Groq and Cohere CFOs: Can Artificial Intelligence Be Profitable?
Channel: Alex Kantrowitz
Published at: 2025-11-28
YouTube video id: -0t_NoQt6FM
Source: https://www.youtube.com/watch?v=-0t_NoQt6FM
What an amazing crowd we have here with us. We're going to put this on YouTube. So, I need you to make some noise. Let people know you're here. Let's go. My name is Alex Caneritz. I'm the host of Big Technology Podcast and I'm thrilled to be here for a discussion about whether there's ROI in AI with France Chad Chadwick and Simon Edwards, the CFO of Gro and Coher. Welcome. >> Thank you. >> Thank you, >> fellas. Let's start with an easy one. MIT came out with a study that said 95% of businesses trying to implement AI are not getting an ROI. Then a couple weeks later, Wharton, a similarly prestigious business school, came out with a study that said 74% of businesses are getting an ROI from AI. Both of those can't be right. So, who is closer to the truth, Simon? >> Well, look, I think I would say show of hands. Who here thinks your job cannot be disrupted by AI in the next 10 years? So I think I think everyone believes in the future of AI. I think everyone understands the use cases that that could become a reality. I think there's a lot of experimentation right now to find how those use cases can be applied in a in a practical way that scales for these enterprises. So I think it's not a case of if but when. Um, and I just think there's a lot of experimentation uh between now and that kind of end state. >> Okay, but I'm going to go back to you now. Which one do you think is closer to the truth, the MIT study or the Wharton study? 95% not getting an ROI, 74% getting an ROI. >> I think 74% are getting an ROI. I live in the Bay Area and I think I can't I rarely see a company in the Bay Area right now who is not applying AI to drive productivity for their engineers or drive productivity for their customer support reps. So I think a lot of these use cases are today demonstrating true value to you. >> I think what you got to do is you've got to look at the date stamp on both of those studies. The date stamp of the earlier study was where a lot of companies were looking at proof of concepts and they were trying something out and they may not have had the internal infrastructure to really take something from the proof of concept and build into something that would prove out the ROI. The second study, the Walton study, what you're seeing now is where companies have gone beyond proof of concept and they're actually putting things into production where they've built a process controls and more investment into the use of what can be built within AI. So obviously I'm going to say this, but very much on the Warton study, I think the timeline between the first and the second proves out the journey that the companies have been taking. >> Okay. Okay, but I want to ask you a followup on this Wharton study. I think it's very optimistic for AI. 74% of businesses say that they're getting an ROI from AI, but then you look at the uh different ways that they broke down the data. One thing stuck out to me. VPs and above are much more likely to say that they're getting an ROI from AI than managers and below. That to me just suggests that people in the sea suite are b drinking the Kool-Aid and they they are like of course we're getting an ROI from AI and then the people actually using this stuff and implementing the projects are telling the truth. So what do you think about that Franis? I I you've got in my mind I think you've got to actually look at the results and go back to the companies and sort of ask them you know we've you've built in production what are you seeing as the results and we are what we're seeing at cohhere with the companies we work with we we build AI tools and models for enterprises what we're seeing is they keep coming back to the table with more and more case studies in indeed a customer a large customer a global customer we have today reached out and said, "We've built all of these things. We built four or five different case studies in the last two weeks. They're proving the ROI, and they're going to continue building and building on top of that." So, I think it's it's a continuum. It's a journey. There may be differences in how people view the results, but um the journey is just one that has to be continued because otherwise companies are going to lose their competitive edge to their competition. And I think I would build on that by asking how do you measure ROI? If every single one of your employees you just pay for a chat GPT pro license, your costs went up, but where do you get leverage in your business? Is it you need less people as you scale? Is it you create more of a competitive advantage and therefore able to drive revenue growth at an accelerated rate? Is it that you can reduce your gross margins because you're more efficient with the way that your solution is deployed? Right? thinking about how you measure ROI is something that I think we're still in an early stage on. Um, but that's something we spend a lot of time thinking about with our customers. >> Yeah, the Wharton study was interesting because yeah, I should caveat 74% of companies that measure whether they get an ROI said they were getting an ROI. But that was only about 75% of companies that actually measure. Let's just take a quick poll. I'm curious in the audience if your company uses ROI. Doesn't have to be exact, but how many of you think you're getting an ROI today by show of hands? >> I think that scientifically was like 15% or so. Is that the reason you guys >> Sorry, ask the other side of the question. How many people don't think they're getting an ROI? >> Okay, let's let's do that. How many of you think you're not getting an ROI from AI today? >> I mean, maybe that's why you're here. So, we just have a shy audience. Is that what you're suggesting, France? >> Just saying. Yeah. >> More hands