Advancing AI 2026 | Darrick Horton, Tensorwave

Clips
More from Advancing AI 2026

Darrick Horton

Founder & CEO

TensorWave

play_circle_outline TensorWave: Building an All‑AMD NeoCloud Driven by AMD Roadmap Confidence and Strategic Alignment

play_circle_outline Preference for general-purpose GPUs for flexibility amid evolving AI architectures

play_circle_outline Comparison to CoreWeave; positioning as AMD-focused NeoCloud competitor

play_circle_outline TensorWave Raises $350M at $1.55B Valuation, Eyes Series C as Helios/MI455X Plans 1–2GW, Hundreds of Thousands GPUs by 2027

Info

Darrick Horton, Tensorwave

Darrick Horton

Founder & CEO TensorWave

search

(INTRO)

Dave Vellante

>> Hi everybody, welcome back to San Francisco. My name is Dave Vellante. We're here at Moscone West, and you're watching theCUBE's live coverage of AMD's Advancing AI event, 2026. 14 ,000 people registered for this event. They tried to pack in, they did pack in for Lisa Su's keynote. Moscone West doesn't fit 14 ,000. They need a bigger boat here. Next year I'm sure we're going to be across the street at Moscone North and South, but sold out event, actually free event, next year I'm guessing they're going to charge because the demand is so high for the content and the innovations that are going on here. I'm excited to have Darrick Horton back at theCUBE. He's the co -founder and CEO of TensorWave. Darrick, good to see you again.

Darrick Horton

>> Great to be here, thanks for having me.

Dave Vellante

>> Yeah, you bet. So, TensorWave is the first all AMD, can I use the term NeoCloud? Okay, great. We're going to call you NeoCloud. That's an accepted term.

Darrick Horton

>> That's right.

Dave Vellante

>> Why all AMD? Why that choice?

Darrick Horton

>> Yeah, that is a great question. That really goes all the way back to the foundation of our company. We have decided to be AMD exclusive from the beginning as a matter of strategy and alignment. When we founded the company in 2023, at that time there was massive increase in demand for GPUs. ChatGPT was still pretty new. Everybody wanted more GPUs. But at the time, AMD hadn't released anything. The only solution on the market was NVIDIA. And NVIDIA didn't have enough. Nobody could get it. So we saw, my co-founders and I, we saw this massive demand for AI compute. We also saw that customers didn't really care whether it had an NVIDIA logo or not. They just needed it to work. They needed it to be available right now. And they needed it to be easy to use and relatively simple to plug into their workloads. And so we looked at that and we said, okay, we can be another NVIDIA cloud and have the same problems as everybody else. Not really add any net value to the world. or we could try to do something different, add some value, solve the problem from a different angle, and bring about some competition to the marketplace. Something that's a little more sustainable in the long run. And so we really analyzed what solutions would meet those criteria, what solutions could we bring to market that would actually be credible alternatives. Viable alternatives to NVIDIA at the time. And at the time, there was nothing. AMD hadn't released their first product yet. But we really analyzed the upcoming products, AMD's product, as well as Intel had some things at the time. And there were some various startups that were announcing products as well. And we felt that AMD was really the only credible challenger, the only group that had the right mix of expertise and skill and experience to actually develop and ship a product that could compete with NVIDIA at scale. And so across the board, that's what we saw. And so we decided to go all in with AMD for their first launch, MI300 in 2023, 2024. And we partnered extremely closely with them. Now, why have we stayed with AMD?

Dave Vellante

>> Well, before you go there, I'll follow up with that question. It's a good question, I'll ask it. But you guys had to, in 2023, take a little bit of a risk. You had to dig deep into the roadmap. I suppose having an executive like Lisa Su helped, they invested as well. And so that doesn't hurt either, but still, if they invest in you and they fail, that doesn't do you any good. Now you're stuck holding the bag. What kind of process did you go through that got you to roll the dice and give you confidence that this was going to, because you're right. NVIDIA, everybody knows the story there. Broadcom, you're going to build custom chips with them. You're not going to do that, I presume. They're going to spend their time on big whales like Google, ByteDance. So there really wasn't, there was some, we saw Andrew Feldman up on stage today, with Cerebras, but that's kind of top of the pyramid, not general purpose. So you had to have, you had to bet that it was a big enough TAM and they were going to execute. What kind of process did you have to go through, Darrick, to confirm that?

