In this analysis segment from AMD Advancing AI 2026 in San Francisco, theCUBE's Dave Vellante and John Furrier close out day one by unpacking AMD's transformation from a chip supplier into a full-stack AI systems company. Vellante traces the arc from the Zen microarchitecture breakthrough through $60 billion in acquisitions, including Xilinx and ZT Systems, that seeded AMD's ecosystem with software such as ROCm. He argues AMD doesn't need to dethrone NVIDIA — it needs to become the essential second source in AI infrastructure, giving enterprises negotiating leverage the way Amdahl once did against IBM mainframes. Furrier frames the shift toward inference, agentic workloads and economic benchmarks over pure technical benchmarks as the real story behind the AI factory conversation now dominating the mainstream.
The conversation also explores the rise of Databricks as a potential rival to frontier model providers, with Vellante detailing how its Genie agents, Unity Catalog and emerging "ontology" layer give it more of the software stack than most LLM vendors. Furrier and Vellante debate whether systems of intelligence, not raw model capability, represent the real enterprise moat, and how model routing between CPUs and GPUs is reshaping IT architecture. They contrast AMD's open, chiplet-based approach with NVIDIA's tightly integrated NVLink ecosystem, likening the dynamic to Android versus iPhone. From CapEx skepticism on Wall Street to the belief that infrastructure will ultimately give way to a software factory built on top, the hosts close by describing the real prize as an operating system for human work.
Forgot Password
Almost there!
We just sent you a verification email. Please verify your account to gain access to
Advancing AI 2026. If you don’t think you received an email check your
spam folder.
In order to sign in, enter the email address you used to registered for the event. Once completed, you will receive an email with a verification link. Open the link to automatically sign into the site.
Register for Advancing AI 2026
Please fill out the information below. You will receive an email with a verification link confirming your registration. Click the link to automatically sign into the site.
You’re almost there!
We just sent you a verification email. Please click the verification button in the email. Once your email address is verified, you will have full access to all event content for Advancing AI 2026.
I want my badge and interests to be visible to all attendees.
Checking this box will display your presense on the attendees list, view your profile and allow other attendees to contact you via 1-1 chat. Read the Privacy Policy. At any time, you can choose to disable this preference.
Select your Interests!
add
Upload your photo
Uploading..
OR
Connect via Twitter
Connect via Linkedin
EDIT PASSWORD
Share
Forgot Password
Almost there!
We just sent you a verification email. Please verify your account to gain access to
Advancing AI 2026. If you don’t think you received an email check your
spam folder.
In order to sign in, enter the email address you used to registered for the event. Once completed, you will receive an email with a verification link. Open the link to automatically sign into the site.
Sign in to gain access to Advancing AI 2026
Please sign in with LinkedIn to continue to Advancing AI 2026. Signing in with LinkedIn ensures a professional environment.
Are you sure you want to remove access rights for this user?
Details
Manage Access
email address
Community Invitation
theCUBE Insights Day 1
In this analysis segment from AMD Advancing AI 2026 in San Francisco, theCUBE's Dave Vellante and John Furrier close out day one by unpacking AMD's transformation from a chip supplier into a full-stack AI systems company. Vellante traces the arc from the Zen microarchitecture breakthrough through $60 billion in acquisitions, including Xilinx and ZT Systems, that seeded AMD's ecosystem with software such as ROCm. He argues AMD doesn't need to dethrone NVIDIA — it needs to become the essential second source in AI infrastructure, giving enterprises negotiating leverage the way Amdahl once did against IBM mainframes. Furrier frames the shift toward inference, agentic workloads and economic benchmarks over pure technical benchmarks as the real story behind the AI factory conversation now dominating the mainstream.
