Latest / Super Intelligence News and Report / Nvidia's $170B Neocloud Bet Meets a Credit Squeeze as Anthropic Nears Its IPO
Transcript
- Super Intelligence News and Report — Nvidia's $170B Neocloud Bet Meets a Credit Squeeze as Anthropic Nears Its IPO Welcome to Super Intelligence News and Report your daily look at what's happening in super and artificial intelligence. Today is Sunday, October 25th, here are the headlines. Host 1: Welcome to super intelligence news and report for Sunday October 25th we have a ton of ground to cover starting right off with some major policy moves Host 2: yeah specifically down in Texas because on October 23rd Attorney General Ken Paxton actually sued OpenAI and Sam Altman personally Host 1: wait personally personally I mean what is a state AG going after Altman for Host 2: so it's basically a clever legal backdoor the state filed a 67 page petition citing eight specific violent incidents Host 1: Okay. Host 2: And they officially filed it under consumer protection law. You know, but they are seeking a court appointed evaluator for any new models. Host 1: Oh, I see where this is going. Host 2: Right. So Texas is trying to build a state licensing system for frontier models through consumer protection law. And just for some impartial context here, Elon Musk spent nearly $200 million backing Paxton's campaigns. Host 1: Wow. Okay. Well, speaking speaking of intense pressure, Anthropic is really really feeling it on the product side right now. On October 24th, two key leaders on the Claude code team, Tariq Shahipar and Anthony Morris, they both just left. Host 2: Right. Host 1: And concurrently, OpenAI just aggressively cut their prices and they actually took 52% of Rippling's coding work straight from Anthropic. Host 2: Yeah. And the timing on that is just brutal. Because Anthropic's S1 is likely due likely due today, October 25th, for a mid-November IPO. That's reported not confirmed, by the way. Host 1: Right. Host 2: But OpenAI is basically aggressively squeezing Anthropic's margins right as investors prepare to examine their books. Host 1: Man, that is rough. Host 2: But wait, let's jump back to policy for a second. Britain just announced something huge on October 24th, right? Host 1: Yeah, a massive AI package, over 600 million pounds. Host 2: And 415 million of of that goes straight to security. Which includes a 100 million pound biosecurity lab. Host 1: Which sounds great on paper Host 2: it does but anthropic and open ai are actually actually providing up to 200 million pounds of that package in the form of form of free model credits Host 1: wait Wait what so the UK is heavily funding AI security but relying on relying on free credits from the very labs they are supposed to evaluate Host 2: Exactly. Host 1: That raises major independence questions. And frankly, Anthropic's week just week just keeps getting worse. On October 23rd, the Elvis Presley rights holders sued them. Host 2: Over training data, I assume? Host 1: Exactly. Over 274 songs allegedly used in Claude's training. And it's the exact same lawyer, Richard Bush, who also recently sued them over 853 reservoir songs. Host 2: That is a lot of legal trouble. Host 1: Yeah. Anthropic faces mounting high profile profile copyright lawsuits right before its IPO prospectus drops. Host 3: It's just pressure from every single angle. Host 1: It really is. But, you know, Anthropic's financial pressures mirror a massive widening AI financing squeeze across the whole industry, which brings brings us to our main story. Host 2: Right. Because NVIDIA is facing its first real test as an AI financier. They have roughly $170 billion on the line backing these neo clouds. Companies like CoreWeave, Nebius and IREN, plus the labs. Host 1: That is massive exposure. Host 2: It is, especially when you factor in a guarantee of up to $105 billion for OpenAI's new Ohio campus. But the cracks are really showing. Host 1: Yeah. Host 2: In October, CoreWeave's bond yields topped 15% and their credit default swaps passed 11%, which heavily signals distress. Host 1: Okay, but let me push back on the panic here. Let's compare NVIDIA to a casino bankrolling its own high rollers. If the gamblers go bust, does the house necessarily lose? Host 2: Well, they are holding the bag on the debt. Host 1: Sure. But H100 rental costs are actually up 44% sitting at $2.88 an hour if demand was crashing prices would drop doesn't this just prove we have a capital shortage not a customer shortage? Host 2: You know I actually agree with that read Amazon's Amazon's Andy Jassy just confirmed AWS is completely sold out into 2027. Host 1: Right. Host 2: And NVIDIA is reportedly already in talks to invest hundreds of millions into Firmus. Again, that's reported, not confirmed. Host 1: So the demand is absolutely ironclad, but the credit markets are just terrified of the debt. Host 2: Exactly. Which leads you with a final thought to mull over. If the tech undeniably works, and compute demand is completely sold out, but capital markets suddenly refuse to fund the debt, who ultimately ends up owning the infrastructure of the future? Host 1: Right now, today, the most dangerous risks in in the artificial intelligence boom are being like quietly and meticulously shuffled onto somebody else's balance sheet. Host 2: Yeah, it is effectively a massive game of financial hot potato. potato. And well, the music