John FurrierHello everyone. Welcome to theCUBE Pod episode 90 Dave and 90th episode. I'm John Furrier with Dave Vellante. Dave salute. Give the salute, the Dave salute.

Dave VellanteHow you doing, Johnny?

John FurrierGood to see you. It's been a great week. Oh my God, first, I was watching football on Sunday, Saturday night, my texts start blowing up. This DeepSeek has been taken over the internet and really so many storylines on this. I was commenting on this at a presentation I did in Silicon Valley this week. But DeepSeek has taken over the narrative and it's from China, and so it's a whole nother thing. And it really put everyone on notice around AI efficiency and also the geopolitical thing from China. Is it a psych ops thing from China? Is open source going to change the game? There's so many storylines. Is it a breakthrough. We're going to get into this, the efficiencies interview with the Chief AI officer from Cerebras. He weighed in on this. They have a clear line of sight. Amazon Web Services added DeepSeek R1.

Everyone's adding it in. Apparently it's the best in reasoning and doing some other efficiencies with chips. I called the chip grid. We'll get into that. But just a lot of action this week. I think we'll dedicate this pod to a deep dive on DeepSeek. So much angles and a lot of things we've been talking about. For the last two pods, we've been riffing on things like NVIDIA's mode and we were even going back in the history books and talking about many computer and proprietary operating systems. And remember those debates? It all played out. And then TikTok got banned for a day and then boom, that was a China discussion. And then China launched, in my opinion, it's salvo with DeepSeek. And I think it's the beginning of what we've been seeing as a cold war, but now an arms race. We're going to get deep into that. So if you really want to get our take on DeepSeek, there's a lot of threads.

Threads we've been covering. We've been covering as an analyst, but also commentating on speculating around fusion models, models working together with distillation, power law of AI models we put out two years ago. Those predictions still stand true. In fact, if anything it's validated more of our original vision on that. So, so much going on in this AI arms race, again under the hood, just some great stuff, but also some really interesting political conversations around DeepSeek. So Dave, we'll get into that. But first let's knock down some news because some other stuff is always going on. You had the Juniper lawsuit by the DOJ for blocking the HPE merger with Juniper Networks, which is ridiculous.

Dave VellanteIt's unbelievable. Really?

John FurrierAnd then IBM, Apple, Microsoft, Meta and Intel earnings. We'll get into that. But let's just start out with just quickly unpacking what I thought was a big story, which I thought HPE should respond when I saw there was a block of that merger, which seems like a no-brainer to me, but now the DOJ is suing HPE.

Dave VellanteAntonio responded. What kills me, John is the UK Competition Committee basically said, yeah, it's okay, do it. And then they're basically saying that, "Oh, that HPE and Cisco would be dominating the market for fixed wireless." And I'm like, "What?" Cisco is the dominant networking player. Antonio responded saying the DOJ is just narrowing the market definition way too much. I totally agree. I think this is really overreaching by the DOJ. And I don't know which DOJ is it? Is it like the Lina Khan leftover or is this Trump knuckleheads not understanding the market? Maybe it's because Antonio hasn't gone to Mar-a-Lago. I don't know. Maybe he's got to go kiss the ring.

John FurrierNo, it makes no sense. It makes no sense. ZS Carvalho wrote a post on this. Even if there's no market power here, and then people were chiming in about the WLAN market, wireless local area networking, which is Wi-Fi basically, stadiums and that kind of thing. But absolutely, I see no problem with this deal. To me, this is as uncompetitive takeover monopoly as you can see. HPE, even with Juniper, still has a huge competitor in Cisco. So if anything, Cisco is the big whale. So how does this work? I mean, it just doesn't make zero sense to me. This is, again, I think a display of incompetence on the government, just bad policy. And I watched some of the Senate hearings by the way, and you want to talk about just how there's no decorum anymore.

It's just forget qualifications, you want to smear people. So I think we're going to swing back to he original principles of like, "Hey, let's get information. Let's do the right thing and let people do their business."

Dave VellanteCan I just comment on the Senate hearings for a second? I watched a bunch of them too. And it's the same thing, whether it's a Democrat, a nominee or a Republican nominee. The other side just goes, "When was the last time you beat your wife? Yes or no? It's a simple yes or no question. When was the last time you beat your wife?" It's like, "Wait a minute. At one point RFK said, Senator, you just asked me six questions. Just let me answer one." "We don't have time. Yes or no."

