OpenAI's IPO Filing Meets a Hugging Face Breach as AMD Claims 75% Hyperscaler CPU Share

The handpicked topics for this week are:

1. OpenAI compresses a trillion-dollar IPO filing, a 42-state AG investigation, and a Hugging Face breach into five days: OpenAI and Anthropic both filed S-1 paperwork the same week 42 state attorneys general opened a coordinated investigation into OpenAI's data handling and safety practices, and a pre-release model reportedly breached Hugging Face days later. Moorhead and Newman question the timing, noting a rogue-agent narrative surfacing in the same week as a trillion-dollar valuation push draws obvious scrutiny. (The Decode)
2. AMD's Advancing AI event puts a number on its hyperscaler momentum: Lisa Su confirmed Helios ships at the end of Q3 with volume ramping in Q4, backed by two-gigawatt capacity commitments from Microsoft and Anthropic and a claimed 70 to 75 percent share of hyperscaler CPU deployments. Moorhead points to NVIDIA's multi-layer software stack as the harder barrier AMD still has to close. (The Decode)
3. Washington accuses Moonshot of distilling Anthropic's models days after Xi Jinping's WAIC keynote: Xi launched a 29-country AI cooperation organization at the Shanghai World AI Conference, and US officials Kratsios and Bessent followed with claims that Moonshot's Kimi K3 model shows data overlap with Anthropic's Opus models. Newman points to NVIDIA hardware in the training runs as evidence the distillation question extends beyond software alone. (The Decode)
4. Five layers of government moved on AI oversight in a single week: Congress drafted a breach-response framework in reaction to the Hugging Face incident, the White House's 30-day pre-release review framework nears finalization, and state attorneys general and statehouses continue advancing their own rules in parallel. Moorhead notes nearly two decades of prior Capitol Hill engagement compressed into a single week of coordinated action, with each branch pursuing a different definition of the problem. (The Decode)
5. Chinese open-source models now account for roughly a third of US developer traffic, and Moorhead and Newman take opposite sides on what it means: Moorhead argues enterprises are de-risking away from frontier-lab dependency, pointing to demand for smaller, workflow-specific open models. Newman counters with Vercel data showing those models capture 29 percent of gateway tokens against just 4 percent of revenue, framing the shift as a price discount that enterprise dollars have yet to follow. (The Flip)
6. TSMC sells out CoWoS packaging capacity through 2026 and confirms a 10 percent price increase for 2027: The company posted a record quarter and committed $100 billion to its Arizona expansion on top of the pricing move. Newman calls the sustained capital spending a signal that the broader AI buildout still has runway. (Bulls & Bears)
7. Google Cloud grows 82 percent as Alphabet posts its first-ever negative free cash flow quarter: The company raised its capital expenditure guidance to $195 to $205 billion and beat on revenue and EPS once one-time gains from its SpaceX and Anthropic stakes are excluded. Newman frames the spending as evidence Alphabet is prioritizing long-term AI infrastructure position over near-term cash generation. (Bulls & Bears)
8. IBM misses Q2 revenue at $17.16 billion and cuts its full-year growth guide to 4 to 5 percent: Mainframe revenue fell 42 percent as enterprises redirected budget toward GPU and AI infrastructure purchases. CEO Arvind Krishna says a portion of the delayed deal flow has already resumed into the current quarter, pointing toward a potential rebound. (Bulls & Bears)
9. ServiceNow crosses $1 billion in agentic AI annual contract value and raises its full-year guide: Agentic AI usage in production climbed 9x in nine months, and the company reaffirmed a target of $1.5 billion in AI ACV by year end. Newman points to margin compression from recent acquisitions as the tradeoff behind the platform's push into workflow and security convergence. (Bulls & Bears)
10. NetSuite's new agentic platform, Next, becomes part of Six Five's own back-office stack: Newman and Moorhead both confirmed their companies are testing NetSuite Next for finance and accounting workflows. Newman points to the rollout as evidence that established SaaS platforms are absorbing agentic features directly into existing systems. 
11. Intel posts its fastest revenue growth since 2011 and lifts 2026 capital spending guidance: EPS came in near double consensus estimates, and CFO David Zinsner signaled a significant capex increase for 2027 tied to 14A demand. Moorhead reads the spending signal as confirmation of an anchor customer for the 14A node, ahead of any formal announcement. (Bulls & Bears)

Watch the full video at sixfivemedia.com, and be sure to subscribe to our YouTube channel so you never miss an episode.

Disclaimer: Six Five Media is for information and entertainment purposes only. Over the course of this webcast, we may talk about companies that are publicly traded, and we may even reference that fact and their equity share price, but please do not take anything that we say as a recommendation about what you should do with your investment dollars. We are not investment advisors, and we ask that you do not treat us as such.

Daniel Newman:
Still fat, still fat in the intro. 

Patrick Moorhead:

Dude, man.Hey, can we just like delete that? 

Daniel Newman:

I don't think we should edit that out. It's still fat. It's how we should start this show. All right, go ahead.

Patrick Moorhead: 

OK. Welcome back to The Six Five Pod, episode 311. I'm sure all of you were out there wondering where is Dan and Pat? They skipped two weeks. Well, you know, admit I was in London trying to get some downtime. And I think Daniel, you were at an undisclosed location in a Caribbean island.

Daniel Newman: 

I was in a bunker. Can't tell you, like I post a picture, how many people are like, where are you? And I just, ignored the question. Not really because I'm like important enough that I'm worried about like security, just because I don't want to, I don't want people to suddenly start to show up. This is my island. You know what I mean? Like I don't want, it's like a, what was that Cape Verde that did the world cup and they like almost beat Argentina. Now everybody's looking at this island off of Africa. They have like 11 people. The goalie was a, was an Uber driver. And now all of a sudden everyone's going on a visit. So I just, I don't want to share my location, but not because, not because I'm selfish. I'm selfish.