Darrick Horton

>> That's a great question. So, kind of filter through all the different options that were available globally. The first big question is, GPU or not GPU? That alone carves off a big chunk of the potential market. And we felt very strongly and still do that GPU was the way to go for our anchor. Because ultimately, for one key reason, which is the entire world changed essentially overnight with the release of Transformers. The T in ChatGPT. The fundamental technology that made LLMs possible was because of one paper released by a team at Google at the time, I believe. And that was really just an architectural shift in how we think about AI. And then everything from then to now is a direct consequence of that one architectural proposal, which means that at any time that can happen again, and our money is on, it will happen again several more times before we ultimately achieve the hypothetical AGI, the end state of this is how you build an AI model in the most efficient way. And so if that is your presupposition, you're assuming that everything we know about how to architect an LLM, an AI model, is going to change fundamentally three or four or five more times, you need to have hardware that is flexible enough to accommodate those changes. So if we go deploy a bunch of hardware that's hyper -optimized for transformers, and then tomorrow transformers are dead, that's not great. And we believe we're very far away from AGI, relatively. We have many years to go, So many foundational shifts in how we architect models before we achieve that end state. And for that reason, we picked general purpose. So that allowed us to say, okay, carve off everything that's not general purpose. You get rid of the Groqs and the Cerebras and a lot of the other startups. So then you're left with, at the time, AMD and Intel. Those were the two GPU providers that were on the table.

Dave Vellante

>> Yeah, so you reasoned through it. and then I suppose being a startup, get in line to get NVIDIA chips. You're going to be at the end of the line.

Dave Vellante

>> Yeah.

Dave Vellante

>> Okay, so...

Darrick Horton

>> And how is that differentiated. What value are we really bringing if we just get in line to provide NVIDIA chips?

Dave Vellante

>> Well, that just raises another question, but before we get there, which is, well, what is your differentiation? But I'll get there. Why did you stick with AMD?

Darrick Horton

>> Another great question. So part of the discussion and evaluation on choosing AMD in that final step over Intel at the time was alignment as well. AMD, just from an ethos perspective, we felt very aligned. AMD has this strong commitment to open source, strong commitment to community building, ecosystem development that we just resonated with as founders. So that definitely has an element. We have never liked the way that NVIDIA has done business in terms of the practices that they have. Monopolistic tendencies. And so having the right alignment, the right values and foundational alignment is part of why we decided to partner so closely and stay partnered with AMD.

Dave Vellante

>> What do you mean by monopolistic tendencies? Can you add some color to that?

Darrick Horton

>> Yes. There have been many examples over the years from NVIDIA. The Run:ai acquisition is one of them. There have been other examples. The Mellanox acquisition and the bundling of different products together to make it harder for others to compete. This is not going to be a net positive for the world at the end of the day. It's just going to slow down progress. And so we strongly prefer partners that are committed to open ecosystems, open solutions, instead of closed-source solutions. And so that drives a lot of our internal decision -making, and it drives a lot of who we choose to partner with. And AMD has always been extremely aligned with the open ecosystem approach and the open-source approach.

Dave Vellante

>> Wow, interesting thought. I would argue Mellanox and NVIDIA wasn't even, nobody was paying attention to NVIDIA when they bought Mellanox. So I wouldn't give that as much credence. But, they tried to buy Arm, and it almost happened. Can you imagine if NVIDIA owned Arm? I don't think that would be a good thing for the industry.

Dave Vellante

>> That's right.

Dave Vellante

>> The Run:ai acquisition, and I was thinking the other day, I'm like, Jensen's so nice, it seems so avuncular. But I wonder sometimes, I say to myself, well, are they bundling in a way that Microsoft, how they did it? And very clearly they are dominating, or trying to dominate the world. So I appreciate your perspectives on that. I don't think, maybe the Groq acquisition wouldn't have happened if Lina Khan were still in charge, but we digress. I interrupted you. Please carry on.