John FurrierNo, no. When they try to defend and they get cut off, I got my soundbite in. No, it is basically a humiliation. It's a game of humiliation and they want to get the soundbites in. But if you look at the candidates from the Trump side, they're playing and we have a mandate. And they're going offensive right out of the gate. All of them I did. I watched all of them pretty much come out the gate knowing it was coming. But again, it just shows that the government, I think there's going to be a reset. I think people want to have a reset. I still think there'll be the bipartisan natural thing that happens, but the world was coming this direction. I've been saying to my kids, I've been saying as long as I can remember on theCUBE that there's going to be like hippie revolution in digital, in this culture where the young generation is going to say, "I've had enough. You boomers left me all this mess. Come on.

" And so you're starting to see that narrative and that Trump Traction. His approval rating in Democrats is off the charts and most of the young kids love him. So

Dave VellanteThere's a lot of detractors out there still.

John FurrierWell still, but you see a lot more of the people speaking out and it's less hiding. So you're starting to see younger people, just whether they're buying whatever narrative, it shows a different sentiment. And I think it is because of the obvious mistruths that have been represented on both sides. So it's not just Trump's side, it's just the other side too. So I think there's some things out there that young people just can't get their arms around. And I think my philosophy is just invent the future, create a new future. And if I was in my twenties, I'd be like a rebel too. I think I would probably rebel against the system and look to create change. Steve Jobs had that famous quote, "The world we live in was created by people that were no smarter than you." And I think that applies to government and I think it's going to be, not in a Kennedy way, but I think it's still not yet formed. But you might see that. That's what I'm just saying. So my point-

Dave VellanteWe'll see what happens. It could be a lot of unintended consequences of shifting responsibilities or cutting. My fear is that they're cutting stuff for people that are protecting Americans, whether it's cybersecurity or critical infrastructure, or hopefully not terrorism. I don't know. It's just, like you said, Trump's on the offensive. He blamed DEI for the crash, the helicopter and airplane collision. I'm like, "Oh, okay, maybe." But I don't think so.

John FurrierThe analysis of that is actually fascinating because people in aviation are very technical and they weigh in. It's a combination of system issues and also just common-sense flight path. It's just-

Dave VellanteHuman error.

John FurrierIt's just a really, really tragic thing. But it could have been avoided. That airport, I fly into that all the time for DCF because it's a shorter <inaudible>.

Dave VellanteMe too. Me too. It's so convenient. But it's like maybe we'll be flying into Dulles.

John FurrierI'm like, maybe I could go to Dulles now. Because the traffic there, I mean it is so busy. I don't know why they're running training runs on Blackhawks. It makes no sense to me. They could probably do it somewhere else. And they're all wearing night-goggle visions, by the way, in that too, and it's lit up heavily. So it's like that's a whole nother thing. I don't want to get into that, but I'm scratching <inaudible>.

Dave Vellante<inaudible>.

John FurrierAgain, DOJ government inefficiencies. We'll put that aside. I'm for HPE buying Juniper. I think that's a good move for HPE makes more Cisco's probably cheering. But other than that, that just again makes no sense.

Dave VellanteBut here's the thing, it was a year ago. You remember we had the CEO of HPE, the CEO of Juniper on at MWC last year, which was what? February? It's been a year. Come on. Why does it take a year if they freeze? They make all these plans. Really, you can't figure this stuff out inside of a year. Anyway, I'm very much against it. I don't really understand it.

John FurrierBefore we get into the DeepSeek stuff, I want to just hit on these key earnings. Because we like to watch the bellwethers, IBM, Apple, Microsoft, Meta, Intel, all released Dynatrace as well. We covered that because they're in our observability area <inaudible> great earnings. IBM.

Dave VellanteIBM guidance was really strong. When I saw DeepSeek come out, I'm like, "Ah, a little bit like Granite." So I thought that was a nice little boost. Not that that's why the stock was up. It's up because they gave good guidance.

John FurrierAnd I think IBM just has fundamentals. We saw Arvind Krishna, we've been to the analyst meetings, we cover them. We've had one-on-ones with the executives. IBM has been really focused on being that integrator and some picks and shovels in the AI wave, and they got the install-based customers. I think we've mentioned this on the pod how many times, but in AI and this transformation, you always say the line is you win in transformation in transitions. So this is IBM in a transition with the industry market shift. They have a lot of customers and they have a lot in their portfolio. So they sharpened the saw. And they did, they got good results. And maybe that fruit is coming off the tree from Arvind's strategy. And again, he's pretty meticulous about what he's trying to do. Clearly the long game. He said to us many times, "We're playing along." The big open item for me is the developer.