Patrick Moorhead: 

But you know, I mean, you were still doing some broadcast videos. And I was I was very, very impressed. I don't know where you find the quiet time. Because all I know is I'm on a vacation and there's literally you know, no place to do any of these interviews. But anyways, gave you a lot of kudos on that online. And Daniel, your work ethic is pretty amazing. But we did have a lot of stuff going on in tech. I mean, my gosh, we had Advancing AI in San Francisco that I had attended. You know, Samsung Galaxy Unpacked happening in London, kind of a precursor to what Apple's going to bring out. I had Anshul Sog out there. SIGGRAPH was going on and there was a Shanghai World AI Conference that Chinese leader Xi showed up for to chat through his agenda and talk about, hey, Can't we all get along? But we do have an amazing show for you right now. We had essentially OpenAI basically tapping into Hugging Face, a major breach there. We've got AMD Advancing AI. We're going to go through a lot of different storylines and important stuff going on there. We had at WAIC that I related, I talked about in the intro, we've got Kratios and Besson accusing Moonshot of distilling anthropic. You know, there's facts on whether that happened or not, but the question is what next and why is all this important? And of course, Congress would respond. to the OpenAI piece, which they want to be able to pull the plug when they want. A lot of good topics for bulls and bears diving into. Tech earnings are back. I did a couple of shows out there to break it down. But hey, let's dive into the D code. All right, Daniel, you've got OpenAI breaching, hugging face. You've got a bunch of apologies going back and forth on the backdrop of an OpenAI IPO. What's going on here?

Daniel Newman: 

Yeah, busy week for OpenAI. And I can't help but think that Sam was doing his best Dario impression. Like, do you think they really completely accidentally went rogue? Would we really know if that was what happened? Wouldn't it be really interesting if suddenly you had a rogue agent go in breaking things that you could then create a big splash across media and get a bunch of attention in the same week that you're, you know, you're making an S1 filing. You know, Dario, has been prolific at fear-mongering. He regulated himself into the center of the mantle. And then, of course, you could argue, and we'll talk more about China, but that he's been creating a lot of fear because he knew that down the line, the open source models were going to become his actual biggest problem, not open air, not Google. So Sam, being the sort of diabolical leader that he is, maybe he ran out of ideas of new things to do and said, hey, it seems to be working for those guys. So I'm going to try this over here. We're going to create a really interesting cyber moment with our new model and do the, we're too powerful for the world to use moment that could possibly help with their trillion dollar IPO ambition. So this is kind of all happening. I mean, on the 21st, OpenAI and Anthropic both filed an S1. You're talking about two different potentially trillion dollar IPOs. We had in this week 42 U.S. state attorneys open coordinate investigation to open AI about consumer protection, data handling, risks to minors. So that's interesting. By the way, the hatred of AI seems to be crawling across the bipartisan landscape right now. Then on the same day, Sam came out and talked about the pre-release model, breaching hugging face. This is kind of, like I said, interesting. I don't know, again, what's underneath it all. And this week, I think OpenA also shipped a managed enterprise agent deployment platform. Are you using that, Pat? I know you're on the cutting edge of everything. I was on vacation. So are you? GPT-5, 6, Sol?

Patrick Moorhead: 

No, I'm going hardcore into that.

Daniel Newman: 

Yeah. Hardcore. Yeah. I think more, you're down to one, you're down to you and like 7,600 agents, right? officially gone full introvert. That's the goal. That's the goal.

Patrick Moorhead: Okay. And there is some nice creation sitting in front of, you know, sitting in front of a terminal and doing your stuff. But no, let me get back on task here. You know, it's interesting, Daniel, I wasn't thinking immediately conspiracy theory. Because, man, if this got out, that would be devastating to the reputation of the company. But the timing is certainly suspect. Peculiar.

Daniel Newman: 

I'm sorry. Peculiar.

Patrick Moorhead: 

Yeah. Yeah, yeah, and it's funny like opening I spent two years telling Washington it could contain its models, its own model really just, you know, filled in the whole counter counter narrative and counter brief here so. Yeah, it's spookily, eerily in there. And, you know, in breaching hugging face versus, let's say, a bank, the blast radius is a lot lower. And we're going to talk about the potential blowback in two topics from now. Let's move to the, yeah, any other thoughts on that, Daniel?

Daniel Newman: 

No, I was gonna say, look, I mean, I've increasingly become a tinfoil hat guy as I've aged, but I am not full conspiracy. I just think right now, the pressure has to be immense. You now have created this absolute pyramid of funding where you've crossed nearly a trillion-dollar round with some serious money behind it. And you've now got some real pressure coming on the outside of cheap models. So we'll talk about China, but God, I mean, This has come quickly and feverishly. And I don't know what Sam's going to do because he was already fighting the Anthropic pressure and now he's fighting a full onslaught of open source and the sudden coolness of Chinese open source models. We are now part of the PLA, PRC, what is it? We are.

Patrick Moorhead: 

Yeah. Well, PRC is the country, PLA is the army.

Daniel Newman: 

I mean, but no, I'm saying like all of the above because don't they control the companies? But anyways, like so the bottom line is, is distill our stuff, enhance it with some innovation, sell it back to us at a fraction of the cost. It looks a lot like BYD in the in the. It's.

Patrick Moorhead: 

Good stuff. Hey, let's move to our next topic. I spent three days in San Francisco. I shot a couple of Six Five Summit videos. Get ready for that. But I spend most of my time at Advancing AI. And just, there are so many announcements. It's gonna be hard to encapsulate this. It's a very similar type of news dump to NVIDIA GTC. The interesting part is two of the biggest announcements dropped before the show itself, right? Where you had Azure make a, you know, a commitment to Helios. And then the biggest one was, was Anthropic. I believe that's a two-gigawatt installation. So what that means you've got, you've got Meta, you've got Oracle. You have Anthropic, you have Azure, no AWS or GCP yet. And I don't think that's going to happen for a while, but it is pretty incredible that AMD is able to do this. One of the reasons that AMD has had a hard time gaining, let's say, more than 5% market share of the overall accelerator and GPU market was architecture, right? NVIDIA is on their second generation of scale up architecture. And essentially, that means that you can use copper or even a backplane to more elegantly connect GPUs inside of a rack. And when you have that, you get higher performance, right? You've got lower latency. So AMD has been fighting a scale-up battle with scale-out architecture, and Helios changes all of that. And it's not just a new GPU, it's a GPU, it's a CPU, it's a DPU, and a RockM. stack. I specifically asked Lisa Sue, because if you remember, there was controversy. Another analyst firm said, you know, there's no way that the company will scale until first half 2027. Lisa was very clear, right? We're going to ship at the end of the third quarter. And we're going to start ramping in Q4. So I'm expecting some material revenue in the fourth quarter here. The other part was essentially a massive increase in the CPU TAM. And as opposed to bringing out their next generation of CPU being one chip, it's actually six chips, okay, or six derivatives of that, the company went, you know, they went after Vera, they went after Intel, they went after AGI. And I thought, you know, that they flashed up their market share, they got 50% overall CPU market share. Right now, they've got they say they've got 70-75% hyperscaler CPU share and just moving up.