went up for the positive ROI. Okay. All right. We'll take it. We'll take that's that's good data for us. >> Yeah. >> Um on the Gro side, >> yeah, >> there was this moment, was it earlier this year? I can't believe it's been this year, uh that Deep Seek came out and seemed to drive the cost of reasoning models down uh by something like a 40th of what Open AI was uh was offering. And Grock's mission, of course, is to make the compute side of this much cheaper. And we hear all the time about how much the cost of GPUs are setting back these companies. Um why why do you think it's still a question now in terms of whether compute is going to be cheap enough to run the proof of concepts or even the inproduction applications that people in the audience are building and how does that get solved? >> Yeah. So for those of you who don't know uh Grock is a is a we design and manufacture uh silicone and chips designed specifically to run AI inference. So once those models have been trained those models run on AI on Gro's infrastructure and we do it not only at a a rate and a performance that is uh market leading at a cost that's also very differentiated. And so I think we're excited about these new open source models, Alex, because um they give optionality to a lot of our customers. I think the most important thing though is I think we're still seeing the industry is supply constrained. And so it's not a case of Nvidia or Grock or pick your chip manufacturer, but how do all of us deploy as much capacity as needed in order to serve all of these use cases? Because coming back to the the earlier point, there's a lot of experimentation. Everyone sees the future which is AI at scale. How do we serve all of those workloads and how do we serve them all equally? >> But is is compute the bottleneck or is it power? Uh Satya Nadella was recently on this very interesting podcast with Brad Gersonner. Sam Alman also joined. The thing that made the headlines was Sam Alman uh was asked about how he's going to fund 1.4 4 trillion in investment with 13 billion in revenue and he sort of had a fit. Uh but the other thing Satya said later was he has chips that are not plugged in and he wants to plug them in but power is the issue. So is it computer power that's the bottleneck? >> Yeah. I think I saw a statistic that says by 2028 in the United States a quarter of all of the energy produced will be used to run AI workloads. Right? Which is which is an amazing statistic. It's a lot of energy. >> Sorry. >> A lot of energy. >> A lot of energy. I think where we actually view this as a a competitive differentiation for Grock. Um our chips run um in a very price performant way. So in other words, we can serve tokens faster and at a lower cost than our competitors. And we think that's a really relevant use case given kind of the supply constraints you're seeing in power. >> Okay. Francois to you. Um, a lot of people here probably came wondering how they can get an ROI from AI. Cohhere is all about putting artificial intelligence into play in business use cases. What should they do? >> Well, what they should do is work with folks like Coher to build a strategy. >> You can't just say work with Coher. Let's give a real answer. It's you you've got to build a strategy as to where you want to deploy it across your company and and and to Simon's point, think about how you actually are going to measure the ROI because it's all about setting the stage up front and then bringing in the partners to help achieve those results. And so if if you just throw money at the AI question, you might not get the right results. you might not know if you've got the right results, you might not even know what your ROI ROI actually is. So, what we're seeing with a bunch of our large enterprise customers and and and even the government, right, both civil and defense, they're starting to build a framework and a strategy around what what they want to do, how they want to do it, when they want to do it, importantly, how much money they going to invest in this, and then how do they measure it at the back end. Mhm. >> It's we're beyond to my initial point, we're beyond just sort of saying, "Oh, let's build something and see what it does." I think there's still some experimentation that companies are willing to do, but they are spending much more time planning up front. Back to the Wharton study, they actually broke down uh the level of the size of companies and the ROI that they were getting. And maybe not surprisingly, the biggest companies 2 billion plus a year in revenue were getting an ROI from AI at 57%. So much lower than the 74% measured, the ones that were getting the most ROI, smaller companies. And so Franco, I wonder what you think about that in terms of does it tell us a lesson that the companies that are smaller and therefore they can adapt a little bit more easily, are they the ones in position to get the most bang for their buck when it comes to building AI today? And is there a lesson that all of us can take in terms of that adaptability aspect? Yeah, we all know and we all talk about the speed of AI and the AI adoption and what you're doing to build it into your enterprises. I think there's there is a true sort of nimleness that smaller companies are able to adopt to to sort of put in place their plan and then implement it across multiple different functions within a company to be able to get that greater ROI. the more that you can integrate agents and multi- aents across various functions within your companies, the greater ROI you're going to get. And I think that proves out at a smaller company. The larger companies, they've got more uh infrastructure that's been built for over a period of time. And so peeling that away to be able to then build your AI tools and your AI models internally just takes a little longer. And I think there's a real interesting nuance there, right? In that smaller companies are likely