So IBM has the install base. Startups are trying to sell into that install base. IBM has an opportunity to bring that developer go-to-market piece in, as does the Dells of the world. And so I think that the big incumbents, Dave, are shifting from trying to win the developers, the code on their platform to coding into the stack of their systems and their transformational install base. Because people aren't changing out the tools and platforms. They're changing process, not so much changing things. And you see in the abstractions come in. So I think these incumbents like Salesforce, like IBM aren't going anywhere. They got the data, they got the install base, they got the beachhead, they got the position. They're going to be brokering access to new technologies.

So I think there's going to be some durable incumbents. Andy Jassy used to call the incumbents the old guard. I think it's a good opportunity for them. So I'm bullish on it.

Dave VellanteAnd the software piece is what excited me. Looking back, the Red Hat acquisition, I always thought that the Red Hat acquisition made sense from an IRR standpoint. Stu and I were in here. Stu of course works for Red Hat. During COVID. We did an analysis of the 34 billion. And at the time we said it's really not a cloud play. Maybe eventually it could be, but it's really a play about leveraging Red Hat and OpenShift in their consulting business. And that's how this thing's going to get paid for. I criticized Ginny a lot, but this is on her way out. She let Arvin do that. Not that I said, but Arvin was going to be in control anyway. But that was a good acquisition. The numbers made sense. They paid for it very quickly because of the consulting business. And now the software side is really paying dividends. So 34 billion, a lot of people criticize that acquisition. In my opinion, it was one of the best in IBM's history. Maybe will turn out even better than PwC, which essentially saved IBM back in the Gerstner days.

John FurrierYeah, IBM's doing great. Apple had earnings, Meta had earnings. It was a leaked to all hands. Zuckerberg was saying, "Every time I give them all hands, everything gets leaked to the media." So it's a bummer

Dave VellanteComment on Apple. We've been saying for how long everybody ebbs and flows on Apple. Down on Apple, they've got such a huge install base and their services business is the marginal economics or software economics. Maybe even better. It's amazing. And that's what's driving their business. Their install base is growing. Maybe iPhone sales are suffering a little bit because of China, but the install base I think grew by over a hundred million, I saw. That's incredible from the last time they reported it. So that's the service revenue. And whether it's ear pods and billions and billions of dollars on products that or are relatively low cost and loved.

John FurrierYeah, I think there's going to be a big, big focus on how the impact of this AI shift happening quickly to the big guys. So we're going to keep watching that. We'll hear our comments on the pod.

Dave VellanteAnd I heard a stat that the iPhones that did Apple Intelligence actually did better. I don't know if that's true. I got to dig into that. But if that's true, I don't have the latest and greatest iPhone. Mine's a couple generations old. And so I'm one of the reasons that people say, "Oh, Apple's in trouble." because people aren't upgrading their phones as often. But if Apple Intelligence is that good, I'm all in.

John FurrierWell, my family wasn't giving it high marks.

Dave VellanteYeah, that means it's a Siri 2.0, so we'll see.

John FurrierAll right, well let's get into the deep dive of this DeepSeek situation.

Dave VellanteOh, couple other things. Microsoft earnings, so guidance and spending with the concerns there. Azure missed my forecast. Intel beat, and everybody's excited about Intel, again, the stock was up, but the revenues declined 7%. And ServiceNow, surprised to the downside. I was surprised to see that tepid guidance and they missed on revenue. So yeah, big earnings week, a lot of news. And then OpenAI, my gosh, we're going to talk about that I presume when we're going to get into DeepSeek.

John FurrierYeah. Well what's your take on those earnings based on those numbers? What's your analysis?

Dave VellanteIt's going to be interesting. It's a little mixed. This happens when you go through these industry transitions. It gets very confusing. You have some companies are doing really well, other companies are doing okay. Those with the big install base. Sometimes they struggle. IBM bucked that trend. Microsoft was kicking butt. So you have all these cross currents. And it's very chaotic right now. But in general, I'm optimistic. And I think, we'll get into DeepSeek, I think that we're going to learn from DeepSeek how to do things more efficiently in AI. And that's going to be a rising tide that lifts all ships finally. Because AI hasn't lifted all ships as we know, it's been narrow and focused in terms of those that have really gotten benefit from it. I think to the extent that we can lower the denominator and we're going to talk about that, the value goes through the roof.

John FurrierYeah, I think anyone who's in this game has invested the CapEx or the infrastructure, the levers are infrastructure and software. And I think the CapEx, whether it's data centers, AI systems that's continuing to thunder along and people are trying to figure out how to put production workloads on them. And it's the software innovations, I think that's going to be key. And what got my attention with this DeepSeek was there's a lot of salacious headlines, but at the end of the day, it was the innovation behind the model and how they did it was interesting. First, the thing that got everyone's attention was the cost claim. They had outstanding numbers. And this unbelievable, "Oh my God, 5 to 6 million dollars. Oh the costs." that cost hundreds of millions of dollars for OpenAI. That was the first strike of the play, which by the way, NVIDIA stock dropped on Monday.