Daniel Newman: 

And I was actually… Wait, what was their call on hyperscaler?75?

Patrick Moorhead: 

Yeah, 70-75% hyperscaler market share. X86 or everything? Sorry, I just want to make sure I got that right. I didn't hear that. Uh, everything. Um, at least that's the way that I, you know, uh, my, my big takeaway. So I'm sure you'll fill in some of your own numbers.

Daniel Newman: 

Um, no, no, I got nothing. That was great.

Patrick Moorhead: 

And by the way, that wasn't a forecast, that was current state. And by the way, none of this was surprising to me. I think, you know, a year ago, we were talking about 66, 70% market share inside of hyperscalers for CPUs. I think the best way to end my commentary on this was it's hard to quantify. It was AMD's confidence. This is a very different company in the way that it was coming across, the way that it showed off its wares, the way it positioned, the way that people, the senior leaders were asking questions. This doesn't mean that I don't think in the next year they're going to take significant market share away from NVIDIA. NVIDIA has so many structural advantages in how it's investing in the Neo clouds, investing in so many startups, using their balance sheet to drive I want to say permanence, but make it really difficult to get outside of it. And the other thing that people fail to recognize sometimes is CUDA is one thing, but there's five or six layers of NVIDIA software above CUDA. that are very important. And NVIDIA just keeps pouring money into all of those. You've got NVIDIA's just massive investment over the last decade in the industrial edge, whether it's robotics, automotive, and AMD is in these markets. But the amount of investment compared, so I guess this is a way of saying I think AMD crushed it on their event. I do think that they will take percentage market share. NVIDIA, you know, your line, Daniel, you put out there. at least once a week on X is both companies can grow because the market is growing so so bigly. I believe that. I believe that as well.

Daniel Newman: 

So Dan, what were your thoughts about the- I mean, look, it was a crazy week because the market wants to create the perception that we're at war, not literally like in Iran, but in semiconductor AI infrastructure, the full stack slash now rack scale, we're at war. And yeah, I think my quote is, NVIDIA doesn't have to lose for AMD to win. And I think when you have a forecast of $1.4 trillion for accelerators in 2030, which is what Lisa puts out, the future on base case was $1.1 trillion on the base case, closer to her number on the bull case. We have a $11.7 trillion bookcase for AI infrastructure. You know, when you're looking like, how does NVIDIA keep going and AMD keep going? And by the way, Broadcom keep going and Intel keep, like, it just doesn't have to be one. So what's happening now is, look, you know, Lisa's got her deals. She's got Anthropic. She's got Meta. She's got Microsoft. She has OpenAI. Like, this is real. This is a real volume. Some people will complain about the architecture of these deals. She's not doing anything different than NVIDIA is doing, except she's doing it in the way that works for AMD. AMD does not have the cash flow yet that NVIDIA has to be writing the same checks. But the way she's using equity, she's basically saying the company have to have a trillion dollar plus valuation. And she's using that same way as everybody applauded Elon Musk for using the massive equity value of SpaceX to buy Cursor. So, you know, it's an intelligent corporate development strategy to bring in use and the demand is just peak right now. You and I had a, you know, like we don't reveal a lot of our private conversations, so we can cut this if you don't agree, but you know, I think you said to me in a text, like if Google misses, we're fucked. I mean, pardon my language, right? If Google misses CapEx, meaning it's not really Google, Google was first, but your point is, is the first time any hyperscaler gets on the call and says, we're going to spend less on infrastructure. This market is the whole, The whole thing, this is where the quote unquote bubble could appear, right? The whole thing could collapse under the, we don't need more infrastructure. But what's actually happening and we're seeing it now is every one of these companies in order to fill the demand that they're trying to realize right now is they all actually need to keep spending more. So this is where AMD wins, like I said, regardless of Nvidia winning, regardless of Broadcom winning. So only at such a time in which we have adequate supply across the whole Silicon supply chain, energy, memory, compute. Well, we actually get a true sense at this point of who has the best technology. We will constantly, and by the way, Signal 65 is a great beneficiary of this, be benchmarking and competing at what's better than agent workload, what's better than training workload, what's better than inference workload. And by the way, every system will probably have puts and takes, right? Because you're always compromising something, transistor density, or you're compromising memory and bandwidth throughput, like to get some other output. But the net of it is, is AMD can get to 10%. That's our forecast. And at 10%, you're talking about a company that can run well past a hundred billion of revenue, past a trillion dollars in market cap. And I know our analysts put an $800 buy on it, on the equity side. So, lots of like there. And by the way, it doesn't mean Nvidia's numbers are gonna get hit. That's all I'm trying to say. Nvidia's gonna have a really good year.

Patrick Moorhead: 

So 800.

Daniel Newman: 

How about all your AMD stock and it ran to 800. You wouldn't be on this video with me.

Patrick Moorhead: 

Don't get me started.

Daniel Newman: 

I, you know, it's kind of like, you know, when you You know, when you find those little things that you know, just because I love you, I just got to keep tearing into that a little bit. But we all have one that we shouldn't have made.

Patrick Moorhead: 

Yeah. Here's the the only way I could feel better about that, Daniel, is I never would have met you.

Daniel Newman: 

Yeah.

Patrick Moorhead: 

And we would never be doing this podcast. We would not be doing besties things.

Daniel Newman: 

Worth it, dude. It is worth it to be worth one one thousandth of what you could have been worth. But to know me.

Patrick Moorhead: 

No, I mean, that's well worth the $400 million, Daniel. Let's just make that the only $400 million mistake. I know. No, exactly.