music is definitely starting to slow down. Host 1: Right. And if you you want to know who actually survives this next phase of the AI build out, you know, just looking at parameter counts and benchmarks, it just isn't going to cut it anymore. Host 2: No, not at all. You have to follow the debt you have to follow the lawsuits. Host 1: And the unconfirmed IPOs. Host 2: Right. Host 1: Because the overarching theme across the entire industry right now is just the suffocating cost of capital. Host 2: Exactly. Host 1: So, for today's deep dive, our mission is is tracking exactly where this risk is hiding. We are looking at NVIDIA effectively acting as like the AI industry's central bank. Host 2: Which is wild to think about. Host 1: It is. Plus this cutthroat price war between OpenAI and Anthropic as cash gets tight and the mounting legal liability spanning from Texas to the UK all the way to Elvis Presley's estate. Host 2: Which I know sounds completely random, but we will get there. We're We're seeing the underlying mechanics of how this industry actually funds itself being stress tested in real time. Host 1: Yeah. Host 2: And to really understand how precarious this structure is, you have to start at the bedrock, you know you have to look at the hardware layer specifically NVIDIA. Host 1: Right, because over the past two years, NVIDIA didn't just sell chips, they quietly became a lender. Host 2: Yeah, a massive one. Host 1: And this wasn't some minor side project for them. It became core to their entire business model. Host 2: Well, when you look at the macro environment, NVIDIA was operating in, you can kind of see the strategic necessity of this pivot. Because NVIDIA's biggest traditional customers, the hyperscalers, Host 1: So companies like Amazon, Google, Microsoft, Meta, Host 2: right, exactly. Host 1: Right. Host 2: They are all aggressively designing and deploying their own custom silicon. Amazon has Tranium, Google has their TPUs. TPUs. Host 1: So NVIDIA basically saw the writing on the wall. Host 2: They did. To sustain their own explosive growth, they couldn't just rely on those four massive tech giants forever. They had to cultivate a completely new second wave of buyers. Host 1: Which is where the neo clouds enter the picture. Host 2: Yes. Host 1: And just for anyone, you know, not totally in the weeds on cloud architecture, neo clouds are specialized data centers. So companies like CoreWeave, Nebius and IREN. Host 2: Right. Host 1: And they are built almost exclusively to rent out heavy duty AI computing power. They don't do everything under the sun like like AWS does. Host 2: No, they just do AI compute. Host 1: Exactly. But setting these up requires literally billions in upfront capital. Host 2: And those newer smaller players simply didn't have the balance sheets to buy you know hundreds of thousands of H100 GPUs. Host 1: They just They just didn't have the cash Host 2: Right so NVIDIA essentially had to fund these buyers these buyers so that those buyers could turn around and afford to buy NVIDIA's own hardware. Host 1: The dynamic here is just wild. Host 2: Yeah I mean it's like a car dealership personally personally guaranteeing your bank loan just so you can buy their most expensive top of the line of the line sports car. Host 1: Yeah that's actually a great way to think about it. Host 2: Yeah. Host 1: But the absolute only way you can generate the cash to pay back that loan is if you drive that specific sports car for Uber 24-7 without ever taking a single break. Host 2: It's spot on. and that analogy holds up incredibly well when you you look at the mechanics of what NVIDIA actually did. Host 1: What did they actually put on paper? Host 2: Well, they Well, they took equity stakes in these neo clouds, they guarantee leases, meaning NVIDIA legally promised to step in and pay the landlords of these massive data centers if a tenant like like CoreWeave defaults. Host 1: Wow. Host 2: Yeah. They They promised to buy back unsold computing capacity. They even set up this this 500 billion dollar financing framework working directly with large asset managers to grease the wheels. Host 1: So the scale of the exposure here is just staggering. NVIDIA NVIDIA currently has roughly 170 billion dollars quote unquote on the line for these neo clouds and AI labs. Host 2: Right. But But we do have to look at the nuance here because these are mostly contingent up to amounts. Host 1: Meaning Meaning they aren't cutting a check for that entire amount tomorrow. Host 2: Exactly. They only actually pay if a counterparty defaults. But still, I mean, the theoretical liability is massive. That 170 billion includes up to 105 billion dollars just to backstop OpenAI's planned 10 gigawatt campus in Ohio. Host 1: 105 billion for one site. Host 2: Yeah. And think about the mechanics of that specific Ohio deal. That site is contractually bound to use NVIDIA hardware for at least 20 years. Host 1: Wow, 20 years in tech is basically an eternity. Host 2: It really is. And on top of that, NVIDIA committed to buy up to 6.3 billion dollars of CoreWeave compute if CoreWeave somehow can't sell it to outside customers. Host 1: Which brings us to a statistic that honestly it completely recontextualizes NVIDIA's recent earnings. A massive 45% of NVIDIA's Q2 data center revenue came from these non hyperscaler customers. Host 2: Yeah. Host 1: That is roughly 40 billion