Darrick Horton

>> So I apologize for that. But just one more thing on this. Why did we stay with AMD? And this goes back to the strategy. We walked through why we selected them. And how we're aligned with them. At the end of the day, there's one more thing that we think is very, very important, which is offering the best experience to our customers. And how do you offer the best experience to your customers? How do you ensure that the product you bring to market is best in class? And what we believe the best way to do that is focus. Focus on doing what you do better than anyone else. And in order to do that, you need to not have distractions. There's a lot of groups that have started that maybe they were NVIDIA Clouds and now they're deploying a little bit of AMD, or they're deploying AMD and NVIDIA and Cerebras and all these other things. And what we see from those groups is they're not the best at anything that they do. And in this space, if you're not the best at what you do, the customer experience is going to suffer. It is non-trivial to build and operate these clusters at scale in a way that is reliable, in a way that is seamless for the end customers and ensure that they don't have to think about it. Infrastructure, you should not have to think about. You don't have to think about when you turn on the light switch, your light's coming on. You don't have to think about water coming out of the faucet when you turn it on. You shouldn't have to think about GPU infrastructure when you try to load your model or do inference. And if you're a subpar cloud provider, then your customers will have to think about you a lot. The best thing that I can hope for our customers is that they don't have to think about TensorWave.

Dave Vellante

>> So your differentiation stems from your focus and the differentiation is simplicity for your customers?

Darrick Horton

>> Simplicity is one element of it. Ease of use.

Darrick Horton

>> Tell me more.

Darrick Horton

>> Ease of use is so critical. You don't want your customers to have to think about you. AI is complicated.

Darrick Horton

>> Yeah, AI is complicated.

Darrick Horton

>> They want to work on building models. They want to work on doing AI. They don't want to worry about is power good. Or is my cooling system optimized for my GPUs? Or what are the optics doing in my network?

Dave Vellante

>> Are you saying your competitors force your customers to worry about that stuff?

Darrick Horton

>> Yes, they do.

Darrick Horton

>> Really?

Darrick Horton

>> They force them to worry about it because they're constantly having problems. And they're constantly having problems because they're trying to do too many things. You can ask the inverse question, which is not why are we focused on AMD, but why is CoreWeave focused on NVIDIA and not doing AMD? And when you analyze that question as well, you get the same answer. You get the same answer. CoreWeave is the biggest and arguably most popular NeoCloud in the space. And they are world renowned. They have the best reputation with their customers. Their customers represent them as being bulletproof, very reliable, easy to use. And why is that? It's because they have a relentless focus on doing what they do better than anyone else. What they do is NVIDIA GPUs. And they are first to market and have been driving a lot of that from the early days. And we are very similar in that regard, but on the AMD side.

Dave Vellante

>> Interesting. Interesting comparison.The focus. That is key.

Darrick Horton

>> Having the best customer experience is key to being successful long term. and as a consequence of that, right, by having the best experience, by making it seamless, you can actually end up charging more. The economics get better. Customers want to stick with a provider where things just work. They don't have to think about it. And they're willing to pay more for that.

Dave Vellante

>> It's a very mature thought process for a pretty young guy. I loved Chase Lochmiller and his team, but you're right. They started out building data centers. Now they're building, cloud services. I get it, there's massive demand, go for it. Yeah. But I often wonder, the world is asking, okay, when supply and demand sort of come into balance, whither the NeoClouds, they're not all going to survive, there's going to be some acquisition. Your comparison to CoreWeave is interesting, very focused. If you went down that same path, you were going to be second to them, now you have a way to be a leader in a segment. you may be narrowing down the segment, that's okay, it's a huge market.

Darrick Horton

>> That's right, that's right. We still believe AMD's going to continue to do extraordinarily well. And even has the ability to overtake NVIDIA, but that might be a hot take.

Dave Vellante

>> Well, based on what Lisa said today, that's their intention.

Dave Vellante

>> Exactly. I wrote a piece this weekend, I followed AMD for decades, and spent a week really doing a deep dive, and my conclusion was they don't have to beat NVIDIA, they just have to be an indispensable second AI platform. Based on the keynote today, they're trying to beat NVIDIA. They made a lot of claims.

Dave Vellante

>> That's right.

Dave Vellante

>> Personally, I don't think that's going to happen in the near term, it doesn't have to. They get 5 % of a multi -trillion dollar market, their revenues are going to double or triple, and you guys are going to be right there with them, making it easier to do AI.

Dave Vellante

>> That's right.