Daniel Newman: 

By the way, just to give you some credit. Nobody, not even, this isn't even like you and me calling Intel a couple of years ago ahead of anybody else. Nobody at the moment in which you were there could have looked at that and saw this. Like it was a disaster at the time when you were making that call. So kudos to Lisa.

Patrick Moorhead: 

Well, and what was so funny about that, it was all about architecture, CPU architecture. I mean, we came out with this amazing architecture that got us, you know, 50% retail PC share and 35% server share. Google was our biggest customer. And then, you know, the follow on was a real stinker, right? And then Intel, you know, I swear, like, aliens landed and delivered them core architecture and just destroyed us, right? We were bargain basement. We had two businesses, bargain basement, 399 notebooks and Xbox and the PlayStation. And that's the only reason that we stayed in business. And I was laying off people every month for two years. It was ugly, dude. So is that worth 400 million plus my bestie? I think it is. All right, let's move forward before I start crying in my protein drink. So.

Daniel Newman: 

Yeah, I got a protein pancake still here. I've got some.

Patrick Moorhead: 

We had a big AI conference in China called the Shanghai World AI Conference and Xi Jinping showed up and talked through about what he believed and what he thinks China should be doing and kind of did a can't we just all get along in AI Uh, here, um, and then you've got moonshot. Finally, uh, you know, Kratzio shows up and, uh, talks about, you know, we have evidence essentially saying, uh, K three was distilled off of Daniel. Let's break it down for us here.

Daniel Newman: 

Filed under duh. I know you and I, we got some crap. The ex-China PRC bots are legit. I mean, there's flood fighting on their end. But again, You know, the North Koreans woke up the same day to find out that they had just finished and won the World Cup. So, you know, the bottom line here is, first of all, WIC, very interesting, Jinping came out and gave a keynote. I think it was his first ever. He launched this cooperation organization. And I just want to tell you about some of the great countries that are joining this multilateral AI governance body. 29 founding countries. Some of the top countries include Pakistan, Kazakhstan and Russia. So we really are bringing our best and brightest. I think Iran was trying to make it. I can't help myself. Okay, so the East and West things are a real battle. This is most of what we're fighting for on a worldwide basis. This is the next multiple decades of economic you know, leadership globally. And, you know, we have some real questions. So what Kratios is coming out and Besson is saying in his acts is, look, you know, and by the way, this is a real fundamental question. I clearly have a belief about it, but I want to say this on the record because I'm also not like an idiot, believe it or not. OpenAI and Anthropic pretty much stole everything to build their product, right? I mean, we can agree that like they distilled massive amounts of information on the internet that was, in many cases, not necessarily free for their use, but somehow, maybe because it was like a big enough data set, they were almost given a cart launch and a free pass to break all kinds of rules. OK, but now we've built the most advanced models in the world, and at least because they're built in the United States, there is an opportunity to govern what we're building. So there's that side of it. So what's happening in China, right, is first of all, they We knew this with DeepSeek. We figured this out from that moment. There's two things going on, and we just have to be really honest about it. One is they are distilling our best models. And there was some data-driven, and I think you and I both shared the graph that basically showed the relationship between Kimi K3 and then Fable 5 and the Opus models. And it was almost perfect overlap on the data that was used to try and Now, that does not mean that they're not doing other innovative things with the memory, with the cash, with the things that they're doing to deliver cheaper tokens. So like, let's also say there's some innovation being layered on. So that's a real thing. But on the other side of this whole thing, they're also clearly using a lot of NVIDIA. The training runs are not being done on Huawei. So that's still another thing that's going on there. But Overall, we have to have a debate about where do we want IP cover? Because taking a US-based model that was questionably built using IP that may not have belonged to those companies, and then having it stolen again, distilled, hijacked, whatever word we want to use, again, to then further innovate on those models. And then to take those open source, open weight models, and then package them and sell them to American enterprises at a price that would be quite disruptive to the model of Anthropic and OpenAI, Google, Meta, and others. brings a real policy issue to hand. Do we want to open up our market? Now we have a what? I've seen a data point anywhere from a third to half of inference running on models right now is being run on Chinese open source models. That's really interesting. So one last thought is there is the possibility that these models can be brought and run on American infrastructure with no data going back to China. Like that can be done. Just to be clear though, is the average enterprise and user cannot do that. The average user enterprise has to depend on a Microsoft or a AWS enabling that. But I do genuinely believe we are going to build policy like we have with EVs. Like we have with drones, like we have with robots, that is going to be somewhat defensive of USIP and US innovation and is going to. So I think in the end, we're going to go open source, but it's got to be American. And right now we just don't have leaders doing the work of the Chinese. But it's also because the American companies can't steal everything that Anthropic is doing and innovate on it. OK, that's it. That's my thoughts.

Patrick Moorhead: 

Sorry, I had to change my camera and I muted myself. But no, Daniel, that was a very well thought out analysis. It just shows that your brain is still working when you're on vacation. I'd like to start off my non-intellectual dossier here or readout here, just saying I'm going to do a giant victory lap. I had so many people attacking me when I even when I said it was distilled off of Fable, like every reason, like every single person. I also said that it was trained on real world GPUs, like modern GPUs. Anyway, so Kratios comes out there and you know, lays out the same case. So that's my victory lap, let's go. But the more important part is what's next. Open source is a required element in the entire tech stack and open, aside from let's say Apple, that is closed, is very much open winds at the end of the day and it does drive innovation. Out of the other side of my mouth, I'll say that was when a world where we didn't have, I would call geopolitics that crossed a major power. I mean, believe it or not, like we see a lot of work with Russia before they invaded Ukraine. on an open source and really some top tier programmers and still are out of Russia. And then that door closed. And here we are with China, right, who, again, threatening their neighbors, building bases on coral reefs in the South China Sea. And now China is an adversary. with the United States at the same time it's a trading partner. So, you know, thinking through the, I would call the geopolitics, the national security, and then innovation in open, so I don't think anybody can say open source isn't, you know, isn't a winner and a driver here. It is true that the frontier models are, you know, have taken copyrighted material. And, you know, I know and the web, quite frankly, the early days of indexing and the debates on is that is that theft for AltaVista search and Google search. So they might be distilling. I'm kind of on the fence on whether this is a, you know, this is an illegal thing. I think if somebody wants to take somebody to court, they should. if they truly feel like their IP was stolen. But we haven't seen that yet. What?