The conversation also explores the rise of Databricks as a potential rival to frontier model providers, with Vellante detailing how its Genie agents, Unity Catalog and emerging "ontology" layer give it more of the software stack than most LLM vendors. Furrier and Vellante debate whether systems of intelligence, not raw model capability, represent the real enterprise moat, and how model routing between CPUs and GPUs is reshaping IT architecture. They contrast AMD's open, chiplet-based approach with NVIDIA's tightly integrated NVLink ecosystem, likening the dynamic to Android versus iPhone. From CapEx skepticism on Wall Street to the belief that infrastructure will ultimately give way to a software factory built on top, the hosts close by describing the real prize as an operating system for human work.
In this analysis segment from AMD Advancing AI 2026 in San Francisco, theCUBE's Dave Vellante and John Furrier close out day one by unpacking AMD's transformation from a chip supplier into a full-stack AI systems company. Vellante traces the arc from the Zen microarchitecture breakthrough through $60 billion in acquisitions, including Xilinx and ZT Systems, that seeded AMD's ecosystem with software such as ROCm. He argues AMD doesn't need to dethrone NVIDIA — it needs to become the essential second source in AI infrastructure, giving enterprises negotiating l...Read more
exploreKeep Exploring
What is driving the recent industry shift toward inference-dominated workloads, agent-driven continuous compute, and economic (rather than purely technical) benchmarks — is it end-user adoption, the permanence of AI, or some other factor?add
Will a "software factory" emerge on top of the "AI factory"?add
What are your thoughts on the claim that Databricks (and similar data-platform companies like Snowflake) are better positioned than frontier model vendors to capture the AI software stack?add
What are the strategic and technical factors shaping AI infrastructure and enterprise systems of intelligence — including the roles of data/context, control planes, compute heterogeneity (NVIDIA vs. AMD), CapEx, and the distinction between creating intelligence and capturing value?add
Is NVIDIA overvalued, and how durable is its competitive advantage versus AMD (considering CUDA vs ROCm, the importance of libraries, integration/switching costs, and market positioning)?add
>> Welcome back everyone to theCUBE here live in San Francisco we're wrapping up day one of AMD Advancing AI event. All the top AMD execs are here, their ecosystem, the industry partners, customers, but also it's a free event. Practitioners and technologists are here admiring and wondering what will be happening here. And everybody seems to be obsessed with the latest smartest model, the next GPU, but after today I came away walking away with the real question. What kind of computer are we actually building here at AMD? Here to unpack it with me for the closing segment is Dave Vellante, my co -host all day and all week. Dave, what kind of computer are we building? Why is this AI conversation changing? Let's unpack this event, that first segment. is why is AI changing this country? And what kind of computer are we building here?
Dave Vellante
>> Well, first of all, John, we're here in Moscone West. We're upstairs. you got to believe that AMD scheduled this event, what, a year ago? And they picked July because, okay, not a lot going on in July. We'll do July. Not a lot of conferences. 12 ,000 people registered, right? 14,000 coming tomorrow.
John Furrier
>> 14 ,000 for the keynote.
Dave Vellante
>> Yeah, 14 ,000 people registered. Sorry. Lisa Su's keynote is tomorrow. It's kind of weird. This is day one, day two tomorrow. tomorrow, I don't know where they're going to put 14 ,000 people. They need a bigger boat, as you said earlier. And so we're seeing clearly the transformation of AMD. they've now reinvented themselves a couple of times. Famously, Jerry Sanders, the founder, said real men have fabs. Finally, they shed the fab, sold it to GlobalFoundries, and that started it. They brought back some key developers and engineers, and they built Zen. And the Zen microarchitecture set them on a path and it just sort of coincided with Intel's troubles. And so they won that x86 battle, they became an extremely viable and durable company. Then they can't take that same playbook and go after NVIDIA. NVIDIA's a different company, they're not wounded as Intel was. And so they had to compress 15 years of ecosystem development by NVIDIA into five years. And they did that with a really aggressive R &D and innovation roadmap. They spent $60 billion in M &A, $49 billion of which was on Xilinx. And they seeded the ecosystem with software like ROCm. And they put those pillars together. And now, to answer your question, they are now in the systems business. They're no longer in just the chips business. And it's a complete transformation and a reinvention. And my premise, John, is that they don't have to take out NVIDIA. they just have to become the essential second player, the second source in this AI wave.
