Friday Chatter features anonymous conversations between two or three people discussing industry rumors and providing both forward- and backward-looking insights into the market.
INTC
U: Intel. Actually, the two big recent news items are both about Intel and Nvidia, right? One is Nvidia investing in Intel. If you read between the lines, it’s really about NVLink. Intel basically ditches OpenCXL and instead uses NVLink so their x86 CPUs can talk to Nvidia GPUs—on servers and potentially on laptops. That could mean dropping the integrated GPU business Intel has been pouring 20 years into, and maybe even axing their discrete GPU unit too—since now they can just bundle RTX IP on the CPU package. And that’s what triggered the 30% jump in Intel’s stock?
V: I mean, I don’t know. It’s tough—Intel’s fundamentals are pretty bad. Nvidia, the government, and SoftBank throwing cash around might keep Intel from going bankrupt for now, but long-term? Hard to invest in. If you just want to trade Intel, fine—there’s plenty of news flow. People speculate Apple might invest. People speculate a “partnership” with TSMC. All of this feels like déjà vu from 2020 when Pat Gelsinger came back as CEO. Stock ran from $40 to $60 then, and honestly Intel was in a way better position back then. The world was still CPU-powered. Today? It’s GPU-powered. So sure, trade the retail-fueled gamma squeezes if you want. But investing in Intel here? I think anyone saying it’s a good idea is either not being honest… or has insider info. I’ve got neither. I’m not gonna sit glued to my screen trading every headline without conviction. So I’ll pass—I’ve got easier ways to make money.
U: But Nvidia does look smart—at least from a gamer’s perspective. Spend $10 billion, wipe out a competitor. Intel actually had something going with their dGPU business—the B60 launched recently and gamer reviews were positive. Sure, nobody really cares about that market, but I think it’s an incubator: if your dGPU works for gamers, your GPGPU will probably do fine too, since compute is inherently simpler. Look at Apple—their GPU is decent for games, which means their GPGPU can perform too if they want. But Intel basically gave up. Can’t blame them—costs too much to develop. Then along comes Nvidia offering an easy exit. It reminds me again why a $500k Trump Club membership “investment” pays off: you get the insider info to trade on. Pretty exclusive crap though. One news drop, and you’ve made the money back.
V: Mhm. I actually put on a small Intel trade back in May. But end of the day, it didn’t fit my style—I wasn’t comfortable holding it, so I dropped it. And my stance hasn’t changed: Intel is extremely hard to invest in. Compare it to AMD. Back in 2016 AMD was near insolvency—trading at $2. Between $2 and $10 it was basically gambling. They whipsawed until 2017 when they finally launched a competitive product—and then the stock 10x’d. For me, I’d never put half a million into a $2 gamble. But from $10 to $100? That’s an investable range. That’s risk/reward I can commit real capital to. Intel today is not there. You can’t just rely on 10% government handouts in a capitalist market. Companies are valued on competitive products and competitive processes. And so far, Intel has shown neither. It is what it is.
U: Thinking about Apple investing in Intel takes some imagination. Nvidia’s deal made sense—worst case, eliminate a competitor; best case, increase Nvidia adoption. And Nvidia doesn’t even have to commit—they could just recommend Intel CPUs to system integrators while still using their own ARM chips. For Apple, though? They tried working with Intel on modem chips—it flopped, so they built their own C1 chip. So if Apple invests, you’d have to imagine a deal structured like Nvidia’s: worst case, remove Intel as a threat; best case, find some synergy. They’re not just going to throw money in and hope for returns.
V: Yeah, from what I’ve heard—Apple’s never going back to Intel CPUs. They’ve got M-series now. And Apple won’t risk Intel’s process tech either—they’re married to TSMC. The most likely angle is Intel’s advanced packaging. But honestly? That’s negligible revenue. Doesn’t move the needle for Intel in the big picture.
U: And from Apple’s side, there’s really no upside either. Packaging isn’t their cost center. They’re not going to gain leverage against contractors through that.
