. But most enterprises, I sort of below the Global 500, 'cause I think it's, Global 500 is not even 2000, it's 500, 2000. They don't need, to your point, the latest and greatest. So do you see that? Do you see that market bifurcated and what's... Go ahead and I have a follow-up.

Dave Vellante

>> Yeah. For the most part, you have two sets of use cases, one about productivity and those you don't need to build your own model, you don't need the thousands of GPUs. And what we're doing with Cohere is very targeted to get quick ROI, change the behavior, change your customer service, make sales much more productive. You don't need to build your own. However, there's a second set of use cases about being either building a model to improve your market research or DNA analysis. So there's a whole bunch of use cases that are also happening, but you need both. So for the thousands of enterprises, they need the productivity now, but many of them need to be able to train models so they can get more competitive.

Dave Vellante

>> So how do you think about, Autumn, going from the 3% to 5% today to going to that 100%. We're in the world of parallelism. Does it happen in parallel with those big 500, they're going to go and then the long tail, thanks to your solutions, is going to be able to adopt more quickly or is it more sort of a sequence, a linear sequence of events where the big guys are going to prove it out on-prem and the long tail is going to wait to jump off the iceberg? How do you see that playing out?

Dave Vellante

>> I see it very much as it's in parallel because one of the great things about this technology is that we've reduced time to value. When you look at how far it's come so fast in the last few years, to your point, everything feels like it's accelerated, so a day is a year. So I think practically, from our perspective, in the industry, it's very much happening in parallel. And that's why so many companies need, they need these efficient solutions because you can't wait to try something out and let your costs balloon, and you need something that's simple enough to get to someone who is maybe tech-savvy, but doesn't code as a developer. They just want to use the models and start talking to achieve-

>> And important, the same big banks are also looking at Cohere.

Dave Vellante

>> Yes.

Dave Vellante

>> It's and, not an or.

>> Yes, it's not an or.

Dave Vellante

>> That's kind of my follow-up is because you guys are a part of this evolving ecosystem, it's critical for that 3% to 5% to go to 100%. When you work with Dell and do your testing and do all your lab stuff, how are you doing that? You doing the Blackwells for the big banks and then you're doing everything else and the AMD, whatever other, inferencing, doing both of those, are you picking spots? How are you preparing for this?

>> Yeah, it's very much you're doing all of them. And that's one of the values in how we've been able to reduce time-to-value for our customers because Cohere takes the burden of saying, "We support all these different silicon options and so wherever you are, as a customer, wherever you want to buy into now and make that investment, you can take the same stack and use it today."

>> Yeah. Maybe we should take a minute to explain what comes with that.

Dave Vellante

>> Please, yeah.

>> Yeah.

>> So this is the first time in the industry, and I believe with Aidan, I don't think there's something like it anywhere.

>> Nope.

>> It has a combination of data management tools, which is Compass, plus one of the most advanced models to aim specifically for these enterprise use cases. They're not chasing hundreds of consumer use cases at the same time. Plus advanced search, agentic and everything you need done, so anybody can create an agent. All one set of software on top of Dell compute with connectors to Dell storage, and we made the starting point to be just one server with two H100s. All of that on-prem to connect your data and keep it where it is. So that such a solution does not exist today.

Dave Vellante

>> It's Literally a turnkey offering to get all of those things. You don't have to cobble together the pieces yourself-

>> And that makes such a big difference when you're talking about that time-to-value ratio.

>> Yes.

>> Nobody wants to do that. And then you have to learn each one of those new tools too-

>> Exactly....

>> or figure that out in terms of what you're doing. It is pretty interesting. Ihab, You mentioned a few use cases a second ago, sequencing DNA and a few other things. What are some of the exciting use cases that get you personally excited? I'd love for you both to answer. Yeah, you go first, Ihab.

>> Yeah. I think the sales assistant is very exciting and I'd like Autumn to talk about it because they use it inside Cohere also. When you're a salesperson and you have to prepare for a meeting, it may take you days to look at all the history, who are these people we're going to meet with, what has the company done lately, have they hired people? That is all automated with this, automated connectors to every single transaction that you've done with that customer, all the orders, customer data, connecting to the web, all of these connectors already built into North, this platform. And within 10 minutes, as a salesperson, all you have to do is type three things, the name of the company and the two names or the three names of the people you're meeting. Ten minutes, you have your whole report.

Dave Vellante

>> I could tell you remember our demo really well with you and our sales team actually just showed him exactly what they do, and that's precisely what it is. And the great thing is the sales team was actually able to build this for themselves, so they didn't have to have developers go build this repeatable workflow. They were able to just type and say, "Go do this research," and then "Do do the next stage of research, do all of that."

>> And what I like about this example, which company is going to take all its billing data, all customer records, all communication, all chat and put it in the cloud just to get that answer. Not only is it hard, it's very high risk. So this data never leaves the office or never leaves the laptop in this case.

>> Exactly, which makes a really big difference. Is that the thing you're most excited about, Autumn.

Dave Vellante

>> I love the sales example. The other one that really gets me excited is support scenarios. So customer support, internal support between different departments, and just giving the power for teams that are very overburdened. They have their coworkers that are talking to them, they have customers talking to them and just giving them this power to say, "Hey, instead of handling 10 or 5 support requests and you're getting so burned out today, you can handle 50, you can handle 500 and give that same quality of response." But just giving them that power to control and share that information. And again, it's all with internal data.

>> Absolutely.

>> You have this history of interactions with different customers. You have the specialized policies and procedures around your team that you have access to because the system is deployed straight within your environment.

>> Yeah. The support for Dell is a very cool use case, and we have a demo here because we have thousands and thousands of customers. So it's very likely we found this problem somewhere else except it's highly confidential that you can't find that on the web. Being able to search all support communications and then being able to search all defects, everything we issued, all the versions to software and come back very quickly. Most of the times, when you have a support issue, the technician would spend the vast majority of their time trying to figure out what really happened, have we seen it somewhere else?,Have we solved it? And this is very secure private data.

>> Yeah, it makes a big difference. All right, you two, I have one final question for you. I'm actually very keen to hear what you have to say to this one. When we are at Dell Tech World 2026, what do you hope to be able to say then that you can't yet say today? Autumn, I'll start with you.

>> It's a great question. Dell Tech World 2026, what I want to say is we have 10,000 customers that are running in their private environments. They've moved past these POCs and we've been able to enable a lot of these companies to try that next step, that's it.

>> I love that. We're holding you to that 10,000.

>> 10,000, yeah.

>> That's great. What about you, Ihab?

>> What I would add to it as an extra challenge is those users are not software developers.

>> Ooh. Love that layer.

>> Day to day because that's how it's built. The agentic architecture and user interface we have with Cohere means all you have to do is know the business issue, not the technology.

Dave Vellante

>> So they are business creators/

>> Business creators.

Dave Vellante

>> Okay.

>> Exactly.

Dave Vellante

>> Love that.

>> I know. I love that. That was a good layer. Well, we look forward to talking to you about that next year. Thank you both so much for taking the time.

>> Thank you.

>> Thank you.

>> It was great to have you, Dave.

>> I always love coming here and talking to you.

Dave Vellante

>> It's our pleasure.

>> We love having you, Ihab. I got excited when I saw you milling behind the production desk earlier and then we both talked about how lovely you are.

>> Thank you.

>> So now the whole world knows.

>> Thank you.

>> And you're all lovely wherever you might be watching. We're here in Las Vegas, Nevada, Dell Tech World. My name's Savannah Peterson. You're watching theCUBE, the leading source for enterprise tech news.