John Furrier
>> You know the mainstream, Mark, we're lucky with theCUBE and SiliconANGLE Media and theCUBE Research to see things early. We saw inference three years ago, two years ago we were on the record when Groq came out of stealth, we covered them. They sold to NVIDIA for billions of dollars. But in the mainstream, the past three years, the conversation has been about training data, the big frontier models. Now in the mainstream here, we're hearing words we've been covering. AI factories, agentic AI, rack scale computing, systems. A lot's changed. Let's discuss the training becoming the infrastructure, inference dominating, agents creating continuous workloads, and economic benchmarks are greater than technical benchmarks. The world has changed. Let's unpack that. Inference is front and center. Hello compute. Training, GPU owns that. Agentic, hello? Desktop PC. Economics, show me the money. I don't want to see people spending tokens like drunken sailors. I want to get productivity gain. What is the driver in your opinion that's making the world wake up? Is it the end user adoption? Is it the fact that AI is here to stay forever?
Dave Vellante
>> Well I don't think training stops. I just think that inference explodes, and that opens up whole new opportunities for companies like AMD, who haven't been able to participate as aggressively as they would have liked in the AI wave, but they're certainly well ahead of, say, for instance, Intel. Broadcom's a different animal, but you've got NVIDIA sort of leading the charge. And then, look at it. The market is so big. We were debating Bubble. You called it TAM. The TAM is so enormous. Many, many, many trillions. all AMD has to do is if they can get mid -single digits, they're going to become even more valuable than they are now. They're an $800 billion company, but from a market cap standpoint, there's upside there. And so I think what's driving it is the rise of inference that's bottom-up. We were talking to Raul about the bottom -up productivity boom. It's happening from individuals. You know this. I know this. We're all using agents. We're joking about leaving our laptops open so the agents can keep running. and we're driving productivity. Our whole way we work has changed. For years we've been talking about new ways to work. Well we finally are realizing the beginnings of what that looks like, and that's what's driving this.
John Furrier
>> I'm seeing something that's been controversial here on theCUBE, and I've been seeing also at the NYSE WIRED program. In the enterprise, the model isn't the product. It's an ingredient, it's in an application for the user, and plus the general models, like Frontier models, like OpenAI, ChatGPT, Claude Code is good, but from an end user standpoint, it's great for the general internet. But when you deal with the company's data, look at the rise of Databricks, $188 billion company. Enterprises are realizing there's a lot of data that's not unlocked yet, and they got tons of process. They have intellectual property. So I think there's now general recognition that large language models will be general intelligence, specialty models will have specialized intelligence, and they're not mutually exclusive. We're hearing from AMD execs and the technologists, model routing. You know what that means? This model goes to that CPU. It's not that expensive. It's not a high priority, but we need to have a good enough answer. Don't use the expensive GPU for that. Go to the right AI resource. So you start to see the software intelligence coming in to support the end user bottoms up experience. And now that's an architectural, that was not IT before.
Dave Vellante
>> Well, what's going to be interesting is, will there emerge a software factory on top of the AI factory? And what I mean by that is, to your point, it's not just about the LLMs. And by the way, just as an aside, I think everybody else is claiming the death of LLMs. I don't see it that way. I think they've got plans. I have to believe, John, first of all, they're not capitulating on proprietary closed models. So to me, they have to figure out how to go deeper into the enterprise. Anthropic is obviously doing it. OpenAI is making moves there. But they both have to be able to provide the antidote to what Alex Karp was saying, is that they're going to steal all their alpha. They have to, through licenses and other terms, allow people to run those models and generate their own tokens. that has to happen or they will have real trouble. But will there be the emergence of a software factory running on top of the AI factory? That's where the value is going to be. The value in the CapEx and the core infrastructure in and of itself doesn't exist without applications on top of it. That's where the money is. And I firmly believe that's going to happen. It does take time. But what cycle have we ever seen where the massive investment in CapEx hasn't ultimately delivered a return? I think the reason why I think we are in a bubble is because I think that return is going to take the better part of 10 or 11 years.