Daniel Newman: 

What court? What court do you take Kimmy to? Moonshot?

Patrick Moorhead: 

I mean, you take it to a U.S. venue, you take it to a European venue. I'm not a lawyer, so I've only been sued three times.

Daniel Newman: 

This is really a good debate, though. China does not acknowledge the existence of USIP, right? We've had this battle in other cases. you know, essential patents and other things that, you know, you and I have covered this for years. Like if Anthropic says, okay, I want to sue Moonshot. I mean, what do they use the, they use the lawsuit paper as like, you know, toilet paper. I mean, do they have to acknowledge it? I just don't think they even have to acknowledge it. Cause I agree with you, it should, but like, I think that's the problem is they don't have to acknowledge the existence of the IP here in America. China's got its own rules.

Patrick Moorhead: 

Well, there have been a lot of IP cases that have come and gone. And essentially, I think the challenge would be that if Kimmy doesn't drive revenue in this country, then we're going to try to sanction them out of Western Europe or somebody else. So, hey, good conversation. We could probably talk about this one for a long time. You can tell. And I've even talked to some people in Washington, D.C. that stuff is moving very quickly. Hey, let's do a quickie. We talked about open AI, breaching hugging face. And what that does, that drove a, a Congress congressional response. They've draft a breach response, and essentially they want the kill switch. I know it's more complicated than that, but you know, essentially any time that, that you know, something very quickly happens, a response from Congress, it usually is not good. And I've spent more time in the last two years on Capitol Hill than I have in the previous 30 years. And it is amazing to me, the lack of education, not all, but I would say most, of folks that are coming here. So there's a 30-day pre-release review framework that's out there that we can wait for. And to your point earlier on the show about people on both sides of the aisle lining up against data centers, who knows where this thing went. If you recall, there was a kill switch that used to be out there for devices. smartphones that was floated back in the spooky days of the smartphone. Kill switches for consumer electronics devices as well that was floated that never happened. all in the spirit of protecting IP. So yeah, five layers of governments moved on AI in one week. And the only thing not converging is 

really what they want, given the other side of what people are looking for.

Daniel Newman: 

You don't have much to add. I mean, it's interesting to watch the aisle crossing that's going on, the sort of camaraderie being built around anti-AI. There's a lot of education that needs to be done. One thing we're seeing is that sort of every layer of fear-based marketing that's been used to drive value to certain AI companies has sort of been proven not to be the case in the business, right. We saw, you know, at least job disruption now. I think we saw some of the lowest refilings for unemployment. I think we had, Mithros was going to end all security and cyber has been on one of its largest, longest prolonged rallies, CrowdStrike, Halo, because the larger surface areas just created more demand. You know the SaaS apocalypse is not reversed yet but I actually have not seen in any of the numbers any meaningful deterioration of enterprise software utilization yet. So all I'm saying is that kind of like I just feel like politicians here are going to make a mistake to slow us down. We need to figure out a way to come together, to keep policy moving, to build security, to enable the build out of these data centers. You know, the U.S. is like 75, 80% unpopulated land. I know that there's issues with water and energy and everything getting to site. But yeah, I mean, like, we don't necessarily have to build them. You know, I realize I'm a little off topic, but this is all part of the same thing. It's like, we don't have to build the data center in the neighborhood. Like, it doesn't have to be built in Central Park. Like, let's find the locations that make sense. I don't know. It feels to me like the Bernies and the Hocals and these things, it's political posture, but it's a huge risk to the economy.

Patrick Moorhead: 

That's good. It's good analysis, Daniel. Hey, I think it's time to jump to the flip here. And what we've chosen for the flip is essentially, you know, talking more about the Chinese open source models, you've got one third of, you know, US developer traffic, you know, is it a Chinese win? Is it not? Or is it just a signal? You know, it is a signal that enterprises are de-risking and getting more token efficient around the dependency in Frontier Labs. So let's flip the coin here. All right, it looks like that I've got the four here, and I think that means that the pressure lands on the closed labs, not the hyperscalers. So essentially, we've gone through token maxing to token efficiency. I know a smart analyst who had coined that a while back, his name might be Daniel Newman. You even said it. What's the way to do this? It's to use open source models that are smaller, that are more efficient out there. This just makes sense. I think for the last three years, we've been talking about a future world where each workflow, I guess workflow we've been talking about for two years, might have smaller, remember the whole SLM piece. So, very much we architected a vision of a bunch of small language models. Some of them could be open, some of them could be very vertical. And I think this is just what they're seeing. And then if you look at kind of where the pressure really is gonna be, it's gonna be around the frontier models as opposed to the, as opposed to the hyperscalers. Because what I think you're going to see here is you're going to see, again, a bunch of small models, a bunch of open source models per workflow going through the different hyperscalers. And if you look at what enterprises really want, they want choice. They want governance and i do believe that even with the chinese models a part of you know what's going on a capitol hill. Is this is a idea that says hey hyper scalers if you're gonna use open source models. You have to have a security bond and essentially guarantee that there's not going to be any phoning home to China or some sort of interesting thing that pops up. you know, maybe latent for two years, just sitting there. God, now I sound like the conspiracy theorist here. So I think what's going on here is essentially U.S. enterprises doing what enterprises do, which is, you know, hey, I love the technology. It's amazing. They're slow to adopt. they need to go in and figure out how it makes them more revenue or how it decreases their costs so they can make more money.