Dave Vellante

>> I love the premise, Darrick. As an observer, sometimes investor, I would invest in that company. It's a very, I think, thoughtful and cogent process. Give us the update on the company. How are you guys doing? Where are you with funding? Where are you getting demand? What can you share with us?

Darrick Horton

>> Everything is amazing. We closed our Series B a couple months ago. That was a $350 million raise, $1.55 billion valuation. We're working on our Series C now, undisclosed valuation, but it's going to be much bigger. Very excited about that. I think last time I was on your show, we were about to do our Series A. So we've come a long way since then. The team is growing.

Dave Vellante

>> That was December 2024.

Darrick Horton

>> That's right. We've come a long way. We're scaling rapidly. Lots of data centers, lots of people. I think in 2027, we'll probably deploy one to two gigawatts of Helios MI455X capacity. So we are extremely excited about that. That will represent a massive step function in scale for us going from today, tens of thousands of GPUs in production to end of 2027, hundreds of thousands.

Dave Vellante

>> Do you have a Helios?Do you have a Helios in-house?

Dave Vellante

>> We do have Helios.

Dave Vellante

>> You do? This is complicated stuff. You remember when GB200 racks started shipping, GPUs were melting. there was a 50% failure rate, people's data center floors were breaking.

Dave Vellante

>> Yeah.

Dave Vellante

>> Are you worried about that? Is that maybe some of the ice breaking that helps you?

Darrick Horton

>> It's a little bit of both. So we're not super worried. AMD had the benefit of learning from NVIDIA's mistakes. But also, that's what this time period is for. Production for Helios is probably January for us. Everything between now and then is testing and validation, working through whatever kinks may come up. Anything that we have access to between now and the end of the year is all pre-production.

Darrick Horton

>> Pre -production.

Darrick Horton

>> There's going to be some things to work out. But we're not super worried about it because AMD has put a lot of thought into the design and we've been in lockstep with them for the past year on Helios and what the design will look like. So, it is a big rack. It's heavy. It uses a lot of power. But we're not very worried about it from that perspective and we believe it will be quite successful.

Dave Vellante

>> Well Darrick, thank you for putting TensorWave into even better focus than I had had before. Congratulations and best of luck, love to have you back and track the progress.

Dave Vellante

>> Thank you, great to be here.

Dave Vellante

>> You bet. All right, and thank you for watching, this is Dave Vellante for theCUBE's live coverage, AMD's Advancing AI 2026. We'll be right back from Moscone West, right after this short break.

Darrick Horton, Tensorwave

search

(INTRO)

Dave Vellante

>> Hi everybody, welcome back to San Francisco. My name is Dave Vellante. We're here at Moscone West, and you're watching theCUBE's live coverage of AMD's Advancing AI event, 2026. 14 ,000 people registered for this event. They tried to pack in, they did pack in for Lisa Su's keynote. Moscone West doesn't fit 14 ,000. They need a bigger boat here. Next year I'm sure we're going to be across the street at Moscone North and South, but sold out event, actually free event, next year I'm guessing they're going to charge because the demand is so high for the content and the innovations that are going on here. I'm excited to have Darrick Horton back at theCUBE. He's the co -founder and CEO of TensorWave. Darrick, good to see you again.

Darrick Horton

>> Great to be here, thanks for having me.

Dave Vellante

>> Yeah, you bet. So, TensorWave is the first all AMD, can I use the term NeoCloud? Okay, great. We're going to call you NeoCloud. That's an accepted term.

Darrick Horton

>> That's right.

Dave Vellante

>> Why all AMD? Why that choice?