John Furrier
>> Well, I wrote a post last night coming in on the plane that basically said this is a systems game, not about the chips, you mentioned that. A lot going on here, you're going to start to see end to end we heard from AMD. I want to get your thoughts on a post that Howie Xu wrote on LinkedIn two days ago. He said, Anthropic and OpenAI may be approaching trillion dollar valuations, but their enterprise moat may be weaker than it looks. Enterprise moat. Don't jump in yet. No, no, no, I respect Howie and his views. Microsoft's Satya Nadella says, paying too much for tokens, concern is legitimate, enterprise should be aware of what's going on. Alex Karp. And then he goes, this is Howie talking. But in parallel a shift is already underway. By the way, within the next few years, companies like Databricks may become even more valuable than Anthropic. Here's his reasons. Anthropic's API is less sticky than the Databricks platform. Databricks can replace and combine models as performance and prices change. We hear that from AMD. Databricks is deeper in the enterprise, understands the difference between the models that can route and do those things with AMD and others. Number two, the category he coined called budget frontier models. This is how he coined the term. Budget frontier models. Meanwhile, Grok 4.5 became available through xAI's API on July 8th, while Moonshot introduced Kimi K3 on July 16th. So his point is that combined with a third reason, recursive AI improvement, which is great for knowledge graphs, could commoditize even more of the model training. hinting at open weights, proprietary data, and customizing models. So, Databricks example, budget frontier models, cost, and computer science things like recursive techniques favor some of the new data structures that are powering this. His net is that the stickiness of a value of an enterprise to counter, or not counter, but either support or highlight Alex Karp's rant about ontologies and moat, it's also not a zero sum game, but there is a balance between frontier models generally, and then what an enterprise needs to do. And we heard budget costs, token costs are a massive problem. It costs more to run AI than humans, why run AI when humans are doing it? So that's got to come down, so he's right on the budget frontier. I like his data premise because if you can swap out models Well, Remember when you decoupled storage from compute?
Dave Vellante
>> Yeah, look what happened. You scaled independently. Snowflake, so -
John Furrier
>> Your thoughts on his premise,
Dave Vellante
>> because it speaks to what AMD's doing. Very thought provoking. First thing I would say is, the software stack is clearly transforming. I do see frontier models as a critical cognitive layer in that software stack. So I certainly don't see them going away. but to his point about Databricks, Databricks actually has more of the pieces of that software stack than any of the LLM vendors, the frontier model vendors. What do I mean? They have Genie, which, Genie Code agent and Genie One, essentially BI agent, so they have a user surface. They're building a context engine, which we call the Ontology, they call it Ontology, which is the system of intelligence. they have Unity Catalog, which is the governance catalog, they have the analytics engine, they've got connectors into virtually everything, all the systems of record, and they've essentially got an agent control framework. So in some respects, I would in many ways rather be Databricks than some of these frontier model vendors, because they have a very high bar to hurdle. Now, I will say this, I've often wondered, Okay, why doesn't Databricks go public? Well, they're in their Series J. Series M. They're on their way to Z. And I don't think Databricks is a controlled company like Meta or SpaceX, right? They're not a controlled company, meaning that no one shareholder owns more than 50 % of the stock. I don't think Ali Ghodsi owns more than 50%. They're venture -backed. So it's venture -backed. I think they have over 100 investors, Andreessen Horowitz and many others. So they're not a controlled company. So by staying private, Ali Ghodsi can make himself acquisition proof. Because if I were one of these frontier model vendors, I would have gone out, if they were public, I would have gone and offered, when they were at 100 billion, gone and bought, I would have gone out and bought Databricks for 200 billion, take them out, because they have all the pieces that the frontier models, in my view, need to compete. So I guess my point is, to Howie's comment, They're missing system intelligence. They have the user surface. They don't have the system of intelligence, that Ontology layer. They don't have the process knowledge. That's what Databricks is building. And so in many ways, his post there is quite prescient. I do think token costs are going to keep coming down. Right now, it's obviously a hot topic for the reason you mentioned. If humans are cheaper than tokens, I'll just use humans.