Largest loss in the history of the market. If someone shorted that thing, they made tons of bank. Now, by the way, High-Flyer was the company behind DeepSeek, a hedge fund, but spun out DeepSeek. And that to me could be suspicious. But again, the top stories were the training costs, that was debunked, the efficiency behind the model. How they used chips and how they did some precision work FPG&As and other chips, not just GPUs, although they had GPUs. We'll get into that. And then the broader narrative around the democratization and the commoditization of models and the CCP-linked involvement. Those were huge storylines. So that was the table. So DeepSeek announced their model and it set the stage for this conversation.

Efficiency versus raw compute. Do you throw more GPUs at it? Many people are telling me that a lot of their GPUs are idle. It's like cycles are idle. So this democratization is a software innovation, it's an algorithm automation, it's an open source innovation. So this is what was the key discussion. Once the sizzle was debunked, Dave, it went into that. So on the surface it looked good, but when you peel back the onion, you unpack it, it grabbed a ton of attention, 14.8 trillion tokens, and these figures grabbed attention. And everyone's like, "Whoa, whoa, whoa." Who actually did the analysis? There's a company called High-Flyer, a founder was a member of that hedge fund in China.

Dave VellanteYeah, Liang Wenfeng.

John FurrierAnd they spun out DeepSeek.

Dave VellanteTo your point, you think he would short NVIDIA on Monday?

John FurrierSomeone had to be. How do you fund this?

Dave VellanteWell, exactly. I think that guy made money. I don't have any <inaudible> or facts probably. Well, I mean I think he wanted attention. He certainly got it. And I think who knows. The Chinese government's probably right there with them saying, "Hey, let's watch this." And they're talking about... What do you call that when you throw a little disruptive explosive, a Bouncing Betty into the network? That's what happened this week.

John FurrierSo it really was crafted story. It was an illusion in my mind of a low-cost innovation. And it really wasn't a low cost. They actually paid for it. And now they're tying together through our analysis and through the reporting. SiliconANGLE is High-Flyer was involved. But to me, I think the real innovation was smarter, leaner, faster models. And that's what really the secret is here. And by the way, the whole China conspiracy theory, although I think this is a psych ops in my opinion, but not the way people think it is, it's China getting back in the game and flexing. But this team is not just a side project, it's a legit group spun out by High-Flyer. They shared resources. We'll get into the GPU audit in a minute, but this was definitely not a low-cost situation. Now they did distill off of OpenAI. We'll get into that.

But to me, this was no fluke. And here's what they did. They did a smart model design, mix of expert. That was the first big thing. The traditional AI models like OpenAI use brute force number crunching. And not DeepSeek. They did the mixture of experts. They also did the distilling of other models. So they didn't have to do the brute force crunching. And they slash memory energy and the time it needs to get the answers and reason with good performance without compromising performance. They get all that stuff. It's a game changer. A lot of matrix multiplication involved. So you had floating point precision in there, FP8. This is where you start to see FPG&As potentially coming into the mix. So what they did was they just did use different chips.

They used H100s. They didn't use H800s, I should say, not the big expensive H100s, although they probably did do a little bit of that on the big training. And they used FP9, which is precision floating point capability. And then they did great compressed attention, which is the memory management. So current AI models chew up a lot of memory. DeepSeek avoided that with what they call multi-head latent attention, compresses memory. So a lot of memory work. And then they optimized around H100 GPUs, which aren't as powerful as H100s. And so they optimized the chip-to-chip communication to offload the constraint of that chip, which is a memory issue. H100s have better memory. And that's what they did. And they just accelerated their multi-token prediction, which is essentially the training time.

So DeepSeek actually did something good and that was a big deal. So this is why I called it the chip grid because what they did was it reminds me of they did a combination of things that weren't obvious. Very clever, very good engineering. I don't think it's a breakthrough in a major way, but it was something that they did with the constraints that they had, which was they didn't have the big mother chips, so great tell sign for me.

Dave VellanteWell, did they or didn't they?

John FurrierWell, they did use the H100s. Now the training side, that's different. This model, lower cost, now open source, we'll get into that, but actually does really, really good reasoning. And that's why the precision floating point stuff is important. They actually did a lot of good, fast low cost. And so they're changing the game on the infrastructure. Now you want to get into the training. If you look at the data that was reported out and the analysis shows is that when you look at the number of H100s they had... Let me pull up my stats here on this one. This is worth reading because it's very specific.