Darrick Horton

>> Yeah, that is a great question. That really goes all the way back to the foundation of our company. We have decided to be AMD exclusive from the beginning as a matter of strategy and alignment. When we founded the company in 2023, at that time there was massive increase in demand for GPUs. ChatGPT was still pretty new. Everybody wanted more GPUs. But at the time, AMD hadn't released anything. The only solution on the market was NVIDIA. And NVIDIA didn't have enough. Nobody could get it. So we saw, my co-founders and I, we saw this massive demand for AI compute. We also saw that customers didn't really care whether it had an NVIDIA logo or not. They just needed it to work. They needed it to be available right now. And they needed it to be easy to use and relatively simple to plug into their workloads. And so we looked at that and we said, okay, we can be another NVIDIA cloud and have the same problems as everybody else. Not really add any net value to the world. or we could try to do something different, add some value, solve the problem from a different angle, and bring about some competition to the marketplace. Something that's a little more sustainable in the long run. And so we really analyzed what solutions would meet those criteria, what solutions could we bring to market that would actually be credible alternatives. Viable alternatives to NVIDIA at the time. And at the time, there was nothing. AMD hadn't released their first product yet. But we really analyzed the upcoming products, AMD's product, as well as Intel had some things at the time. And there were some various startups that were announcing products as well. And we felt that AMD was really the only credible challenger, the only group that had the right mix of expertise and skill and experience to actually develop and ship a product that could compete with NVIDIA at scale. And so across the board, that's what we saw. And so we decided to go all in with AMD for their first launch, MI300 in 2023, 2024. And we partnered extremely closely with them. Now, why have we stayed with AMD?

Dave Vellante

>> Well, before you go there, I'll follow up with that question. It's a good question, I'll ask it. But you guys had to, in 2023, take a little bit of a risk. You had to dig deep into the roadmap. I suppose having an executive like Lisa Su helped, they invested as well. And so that doesn't hurt either, but still, if they invest in you and they fail, that doesn't do you any good. Now you're stuck holding the bag. What kind of process did you go through that got you to roll the dice and give you confidence that this was going to, because you're right. NVIDIA, everybody knows the story there. Broadcom, you're going to build custom chips with them. You're not going to do that, I presume. They're going to spend their time on big whales like Google, ByteDance. So there really wasn't, there was some, we saw Andrew Feldman up on stage today, with Cerebras, but that's kind of top of the pyramid, not general purpose. So you had to have, you had to bet that it was a big enough TAM and they were going to execute. What kind of process did you have to go through, Darrick, to confirm that?

Darrick Horton

>> That's a great question. So, kind of filter through all the different options that were available globally. The first big question is, GPU or not GPU? That alone carves off a big chunk of the potential market. And we felt very strongly and still do that GPU was the way to go for our anchor. Because ultimately, for one key reason, which is the entire world changed essentially overnight with the release of Transformers. The T in ChatGPT. The fundamental technology that made LLMs possible was because of one paper released by a team at Google at the time, I believe. And that was really just an architectural shift in how we think about AI. And then everything from then to now is a direct consequence of that one architectural proposal, which means that at any time that can happen again, and our money is on, it will happen again several more times before we ultimately achieve the hypothetical AGI, the end state of this is how you build an AI model in the most efficient way. And so if that is your presupposition, you're assuming that everything we know about how to architect an LLM, an AI model, is going to change fundamentally three or four or five more times, you need to have hardware that is flexible enough to accommodate those changes. So if we go deploy a bunch of hardware that's hyper -optimized for transformers, and then tomorrow transformers are dead, that's not great. And we believe we're very far away from AGI, relatively. We have many years to go, So many foundational shifts in how we architect models before we achieve that end state. And for that reason, we picked general purpose. So that allowed us to say, okay, carve off everything that's not general purpose. You get rid of the Groqs and the Cerebras and a lot of the other startups. So then you're left with, at the time, AMD and Intel. Those were the two GPU providers that were on the table.

Dave Vellante

>> Yeah, so you reasoned through it. and then I suppose being a startup, get in line to get NVIDIA chips. You're going to be at the end of the line.

Dave Vellante

>> Yeah.

Dave Vellante

>> Okay, so...

Darrick Horton

>> And how is that differentiated. What value are we really bringing if we just get in line to provide NVIDIA chips?

Dave Vellante

>> Well, that just raises another question, but before we get there, which is, well, what is your differentiation? But I'll get there. Why did you stick with AMD?

Darrick Horton

>> Another great question. So part of the discussion and evaluation on choosing AMD in that final step over Intel at the time was alignment as well. AMD, just from an ethos perspective, we felt very aligned. AMD has this strong commitment to open source, strong commitment to community building, ecosystem development that we just resonated with as founders. So that definitely has an element. We have never liked the way that NVIDIA has done business in terms of the practices that they have. Monopolistic tendencies. And so having the right alignment, the right values and foundational alignment is part of why we decided to partner so closely and stay partnered with AMD.