John Furrier
>> One of his VC friends made a comment. You're going to like this. The real moat may be the systems of intelligence, combining a rich context layer with unified transactional and analytical platform to power real time agentic feedback loops.
Dave Vellante
>> That person must be reading my stuff. Geoffrey Moore, really, I got to give George credit.
John Furrier
>> It's not George.
Dave Vellante
>> It's Geoff Moore, Geoffrey Moore, I'm just saying.
John Furrier
>> But you're starting to see the digerati weigh in. This is the tension between frontier model capture and enterprise value. The control plane, IBM's trying to do this, you saw their earnings, you're seeing the control plane, I've been joking on theCUBE, everybody's going for the control plane. If everyone's swimming out to the dock as fast as they can, who gets there first wins. But you can't have one control plane, you can't have 10 control plane vendors, or can you?
Dave Vellante
>> this is an enterprise dilemma, I think it slowed the market down. So the client surface is critical, and that's why I'm so impressed with what Ali Ghodsi and Matei and the team there have built out or are building out. But the interaction between the user surface, the client surface, whether it's a coding agent or a business user, has to interact with that context layer, that system of intelligence. And those two pieces will form a loop. Let me give you an example. When a human gets involved and makes some exception, you have to have a human in the loop, that knowledge, that tacit knowledge of the enterprise, the reasoning of which the human went through to get to that decision, now gets absorbed into the context layer. And so you have this ongoing cycle.
John Furrier
>> The question is that as NVIDIA and AMD we heard here today is end to end, much broader compute and GPU base, although NVIDIA is the GPU king of the hill right now, but there's a lot of use cases in the enterprise. Creating intelligence and capturing value are two different things. And my personal philosophy is, the moat is closer to the data. So wherever the data is, you got the moat. So if you're OpenAI and Anthropic, you got the data at scale, it's called the internet. So if you're an enterprise, you got to get closer to this, why we like Databricks. So to get that system of intelligence on top, you always skew to the data source, because you can always bring intelligence in. And so creating intelligence and capturing value has to be thought through at a very technical level.
Dave Vellante
>> And it's data, and now the data actually becomes the way in which workflows are coded. So process is the next sort of holy grail that really Palantir has been doing a phenomenal job of incorporating, and that's what pretty much everybody is going after, not just the data, but you've got to get that process knowledge in there, and that's why Databricks is connecting everything.
John Furrier
>> Well that's the word, the most popular word of the past 60 days has been context. What is context? Data. Well yeah, it's data and it's okay.
Dave Vellante
>> And what's around it. So here's a way to think about context. What analytic engines do and BI systems, they tell you, well what happened? What are the numbers? What are the metrics? What are the dimensions of the business? They tell you what happened. You can maybe infer why it happened. Oh, we lost a sales rep in the region, so our sales went down. It's going to take some time to ramp up. Okay, what happened? Why did it happen? What it's not really good at is what's going to happen next? What should we do next? What's likely to happen next? And that is the next wave of intelligence. That's what intelligence brings in. So, what should we do next? What's the next best action? That's where organizations are really running fast towards.
John Furrier
>> So the White House OSTP director, Michael Kratsios, said, quote, we have information that Moonshot AI distilled Anthropic's Claude for the development of the K3 model.
Dave Vellante
>> Because there was reporting that it did.