V: Yeah, I mean, right now it’s just speculation. People say other CSPs will follow. But honestly, CSPs got burned by Intel’s poor execution for years. Hard to see them handing Intel new ASIC manufacturing deals. Big question mark there. So sure—if you want to trade on sentiment, knock yourself out. Sentiment is sky-high right now. But me? I’m not touching it.
OpenAI
U: So Nvidia invested $100 billion into OpenAI. OpenAI’s planning to spend $300 billion on Oracle for cloud. What else is in Oracle’s deal? The whole “Project Texas” thing is still alive from the TikTok side—that’s basically the new TikTok/USDS subsidy arrangement, government-backed. Where should we start? Let’s start with Nvidia investing in OpenAI.
V: I mean, I don’t think it’s a bad investment. Basically, Nvidia covers OpenAI’s short-term funding gap before IPO. In return, OpenAI runs primarily on Nvidia’s stack. That’s essentially vendor lock-in—though OpenAI has said publicly they don’t want a single-vendor lock-in, even after this deal. Still, effectively, that’s what Nvidia paid $100 billion for.
U: Yeah, but I don’t think vendor lock-in is a real story. It’s a market narrative, not reality—OpenAI clearly doesn’t want to be locked in. Still, for Nvidia, it’s a good story and a smart investment. We’ve seen this before: in 2018 with crypto, Nvidia “seeded” the market. They’ve learned how to float demand—if you have money, float all the players so demand is sustained. They don’t need to bribe customers to use Nvidia hardware for OpenAI. Just stuff the supply chain with service providers, and the bubble can run much longer—even if it’s already a bubble.
V: So you’re saying it’s a way for Nvidia to smooth earnings, which I kind of agree with. But I’ve seen news suggesting Nvidia might lease GPUs to OpenAI. Honestly? That’s probably fake. Why would they? Nvidia would rather sell chips to Oracle, CoreWeave, whoever. Then on the side, fund OpenAI to buy capacity from those clouds. Much better structure than leasing GPUs directly.
U: Exactly. Putting depreciating assets on Nvidia’s balance sheet is a bad deal. Even car companies know to spin up finance arms to hold the assets and lease them. Holding those directly is terrible. So no—I don’t think Nvidia will ever lease GPUs themselves.
V: Honestly, this kind of arrangement always gets bad publicity. Everyone says “this is just like the internet bubble,” because Cisco did vendor financing back then and investors got burned. So now people scream “this must be the top!” But come on—it can’t be the top. This financing cycle is just getting started. How could this be the peak already?
U: I’ve been trying to find the right analogy for this build-out. The late-90s internet boom doesn’t capture the scale or madness of today. Here’s one: imagine an alternate history starting in 1965. Everyone in the U.S. buys black-and-white TVs. Broadcast companies pop up everywhere. Now imagine TV makers selling directly to broadcasters, and you rent TVs from the broadcaster instead of buying them. TVs depreciate fast—every 3–5 years a new colored standard rolls out, so they ship you a new set. You just keep paying rent. Run that out through the 1980s—that’s closer to today’s craziness and scale.
V: Well, I mean, why do you finance guys all lack imagination?
U: I don’t lack imagination.
V: No, seriously—finance people especially lack imagination. I had this debate with some Twitter account. He was ranting that Nvidia is doomed: China can’t buy, customers are building their own chips, blah blah. He calls this “the top.” I told him, this is just like Apple in 2018. Back then, China pushed Huawei, told officials not to buy iPhones. Developers were protesting the 30% Apple tax. Everyone was bearish. Every active fund was underweight Apple. And then? Apple 4x’d. Today, same setup with Nvidia—despite being the biggest company in the world, active mutual funds are underweight. That’s a massive analogy. This guy laughed at me: “Haha, you’re saying Nvidia will be $18T?” I wasn’t even putting a number on it. My point is: in four years, with money debasement, fiscal madness, and this scale of build-out, >$10T Nvidia isn’t impossible. And vendor financing? Not new. Happens every cycle. Watch Titans of Finance on Netflix—they talk about JP Morgan financing railroads. He lent money to builders and made them buy steel from U.S. Steel, which he owned. Exactly like Nvidia today. Who got rich? JP Morgan. Or take iPhones: early rollout was subsidized by AT&T and Verizon. They financed Apple and consumers. Smaller scale, more consumer-facing, but same arrangement. Nothing abnormal.