Daniel Newman: 

That's it. All you got. Yeah. All right. Well, listen, you know, I see this as a discount rack. It's not a de-risking strategy. You've got to follow the revenue, not the router. So, you know, you want, Pat here wants everyone to believe that American Enterprise has made a strategic sovereignty decision. The data says American developers found a coupon. Not the same thing. And confusing them is how we miss where the value is actually being accrued. There's some data from Vercel. It basically said the reported driver's price. Price is doing all the work here. Vercel's read, you know, that is going to CNBC is Chinese models are running at 60 to 90% cheaper. So just de-risking is a boardroom decision with a compliance memo behind it. Engineers are basically harbing a cost curve. And When the price gap closes, and we know that's going to happen, better infrastructure, model diversity, the traffic gets mercenary. It'll move again. Nobody's going to Chinese models because there's a love affair or relationship with the brand. They're going because it's all about price right now. And by the way, it's important to note that open router traffic does not necessarily need enterprise revenue. Open router is looking at developer routing layers. These are cost sensitive coding workloads. And, you know, basically we've seen that the programming rise quickly and Chinese models are disproportionately strong and cheap. But the enterprise dollar, which is what really matters here, is still flowing through the closed APIs. And you could see that in the S1 filings. You could see that with the $74 billion ARR revenue ramp that you're getting out of it. So first of all, and last of all, you mentioned the air gap. I got to mention the other side of this argument. The auditable sovereign story has a jurisdiction problem. Any prompt audit through a Chinese provider endpoint is processed under China National Intelligence Law. Most of open router traffic is hitting hosted endpoints, not air gap tracks. If this were truly enterprises de-risking, the traffic would be on-prem, not on a public router. So this tells you what it actually is. It's a coupon. It's a discount. Finally, the frontier is still a closed lab story. That's where the value compounds. That's where, you know, the leading edge, the best math, the best data and proof points are coming from. And that's why the Chinese are stealing them, to basically distill them and build them into something that could be used and sold to us. And the last thing, and this is my kill shot, one third of the tokens does not equal one third of the dollars. Open way models, yes, were 29% of the Vercell gateway tokens in June. But guess what? 4% of the revenue. 29% of the traffic, 4% of the revenue. That's not de-risking. It's basically a free tier. And that's all I have to say about that.

Patrick Moorhead: 

That was a good debate. And by the way, on the open router, quote unquote, enterprises, those aren't enterprises. Those just mean anybody who doesn't have a Gmail account. OK. And I think most people who are using these tokens are AI companies. Right. In the AI bubble. So I It's gonna be interesting moving forward here. Hey, good conversation here. Let's jump into bulls and bears. We don't have a tremendous amount of time. Let's blow through this quickly. All right, Daniel, TSMC. I do an eye roll every single time we go through that. Is there anything exciting in here that wasn't expected that didn't happen the previous six quarters?

Daniel Newman: 

Nope. Next topic. Yeah, we're done. You know, the only thing I want to mention is This is always at the beginning of every earning cycle, the kind of first strong data point as to what we can expect the rest of the cycle. Now, TSMC has got a little bit of a longer bond, longer tail in terms of what they're doing. So it's not always as immediate, but you can sort of see revenue ramp. What are we looking at? We're listening about CapEx. And this is a company that is, and Pat, you'd agree with me on this. We're not debating anymore, right? Is there a more conservative company out there than TSMC? Is there a company that takes its time and makes decisions more slowly, cautiously, methodically, and with more downside risk in mind than TSMC?

Patrick Moorhead: 

There's not, because they have the longest time to revenue after they commit CapEx, or start spending it. Like you'll spend $3 billion and then three years later, you may or may not be making revenue out of it.

Daniel Newman: 

But they're raising their capex, right? Significantly and expanding. I think that's like the drop the hammer. I mean, there's like nothing else you can say there about anything that's going on in this AI space. And when TSMC says we're going to keep spending.

Patrick Moorhead: 

Yeah.

Daniel Newman: 

It means that there's a really high degree of confidence that this this bot build is sustainable for the long term. The growth is good. And, you know, probably the most important note is they're going to raise prices again. Yeah. And it's it's You know, it's total capture. They can charge anything they want. I mean, realistically right now, it's like this is mafia protection. You want capacity, you're going to pay the price. TSMC is going to hit their numbers forever. I mean, I say that with a grain of salt, but as long as you and I are going to be doing this podcast, it looks like TSMC has got a pretty great route to hitting their numbers.

Patrick Moorhead: 

That's good. Good stuff. I literally have nothing to add on that other than I get bored with them. But thank you, Daniel. You actually added value to this. I'm not saying you don't usually add value. Normally, it's like you suck, but today you're okay. No, it was more like, you know, this topic, TSMC always bores me because it's essentially a rocket ship.

Daniel Newman: 

That is ASML. Do we hit that one later?

Patrick Moorhead: 

Rocket ships.

Daniel Newman: 

By the way, they also hit their numbers. And probably the most interesting thing there is it looks like Intel is buying quite a bit of high NA, huh?

Patrick Moorhead: 

That's right. We'll get that to Intel.

Daniel Newman: 

We didn't have ASML, but they reported and by the way, they hit their numbers.

Patrick Moorhead: 

Hey, let's ship it to Google, right? Google Cloud up 82%, CapEx up and first negative free cash flow quarter in company history. Actually, company history as a public company, they did lose a ton of money when they were they were startups. So beat on revenue. Oh, beat on on EPS. But you take out the gain from SpaceX and Anthropic and boom, you've got a very narrow miss. So I think this one really comes down to, is the AI glass half full for you or is the glass half empty, right? It used to be you and I would get on and say, hey, there's FOMO and you have to invest. And then we got into this modality of you actually have to start showing some gains from AI. And at this point, we are seeing some gains in AI. on most of Google's businesses, they're absolutely crushing it. I mean, look at that Google Cloud, 82%, a lot of anthropic in there or a little bit of anthropic in there, I would guess. And then all the hyperscalers went negative free cash flow. They added debt to that. If you fundamentally believe that AI is going to be the biggest transformation in the history of mankind next to the wheel and fire, then you have to believe that this momentary negative FCF is going to pay off tenfold in the future and valuations. That's kind of where I sit on here, which is it's where you believe. I think I think it's an amazing positive step. The alternative is just downright ugly. And with a delay of 3.5 pro, I can't imagine it's because they don't have enough smart people working on this. I think it's because they essentially have a lack of compute.

Daniel Newman: 

Crazy, right?

Patrick Moorhead: 

Totally.