John Furrier
>> Well,OpenAI is taking the headshot. Well, we don't really have the full data, but that's the White House, that's a quote. It likely acquired GB300 equipped servers and accessed that data and trained in Thailand and trained the AI models. We'll see. But this is again, risk. Creating value and capturing, creating intelligence and capturing value is the key thing. We'll see how it goes. Meanwhile, Google, Alphabet reporting massive CapEx increases, negative free cash flow, by the way, of 5 .9 billion. Cloud computing has a backlog of 514 billion of contracted work that has not been recorded as revenue up from 460 billion in Q1, so that's some backlog that's going to come in as revenue. So, Alphabet stocks up. AMD.
John Furrier
>> No, no, after hours, not.
John Furrier
>> AMD signed with Anthropic, a server deal. Google stock's getting crushed right now.
Dave Vellante
>> It's down 3 % and they crushed earnings. The cloud revenue was up over 80 % and the stock's down three points in after hours. Why? Because they increased their capex outlook.
John Furrier
>> Okay, my take is Wall Street doesn't understand what's happening. IBM got taken to the woodshed for no reason. They missed some deals. People are prioritizing money and probably paid down other CapEx lock-in memory prices. Talk of a breakup was BS. they missed the quarter. By the way, they beat by 1%, but they just, because they had to change their game, they're getting taken to the woodshed. I think that's bad. I think that's a misread by Wall Street. People panicked. The mob went after IBM. IBM will probably get back up again. They were up and then they leveled out after hours. Google, Amazon, all announcing debt deals. NeoClouds are spending. The question is, is the CapEx legitimate, given the build out, when you compare the Telco 5G build out, roughly a trillion dollars, just on 5G. Just 5G, Dave, a trillion dollars. Okay, so if you look at all of telecom in the past 10, 15, 20 year span, it's in the trillions and trillions of dollars. That's just for wireless service. So for AI, why wouldn't I spend for essentially full online population use of AI?
Dave Vellante
>> Well, that's why Wall Street's freaking out about the hyperscalers now. They don't know how to model. Oracle's got negative cash flow. These cash flow machines are taking their cash flow instead of doing stock buybacks or dividends. They're investing in R &D. I love it. They're investing in CapEx and R &D. I love it because they see – I'm always critical of companies. Don't you have anything better to spend your money on than stock buybacks and dividends? To prop up your stock? Is that where you're at? You can't – there's no roadmap that you can invest in? And so, these guys are smart. the hyperscalers, Larry Ellison, they see an opportunity here, and they're going for it. They don't want to be left behind. This is either the biggest scam since the Dutch tulip craze, or this is going to be a massive windfall for—
John Furrier
>> Yeah, but everybody wants tulips.It's called AI. I want to get your thoughts on this to close out. In Paris I interviewed Mark Papermaster, the CTO, EVP here at AMD. He sat down on theCUBE with us. Homogeneous data centers are dead. No CIO could pick one vendor, one chip, one architecture. Agentic workloads demand CPU, GPU, networking and memory all optimized together at the rack level and also end to end. That's what I added, end to end. That's essentially his position. Heterogeneous wins. Your thoughts on AMD right now because I like what they have here. It's an alternative to NVIDIA. And I wrote in my post, we've seen both models. Both could work. NVIDIA and AMD have different approaches.