U: It’s not abnormal—but it’s still a terrible deal for the middlemen.
V: Oh, absolutely. Like the neoclouds.
U: They take on massive debt financing customers who frankly can’t afford those services. The assets they buy depreciate fast. In an alternate world, those products wouldn’t even exist—cost-performance doesn’t justify them. That’s why it’s a terrible deal for the middlemen.
V: True—and not true. Because those exact arguments apply to the neoclouds right now: Oracle, CoreWeave, Nebius. They are the middlemen, yes. They’re taking on debt to buy depreciating GPUs. But…
U: It’s just hard to tell right now. They’re not making much money—but they’re also not bleeding heavily either.
V: Exactly. That’s the point—it all comes down to execution. The middlemen aren’t pointless. Their bet is: land the customer first, then keep them, upsell, cross-sell, layer value-add services, and actually charge. Then you’ve got a solid business. There are over 200 neoclouds now—half of them are ex-Bitcoin miners chasing gold rush 2.0. Will most fail? Yes. But if even one or two manage to become “the fourth cloud,” that’s the dream. Too early to call.
U: I don’t actually worry about neoclouds yet. They’re not making much, but they’re not hemorrhaging either. Still, someone has to lose money here. Back to my broadcast analogy: in real history, consumers financed their own TVs, so adoption was gradual. Poor families simply didn’t buy. Broadcasters stayed powerful. But in the alternate timeline, broadcasters financed TVs for everyone—including those too poor. They became the losers. Consumers got the service anyway, and the real power shifted to TV manufacturers. That’s Nvidia today: as long as they can push updates every 3–5 years with step-change performance, demand—deserved or not—keeps the race alive. Power concentrates at the manufacturer. It’s bubbly, sure, and will burst someday. But we can finance this bubble for a very long time.
V: Right. We’ll probably overbuild eventually, and at some point supply will exceed demand—the opposite of today. But are we there yet? Not close. Everywhere you look, demand still far exceeds supply. The debt-fueled phase is just starting—Oracle, Nebius, CoreWeave. And a lot of that capacity is effectively backstopped by giants like Microsoft and Google, i.e., by real free cash flow. The first real “debt-led” stress point is Oracle; and if you don’t believe OpenAI will become highly profitable, then you can question the whole chain. But for now, we’re early in debt-backed financing, with room to run. Oracle literally raised ~$18B in 40-year bonds yesterday—5x oversubscribed. Can they raise more? Probably. Can others? Probably. I don’t see the party ending soon on that front.
U: And a lot of that debt gets inflated away.
V: Exactly—and we’re just starting an easing cycle. Much of this debt is tied to benchmarks that are drifting down. Consensus is ~100 bps of cuts over the next 12 months.
U: And this government has zero appetite for austerity. The plan—from start to finish—is to inflate the debt away. So it’ll be a long time before the debt becomes a problem, if ever. In hindsight we might look at depreciated assets and laugh: “An H100 at $30K? Cheaper than my coffee machine. Maybe cheaper than my toaster.” Not a big deal.
V: Yeah. So for the past month I’ve been moving out the risk curve—hunting for more cyclical, junky, high-beta plays on the AI cycle. The neoclouds: Oracle, CoreWeave, Nebius. I’ve owned Oracle before on the pop. I’m even looking at “dumb” PCB providers. We’re at the upswing; my view is you increase beta here, not reduce it.
U: Seems reasonable. The easing cycle just started and isn’t ending soon. There’s even urgency to take on cheap debt—subject to the rate you can actually get, of course.
V: Sure—everyone has to choose their own path. It’ll be bumpy: you can get random 10% air-pockets in this stuff for no reason. But the bigger picture still supports this view.
U: We’re nowhere near the end of this cycle.