Dave Vellante

>> What do you mean by monopolistic tendencies? Can you add some color to that?

Darrick Horton

>> Yes. There have been many examples over the years from NVIDIA. The Run:ai acquisition is one of them. There have been other examples. The Mellanox acquisition and the bundling of different products together to make it harder for others to compete. This is not going to be a net positive for the world at the end of the day. It's just going to slow down progress. And so we strongly prefer partners that are committed to open ecosystems, open solutions, instead of closed-source solutions. And so that drives a lot of our internal decision -making, and it drives a lot of who we choose to partner with. And AMD has always been extremely aligned with the open ecosystem approach and the open-source approach.

Dave Vellante

>> Wow, interesting thought. I would argue Mellanox and NVIDIA wasn't even, nobody was paying attention to NVIDIA when they bought Mellanox. So I wouldn't give that as much credence. But, they tried to buy Arm, and it almost happened. Can you imagine if NVIDIA owned Arm? I don't think that would be a good thing for the industry.

Dave Vellante

>> That's right.

Dave Vellante

>> The Run:ai acquisition, and I was thinking the other day, I'm like, Jensen's so nice, it seems so avuncular. But I wonder sometimes, I say to myself, well, are they bundling in a way that Microsoft, how they did it? And very clearly they are dominating, or trying to dominate the world. So I appreciate your perspectives on that. I don't think, maybe the Groq acquisition wouldn't have happened if Lina Khan were still in charge, but we digress. I interrupted you. Please carry on.

Darrick Horton

>> So I apologize for that. But just one more thing on this. Why did we stay with AMD? And this goes back to the strategy. We walked through why we selected them. And how we're aligned with them. At the end of the day, there's one more thing that we think is very, very important, which is offering the best experience to our customers. And how do you offer the best experience to your customers? How do you ensure that the product you bring to market is best in class? And what we believe the best way to do that is focus. Focus on doing what you do better than anyone else. And in order to do that, you need to not have distractions. There's a lot of groups that have started that maybe they were NVIDIA Clouds and now they're deploying a little bit of AMD, or they're deploying AMD and NVIDIA and Cerebras and all these other things. And what we see from those groups is they're not the best at anything that they do. And in this space, if you're not the best at what you do, the customer experience is going to suffer. It is non-trivial to build and operate these clusters at scale in a way that is reliable, in a way that is seamless for the end customers and ensure that they don't have to think about it. Infrastructure, you should not have to think about. You don't have to think about when you turn on the light switch, your light's coming on. You don't have to think about water coming out of the faucet when you turn it on. You shouldn't have to think about GPU infrastructure when you try to load your model or do inference. And if you're a subpar cloud provider, then your customers will have to think about you a lot. The best thing that I can hope for our customers is that they don't have to think about TensorWave.

Dave Vellante

>> So your differentiation stems from your focus and the differentiation is simplicity for your customers?

Darrick Horton

>> Simplicity is one element of it. Ease of use.

Darrick Horton

>> Tell me more.

Darrick Horton

>> Ease of use is so critical. You don't want your customers to have to think about you. AI is complicated.

Darrick Horton

>> Yeah, AI is complicated.

Darrick Horton

>> They want to work on building models. They want to work on doing AI. They don't want to worry about is power good. Or is my cooling system optimized for my GPUs? Or what are the optics doing in my network?

Dave Vellante

>> Are you saying your competitors force your customers to worry about that stuff?

Darrick Horton

>> Yes, they do.

Darrick Horton

>> Really?

Darrick Horton

>> They force them to worry about it because they're constantly having problems. And they're constantly having problems because they're trying to do too many things. You can ask the inverse question, which is not why are we focused on AMD, but why is CoreWeave focused on NVIDIA and not doing AMD? And when you analyze that question as well, you get the same answer. You get the same answer. CoreWeave is the biggest and arguably most popular NeoCloud in the space. And they are world renowned. They have the best reputation with their customers. Their customers represent them as being bulletproof, very reliable, easy to use. And why is that? It's because they have a relentless focus on doing what they do better than anyone else. What they do is NVIDIA GPUs. And they are first to market and have been driving a lot of that from the early days. And we are very similar in that regard, but on the AMD side.