Dave Vellante
>> To quote Matt Baker, it's not a zero-sum game. I think NVIDIA is a remarkable company. I think AMD doesn't have to take out NVIDIA. It needs to be, and I think it's positioning itself as an essential second choice in AI infrastructure. Why is that important? John, you remember the old mainframe days? There was a company back in the day called Amdahl. They went out of business. But in the day when mainframes were everything, there weren't CIOs back then. They called them IT directors or VPs of IT. they would put an Amdahl coffee cup on their desk when the IBM sales rep walked in. Prominently, they'd be drinking out of the Amdahl cup. Oh, yeah, Amdahl was in here the other day. Why? Because it gave them better negotiating leverage. That was always the metaphor for negotiating leverage. So you want to have optionality. You want to have second source. You want to have negotiating leverage. So I see AMD as an essential second choice, second option, second leading company in the AI race. Are they going to catch NVIDIA? I don't think it's an Intel playbook all over again. And NVIDIA is not wounded the way Intel was. They don't have a foundry that was dragging them down the way Intel did. They didn't lose sight when the founders weren't there. NVIDIA is a founder-led company. I don't think they're going to, unless they trip up, that's a different story. But I don't see Jensen tripping up.
John Furrier
>> Do you think NVIDIA and AMD have competing philosophies? Different philosophies?
Dave Vellante
>> I think there are similarities and there are differences, and I'll share that. The similarities are that they're building systems. They're building rack-scale systems, not just silicon. They're building software. And so therein lies the similarity. And I think in that sense, Lisa Su recognizes we have to become a systems company. ZT Systems, ROCm, rack-scale Helios, et cetera. CPUs, GPUs, okay, that's a system. Where they are different, by design, is NVIDIA had the lead. They were able to go with, Mellanox, NVLink, which is a highly tightly integrated system. Tightly integrated, extreme co-design. Yes, call it proprietary. Of course, they use open protocols. Of course, their Spectrum-X is Ethernet based upon open protocols. But the whole system is the mother of all lock-ins. And so if you're a customer, you want to make sure if you're going to risk that high switching cost that you get value out of it. And I think plenty of customers are and will continue to do so. And that's the difference. If you are late to the market, which everybody was except NVIDIA, you have to have differentiation. How is AMD differentiating? It's with open source, the open marketing message, FPGAs, chiplets, which give them advantages. There's tradeoffs of chiplets, as you well know, right? There's the optionality and the engineering flexibility and the customer choice that you get from chiplets. There's also, at the very highest performance demands, you've got a synchronous memory operation versus a more asynchronous memory operation. So there's always tradeoffs in tech. So I think therein lies some of the similarities and the differences.
John Furrier
>> NVIDIA has been selling to the hyperscalers. Okay. So they're winning that. Is that a lockout or a lock-in for NVIDIA with AMD? If AMD is open.
Dave Vellante
>> At the end of the day, John, I think it comes down to who has the best price performance, the best performance per watt, the best price performance per watt per token. per token per watt in the business. And how that shakes out for training, I think NVIDIA's got the advantage, and inference. It's interesting. AMD has made some claims about performance of Venice relative to Vera. Now, they don't have Vera. NVIDIA doesn't have Venice. so they're making these claims from specs, published specs. But at the end of the day, you've got to believe a couple things. One, the fact that AMD acquired ZT for $5 billion is a signal that they saw that they had an opening in their flank, and they wanted to shut that down and close it, so that was why they announced Helios and the ZT acquisition. We'll see how that plays out. If I'm AMD, I'm marketing hard my openness card. I'm going to bang that drum, I'm going to pick that scab, and I'm going to force NVIDIA to defend it.
John Furrier
>> I think that what I like about this conversation here is that it's moving beyond the GPU. I like the systems conversation, but I think the ultimate choice between approaches between NVIDIA and AMD is going to come down to this. The winner will be whoever can, in the fastest way possible, redesign the entire computing experience around how humans actually think and work. And so that's to me the net net, because whoever gets there first, no one will care how it's done. I love my Apple, it's a closed integrated system. Some people like Android, two different approaches. In the enterprise, if AMD can use the ecosystem leverage that they have over NVIDIA, they could make a play to get there first. But it has to be redesigning the entire computing experience around how humans think and work. Because that is what changed the most.