growing quicker and AI enables them to create leverage as they're growing versus larger companies already have that infrastructure that Francois mentioned. And so actually in a lot of use cases as those AI solutions are effective they result in the need to take out cost in order to truly realize the the ROI as opposed to just creating leverage from growth. So Simon, to follow up on that, when it comes to the ROI question, is it the fact that there's this conventional wisdom that AI costs so much to deploy that it's making the ROI picture difficult or is it processes? Is it is it actually the cost of deploying AI not as bad as the fact that new processes need to be developed? >> I think it's the latter. I think you know how many people in the audience right have been involved in these large ERP implementations or SAS rollouts that consume a ton of resources across the organization take a long time and then the the go live is you know 9 months in the future with these AI solutions you can get value immediately but you get value immediately and then have to go completely reconfigure your processes as an afterthought as opposed to design it in >> Google recently uh held a earnings conference call and Sudar Pachchai shouted out a number of companies that were using Gemini and generating many millions of tokens and then someone did the math and they said that's an incremental maybe like couple million dollars and uh this year we've seen so much money go into this field I mean if you think about just big tech's investment alone it was something like 300 billion in infrastructure buildout uh at the start of the year now it's in the 400 billion range if my numbers are are correct. Are we in a bubble? How does this make sense, Simon? >> This was the easy question you told us was coming, right? >> Um, look, I think as we stand here today and look back, no one regrets um the amount of money that was invested in railroads or the amount of money that was invested in fiber networks as part of the internet roll out. And so, I think 10 years from now, no one will look back and say, I wish we spent less money on AI infrastructure. But all those were bubbles. >> I'm sorry. >> Those were bubbles. The fiber was built in the internet bubble. >> Well, I think it's important to to define what bubble means in that context. And I think I think there's two dimensions to it. There's one which is just the infrastructure buildout. And then I think there's the second which is are valuations of private and public companies too high. Are we overleveraged in the economy? You know those I think hindsight will be 2020 in the future. I would say as it relates specifically to the valuation question, I believe that companies that are emerging right now, we will see the next batch of trillion dollar companies. And so I think a lot of venture investors and a lot of growth investors are are picking their bets. Um, but I don't think we'll ever regret the amount of infrastructure that we're building today. >> Franuis, what do you think about the bubble question? Yeah, I think there are potentially certain areas. When when we speak about AI, it's a two letters that cover a large sway of different different things. There may well be a bubble in certain areas and we got to be careful on some of that, but I don't think it applies to the whole industry. And even if there's a sort of a correction that comes, it may impact some companies, it's not going to impact every single company. But but wait a second because AI does go through this process where there's moments of development over exuberance and then a drawback of investment which is called AI winter. So I don't think necessarily that there will just be a continued buildout on the infrastructure that's built right now. We could end up very easily in an AI winter if if the marketing and the investment isn't going to make sense in a couple years. >> I'll come back to your first question. If companies can prove out that they have an ROI on their investment, there'll be many companies that will continue to grow and flourish as providers to those companies >> and the the buildout of infrastructure today is underwritten by the growth projections that these companies are seeing. >> And you believe in those growth projections? Well, I haven't seen their detailed growth projections, but I do believe that as just as we've all talked about AI will disrupt all of our jobs. I don't believe we've reached saturation of AI workloads within large enterprises. And I think there's a lot of head headroom to grow. Whether or not the companies that are investing that capital today will be the ones who are successful in the long term, there's a lot of execution between now and then. >> You know, I I always find it funny to be the one asking these like, come on, like where's where's the ROI? uh questions because at the end of the day the computer we're talking to the computers like sometimes it's easily to forget easy to forget that but there is so much talk and and marketing coming from the industry that you wonder you know how is this going to make sense but then again I was sitting with Dario Amod the CEO of anthropic a couple months ago and he said that the company did uh 1 billion a couple years ago and it's scheduled to do something close to 10 billion in run rate by the end of the year. So this 10xing continues to happen. All right, last question for both of you. Um, are you believers in AGI? Who wants to start? Frantois. >> Well, I I'll start with the cohhere slogan, right? Our slogan is ROI over AGI. As long as our business customers are getting an ROI, that is what we're focused on. >> I'm excited about the future. That's what I will say. I I'm I'm not necessarily the technologist in my company, but I'm excited about the ideas that we're we're a part of. >> Okay, let's hear for our panelists. Thank you everyone. You've been an amazing audience. Thank you. Thank you.