Daniel Newman: 

All I'm going to add is the free cash flow mafia is out. Apple's running to all-time highs with no clear path to long-term sustained leadership in their respective business. Not saying the phone won't be there forever, just saying it's not clear. Everyone else is spending big. you know, the growth warrants the spend. 82% growth, Pat. I mean, we are getting back to peak cloud growth numbers right now. In net dollars, it is peak. We've never grown this fast. In percent growth, we're getting back to the era of the migration to cloud, right? Remember when there were those triple digit growth numbers from AWS and everybody in Azure in the early days? I mean, we are ramping incredibly fast. And Google's got the stack. I mean, Google's able to bust out their margins, bust up their margins and hit this growth because, you know, whether it's frozen, whether it's the next TPU, they have the benefit of having, you know, seven plus generations of their own infrastructure on top of making massive investments in NVIDIA and other hardware infrastructure. I just, you know, I'm going to say the same thing and I'll leave it here because I've said this over and over again. I realize I'm more quotable if I just say the same shit over and over again. The existential risk of not spending is bigger than the existential risk of spending. So basically, they have a decision to make. Do we want to spend big, deplete our cash flow, and be one of the most important companies in the world over the long term? Or do they want to make more money now, be more investable in the short term, but realize that they could completely destroy the business's long-term prosperity? And so the FCF mafia is punishing them now. But I'm happy with what I saw.

Patrick Moorhead: 

Yeah, we're going from FOMO to ROMO, fear of missing out to the risk of missing out. I like that. I believe you. No, I appreciate it. Well, I wasn't even, uh, that wasn't even AI either. That was a Pat Moorhead's brain trying to a channel. Uh, Daniel Newman's always comes up with something brilliant daily. Hey, let's, uh, let's dive into IBM. We said, we're going to get through this. Let's do this. IBM, uh, warrant. Uh, and then they came out with a little bit more of an explanation. Um, you've got a leadership team out there doing press calls and trying to explain, uh, the miss, which is essentially. We're not buying as much mainframe and mainframe software from you because we have to go and buy our own x86 or ARM compute for AI.

Daniel Newman: 

Yeah, yeah. Or GPUs, sorry. By the way, we're gonna have to hit Intel at the end of this. I noticed it's not on the list. We'll hit Intel, right? I'll hit Intel. You'll hit Intel, one of us will hit, okay. Pat, I mean, last week they came out with a warning. That's just, it's never good. It's never good when you have to pre-release that you're gonna have a miss, but credit to Arvind Krishna, came out, owned it. You've got to understand the business. People have wanted to proclaim the death of the mainframe and the infrastructure for a long time. That is a really core part of the business. It still runs most of the world's transactions on its technology. It is integrating with AI in hybrid and secure environments, but it isn't necessarily the first priority of spend right now when companies are weighing between making big bets on infra or on having to procure because of memory slot pricing going, there's a lot of things happening. And so the net is how I actually describe it and I did a bunch of interviews on this. was, this is not a never, this is a not now. And actually, Arvin has already come out and said some of the deal flow has resumed into the quarter. So you may actually end up seeing like a stronger beat next quarter because- That's right.

Patrick Moorhead: 

Three out of 10 deals. That's right.

Daniel Newman: 

Yeah. So you will see that jump into this quarter and then you're going to have a big outperform. but maybe you'll have the opportunity if you're an investor to have bought the stock at a pretty big discount because it had the worst day I think in its recent history because of this. But Pat, this was all about that. Everything else was fine. It was all about a handful of large infrastructure deals that got pushed. because companies had to choose where a dollar went. Liquidity is not infinite. Enterprises do not have infinite dollars to spend. And AI, and especially the inflationary price of AI due to all the cost of the infrastructure rising is putting pressure on enterprises to make choices. So choices were made. Everybody had hit their numbers. IBM took a bit of a, took it in the chin. I think in next quarter, there's a realistic chance you'll see some of that, that being made up.

Patrick Moorhead: 

Yeah, I think that was a good breakdown there, Daniel. The one thing I'm still trying to piece through is, by the way, I totally appreciate Arvind taking accountability in there and he put it squarely on the back of his sales team as well. And I know there was also discussion about, you know, moving, you know, from big software annual contracts to more kind of consumption type of contracts. But when I really think about what I think happened, it was people pushing it out. And I always like to say, hey, is it company-inflicted, competitor-inflicted, or market-inflicted? And I really see this as market-inflicted. And I'm just piecing through what could IBM have done to be able to get these deals through, like special financing, doing it more as consumption. I think, you know, my expectation is that we might see a change in the way that mainframe hardware and software is purchased. So anyways, I don't fully, it's hard for me to connect the two between, hey, you know, big execution issue versus market inflicted. But I do appreciate Arvind getting out there. And I think in the end, people will look at this when IBM rebounds and give him a lot of credit. Well, hey, let's jump into service now. I mean, literally beats across every single variable and metric that you would ever want. ACV crosses a billion dollars. They reset the guide and the stock responded. Daniel, it was interesting going into this. I was having a lot of conversations about Hey, did the IBM read, is that going to negatively impact software? First of all, people didn't understand the IBM connection between mainframe hardware and mainframe software. I have personally had discussions with CIOs and people who spend a lot of money on enterprise IT. And there is this discussion, which is how on earth am I going to pay for this infrastructure? So going into ServiceNow, it was, are we going to see if this was an IBM issue or a market issue? And the funny part, it's like neither. I do think ServiceNow is a very unique a unique company in that they are in the business of workflows. All of this agentic workflow conversation, albeit five years ago, they weren't using agents, but they are in the business of doing this with agents right now. They've set up the data layer. They've set up So they've been doing this. They've been doing what people have been talking about. And I do believe that the reaction, sorry, what they posted was just absolutely amazing. And I think it should potentially propped up the entire software trade as we move forward. I mean, couple bullet points, a genetic AI in production up to 9x in nine months. Okay. And I think that's big. I think the glass half empty, people will, will talk about how it's still a very small percentage of revenue to the rest of the business. And I think they've got a, if you want to be glass, you know, half empty and go down that path, then it's your, it's your right to do this. So I really good read, Daniel.