Dave Vellante
>> This is really, I think, that's a really insightful comment. This is the debate I was having this morning on text with Sid Nag, who thinks NVIDIA is overvalued and screwed. I don't think they're overvalued. I don't agree. I think they're undervalued. I think that's my opinion. But his point was, look it, CUDA is not that big of a moat. Computer scientists know how to replicate it. My point was, well, okay, maybe. And ROCm is good software. It's getting some traction. But the real moat to me is the libraries. It's the maturity of the libraries. And I would liken it to Android and iPhone. Why do we buy iPhones? Because it's an integrated system. Who got there first? Apple.
Dave Vellante
>> Apple.
Dave Vellante
>> But also, the philosophies are different. It's an integrated system. It works. The ecosystem is mature. And I see that as a challenge for AMD. They've got to build up those libraries. And I think there's probably more do-it-yourself work that has to be done to replicate the NVIDIA. At the same time, you're trading off when you go with NVIDIA. You're increasing your switching costs. Whereas with AMD, in their Open Rack, your switching costs will be less, so there's less lock-in. Now, NVIDIA's positioning of that is, they call it a server farm. And they're trying to position it, that's marketing, it's what server farm versus AI factory. Yeah, we'll see how that plays out. Right now, my feeling is, everything that can be built will be sold, and that's going to be the case certainly through 2027, would be my prediction, and potentially beyond that. All right, so you look at the market share between Apple and Samsung. It's pretty close. Well, this was Sid's argument. Dave, there's way more Android smartphones out there than Apple. I'm like, yeah, and who's the number one company? It's Apple. They've got the most revenue. They've got the strongest ecosystem. They've got the highest valuation. Why? And so that, to me, is a reasonable analogy. I'd take either one, by the way. They're both great business models. I just think, look, NVIDIA throws off $119 billion a year in free cash flow. that's bigger than most companies on the planet.
John Furrier
>> Well, I'll salute you. This is an abbreviated theCUBE Pod here.
Dave Vellante
>> All right, John.
John Furrier
>> Wrapping up day one. Tomorrow, you've got Varun Chhabra coming on, SVP, Product Marketing at Dell Technologies.
Dave Vellante
>> Great lineup.
John Furrier
>> Soni from AMD, formerly Pensando, we covered. Omar from Meta. Darrick Horton from TensorWave, one of the hottest companies. I saw him out there today. Andrew Dieckmann, General Manager of AMD. Prith Banerjee, Cube alumni from 2013, formerly of Hewlett Packard, running the labs. Now he runs innovation at Synopsys. Trish Damkroger, who's the SVP GM at HPC and AI at HPE. Alistair Speirs from Azure Infrastructure and a lot more. So, we got tomorrow a big day on theCUBE. Final thoughts from today.
Dave Vellante
>> Well, hats off to AMD. First of all, thank you for this awesome set. The theater of presence is amazing. People have been walking up. They come up the escalator and they see theCUBE. So many people have stopped by. There's palpable excitement in this ecosystem right now. The partners, it's very clear the ecosystem wants an alternative. They don't want one company to completely dominate the chessboard. Okay. The second thing I would say is, again, the transformation of AMD has been quite impressive under the direction of Lisa Su. She came in. She was 44 years old in 2014. and she has helped take this company from a chip company and in a very short amount of time has compressed the cycle through focused R &D and through M &A and through ecosystem development has compressed a 15 -year lead that NVIDIA had down to five years and has positioned AMD to really be that essential second AI platform. So very impressive. Super psyched for tomorrow's keynotes and hearing Dr. Su lay it out.
John Furrier
>> Dave, thanks, great day. I'll just end by saying it's not the next chat bot, it's not the next AI model, it's actually, to use my favorite word, creating an operating system for human work, whether you're providing it to consumers or people at work or both. That's what we're seeing these companies do, is to get that infrastructure built so that the enablement and empowerment for the user, bringing them agency so they can get work done, use agents, be more productive, a boom is coming. We're doing our part here at theCUBE. That's a wrap for day one. I'm John Furrier with Dave Vellante. Thanks for watching.