Dave Vellante

>> Interesting. Interesting comparison.The focus. That is key.

Darrick Horton

>> Having the best customer experience is key to being successful long term. and as a consequence of that, right, by having the best experience, by making it seamless, you can actually end up charging more. The economics get better. Customers want to stick with a provider where things just work. They don't have to think about it. And they're willing to pay more for that.

Dave Vellante

>> It's a very mature thought process for a pretty young guy. I loved Chase Lochmiller and his team, but you're right. They started out building data centers. Now they're building, cloud services. I get it, there's massive demand, go for it. Yeah. But I often wonder, the world is asking, okay, when supply and demand sort of come into balance, whither the NeoClouds, they're not all going to survive, there's going to be some acquisition. Your comparison to CoreWeave is interesting, very focused. If you went down that same path, you were going to be second to them, now you have a way to be a leader in a segment. you may be narrowing down the segment, that's okay, it's a huge market.

Darrick Horton

>> That's right, that's right. We still believe AMD's going to continue to do extraordinarily well. And even has the ability to overtake NVIDIA, but that might be a hot take.

Dave Vellante

>> Well, based on what Lisa said today, that's their intention.

Dave Vellante

>> Exactly. I wrote a piece this weekend, I followed AMD for decades, and spent a week really doing a deep dive, and my conclusion was they don't have to beat NVIDIA, they just have to be an indispensable second AI platform. Based on the keynote today, they're trying to beat NVIDIA. They made a lot of claims.

Dave Vellante

>> That's right.

Dave Vellante

>> Personally, I don't think that's going to happen in the near term, it doesn't have to. They get 5 % of a multi -trillion dollar market, their revenues are going to double or triple, and you guys are going to be right there with them, making it easier to do AI.

Dave Vellante

>> That's right.

Dave Vellante

>> I love the premise, Darrick. As an observer, sometimes investor, I would invest in that company. It's a very, I think, thoughtful and cogent process. Give us the update on the company. How are you guys doing? Where are you with funding? Where are you getting demand? What can you share with us?

Darrick Horton

>> Everything is amazing. We closed our Series B a couple months ago. That was a $350 million raise, $1.55 billion valuation. We're working on our Series C now, undisclosed valuation, but it's going to be much bigger. Very excited about that. I think last time I was on your show, we were about to do our Series A. So we've come a long way since then. The team is growing.

Dave Vellante

>> That was December 2024.

Darrick Horton

>> That's right. We've come a long way. We're scaling rapidly. Lots of data centers, lots of people. I think in 2027, we'll probably deploy one to two gigawatts of Helios MI455X capacity. So we are extremely excited about that. That will represent a massive step function in scale for us going from today, tens of thousands of GPUs in production to end of 2027, hundreds of thousands.

Dave Vellante

>> Do you have a Helios?Do you have a Helios in-house?

Dave Vellante

>> We do have Helios.

Dave Vellante

>> You do? This is complicated stuff. You remember when GB200 racks started shipping, GPUs were melting. there was a 50% failure rate, people's data center floors were breaking.

Dave Vellante

>> Yeah.

Dave Vellante

>> Are you worried about that? Is that maybe some of the ice breaking that helps you?

Darrick Horton

>> It's a little bit of both. So we're not super worried. AMD had the benefit of learning from NVIDIA's mistakes. But also, that's what this time period is for. Production for Helios is probably January for us. Everything between now and then is testing and validation, working through whatever kinks may come up. Anything that we have access to between now and the end of the year is all pre-production.

Darrick Horton

>> Pre -production.

Darrick Horton

>> There's going to be some things to work out. But we're not super worried about it because AMD has put a lot of thought into the design and we've been in lockstep with them for the past year on Helios and what the design will look like. So, it is a big rack. It's heavy. It uses a lot of power. But we're not very worried about it from that perspective and we believe it will be quite successful.

Dave Vellante

>> Well Darrick, thank you for putting TensorWave into even better focus than I had had before. Congratulations and best of luck, love to have you back and track the progress.

Dave Vellante

>> Thank you, great to be here.

Dave Vellante

>> You bet. All right, and thank you for watching, this is Dave Vellante for theCUBE's live coverage, AMD's Advancing AI 2026. We'll be right back from Moscone West, right after this short break.