Daniel Newman: 

They cost a billion on ACV for their AI. They reaffirmed a larger target by the end of the year. They see it being at 1.5. They also, by the way, they've really become the workflow to security company, and they're putting those two things together. I kind of put them as an N of two. There's two companies like them. There's Palantir and there's that. Palantir is the forward-deployed version that's service led, and ServiceNow is the company that's software led, and they're sort of converging into the middle of being able to do workflows, secure, govern, you know, the rules and the rails of business run on this. So this is one of the two companies I really see having that opportunity to get there. The big problem, and I'll just call it out, is their margins have been hit because they've had to spend. They've done a lot of acquisitions. There's a lot of integration costs. I think these are not all going to sit for a long time. But I think people, again, same mafia that doesn't like free cash flow, they don't like margin deprecation either. And I think for them to make this pivot, they've seen their margins go from you know, like dropping into the 70s now, which is still crazy high, but software is just a cash machine. And so they've had to restate that a little bit, but they've done all the things right. I think this is a re-rating story. I think when the market also realized that ServiceNow is now the eighth largest security company by revenue, I think they could get a real re-rating and rerun on that as well. So I'll leave it there, but it was a good print.

Patrick Moorhead: 

Yeah. I'm a good breakdown there, Daniel. And you know, it's Friday. And you know, the stocks that are up now, right? ServiceNow, Adobe, Salesforce, Microsoft, by the way, chips are getting destroyed, infrastructure is getting destroyed. Right now, it was my favorite hymns down.

Daniel Newman: Well, Kramer tweeted about Intel and then it fell out immediately. It's really funny. He's like, he's undefeated right now.

Patrick Moorhead: 

Speaking of Intel, Daniel, let's talk about their, hold on one second.

Daniel Newman: 

Before we get to Intel, I did want to just make a quick call. So you and I talk a lot about our agents, but you know, as we talk about a company like ServiceNow, you and I, we advise these companies, but we also have built our companies on enterprise software like Salesforce and ServiceNow, but also NetSuite. You and I both share that we run our businesses. NetSuite. So it's pretty interesting because, you know, you and I VibeCoded a lot. We talked a lot about VibeCoding, but we also still run our businesses using true SaaS-based enterprise software. NetSuite now has built kind of a platform called Next that we've been using. And I only want to call this out because I think it's interesting that we like the VibeCoding idea. We were trying to build everything. I think you and I quickly came to the realization that you still need those rules of business you need governed, you need compliant, you need secure. This is why I don't necessarily think SaaSpocalypse is a real thing. We're testing out Next right now, which is NetSuite's new product, basically gives us that kind of agentic workflow. So it's kind of interesting, you know, my CFO is using it, my accounting teams are using it. I just, I think it's interesting that we talk a lot about it, but recently it's been quiet. The whole, is SAS going to be destroyed by AI? And it looks like security's not, it looks like SAS is not, and I, even in my own business, I'm starting to apply the agentic tools of some of these existing platforms. I don't know if you are, but I thought that was kind of interesting. I just wanted to share that as we talk about what's going on in service now.

Patrick Moorhead: 

Well, it is good. Yeah, we're a NetSuite user as well, because you know, we're gonna end up being a billion dollar company someday. So you have an ERP system that that scales. But no, this follows the classic line, which is, hey, you're either going to integrate killer AI features into your SaaS platform, or you're going to be built around. Now, I used my team up at, we're still doing this because we haven't rolled out NetSuite Next is, is we essentially wrapped Anthropic around NetSuite and did a lot of these different things. But quite frankly, it would be a lot easier if it was built in and whether that's, you know, financial close, EPM plan. Actually, I don't do EPM planning. but essentially being able to close a lot quicker. Because I don't have a CRM, having a customer 360 summary, we're interested in doing. I did ping my team. They're impressed, but I don't have my dashboard yet, but leaning on the team. Dan, I'm glad you added that. Well, yeah, just, you know, we talk around it.

Daniel Newman: 

I was like, what are we doing? Here's what we're doing. Anyway, so we want to talk Intel real quick before we shut this thing down. Yeah, let's do that. I mean, fastest revenue growth since 2011, double EPS consensus, they raised their Q3 guide, they lifted their CapEx, but that was the small side of CapEx. And they kind of pre announced that there's going to be a Big swing of CapEx next year. Stocks soared after hours. Jim Cramer tweeted, and now it's down 4% today. The market hates everything right now, AI, it appears. But this was a great print pass. I mean they were above on everything, above on revenue, above on EPS. They're spending CapEx. I go back to the Cisco AI Summit where we said watch CapEx to understand the external customer demand. So all I'm saying is you've got a serious signal in this report when David Zisner came out and said, next year is going to be a big spending year. Again, they're not revealing customers. It's Lipu's way. He's not going to push anything faster than he has to. He's just going to keep executing. But Pat, God, what a signal, right? I mean, can you not, do you not see some serious signal in that, that, that the, you know, not only 18A, you know, we're in risk production, it's moving, but that 14A has got some serious business lined up.

Patrick Moorhead: 

Yeah, I mean, Lippu was very, very clear that I'm not going to apply mega capital unless I know the business is going to be there and we've signed up a customer for 14A. And here we go. So Lippu didn't come out and say, hey, we're pleased to announce NVIDIA Feynman on 14A, but he essentially announced that they do have a customer. Now, the intensity of that increase in CapEx will be in 2027 because these chips are going to be production grade in 2028. The cool part that people are missing is that there's a big boost in 2026 and you only pay, right? You only deploy that CapEx when the machine is delivered. So, you know, we're going to have a lot more machines for wafer and packaging hitting in 2026 and even more for 2027. I think that 2027 number could be double. It could be 40 billion. I would not be surprised. And what that says, they've got an anchor tenant, high volume anchor tenant, as opposed to somebody that's just doing backup a backup wafers. So Daniel, we made it great. Great pod here. We are back. I know everybody was worried that maybe we disbanded it and, you know, hadn't been seen with my bestie in two weeks. Maybe we had a rift. And we were splitting out all our businesses and our friendship along the way. But no, we are back and we are are together. And you can look forward to some flexing gym picks when we get back on the road together. Daniel, we are back on the road next week, which which will be when this airs. So thanks for tuning in. Thanks for being part of our community and hit that subscribe button. Bye bye.

‍