dollars in a single quarter. So NVIDIA's NVIDIA's revenue is heavily inextricably tied to the exact same companies they are propping up with these debt guarantees. Host 2: And this is exactly where the squeeze is happening right now because if we look closely at CoreWeave they are flashing some really serious warning signs in the credit markets. Host 1: Okay, what kind of warning signs? Host 2: Well, their credit default swap spread just rose past 11%. Host 1: Okay, let's break down why that specific metric matters so much for them. Host 2: Sure. So a credit default swap or a CDS, is essentially an insurance insurance policy against a company defaulting on its debt. Host 1: Like if I'm worried you won't pay me back, I buy insurance on your loan. Host 2: Exactly. So an institutional investors get nervous about a company's ability to pay, the cost of that insurance goes up. For CoreWeave's spread to pass to pass 11% Host 1: That's high, right? Host 2: That is universally seen in the financial financial world as distressed territory. The market is pricing in a very real tangible risk of failure. Host 1: And you see the exact same panic in their bond yields, don't you? Host 2: Oh, Oh, absolutely. They are topping 15% on their 2031 and 2032 bonds. Host 1: 15%. Host 2: Right. And keep the context in mind here. Those bonds carried an original coupon rate of just 9.75% when they were issued. Host 1: So lenders are now demanding a massive premium because they perceive the risk of holding CoreWeave debt to be like exponentially higher. Host 2: Exactly. and to put that in perspective CoreWeave's own Q3 guidance projects 860 to 950 million dollars in interest expense. Host 1: Wait Wait just to service the debt almost a billion dollars. Host 2: Just for the interest Host 1: just for the interest and what is their projected operating income to cover that? Host 2: just 200 to 260 million dollars Host 1: look we need to have a real argument about what this this implies for the industry because I am looking at these exact same numbers and I am firmly in the bear camp here Host 2: okay lay it out Host 1: this entire structure looks like a house of cards that that is structurally primed to collapse and we are already seeing the foundational cracks the capital markets the actual avenues for these companies to get to get fresh cash they're slamming shut right in front of our eyes Host 3: you mean the the IPO window closing Host 1: yes SK Hynix a massive player they just pulled a 150 billion dollar IPO for its flash memory unit Firmus another Neo cloud cancelled their IPO outright Host 2: Right the equity rods are getting tough Host 1: Memory and storage stocks across the the board have bled 1.3 trillion dollars from the recent highs the equity route is vanishing precisely at the moment their debt is getting exponentially more expensive. Host 2: I hear you Host 1: CoreWeave is drowning in interest payments they cannot cover with operating cash flow. If If compute demand softens even a little bit bit or honestly more likely if these GPUs lose their value as collateral over time Host 2: which computing hardware historically always does as faster chips come out Host 1: Right then NVIDIA's guarantees are going to get called I'm sorry but this is no longer a story about about technological breakthroughs it is a terrifying financing story about bad debt debt Host 2: okay I see exactly why you're connecting those dots but you're completely ignoring the actual product layer here. Host 1: How so? Host 2: Let me lay out the bull case because what we are looking at isn't a fundamental technology failure it is just a highly expensive refinancing phase. The demand for the actual compute the end product is unequivocally measurably there. Host 1: But is it enough to outrun the interest rates? Host 2: Look at AWS their CEO Andy Jassy just stated publicly that Amazon's cloud cloud capacity is entirely sold out for 2026. It's largely sold out for 2027 and a good chunk of 2028 is already committed. Host 1: Wow already? Host 2: Yes. And he explicitly noted that even with 220 billion dollars of capex this year Amazon is still leaving customer demand on the table. They literally can't build fast enough. Host 1: Okay but Amazon is a hyperscaler with like infinite cash. We're talking about the vulnerability of the neo neo clouds Host 2: then look at the neo cloud pricing mechanics. The cost of renting an NVIDIA H100 from a from a neo cloud on Silicon Data's index it just hit $2.88 an hour. Host 1: Okay and what was was it before? Host 2: Well that price is actually up Host 4: 44% since December. In any market, rising rental prices are the exact opposite of what a demand collapse looks like. Customers are desperate for compute and they're willing to pay a premium for for it. Host 5: Right. Okay, but But what happens if a Neocloud simply collapses under the weight of its own interest payments before it can fulfill those rentals? Like CoreWeave? Host 4: Then you look at the contingency plan NVIDIA's CFO Colette Kress outlined. If a Neocloud goes bankrupt, NVIDIA can simply step in, repossess those chips and resell them to someone else. Host 5: Because the latent demand is just that high. Host 4: Exactly. The demand in the market is deep enough to absorb it. Gene Munster at Deepwater Asset Management points out that NVIDIA's forecasting 700 billion dollars in revenue next year. Host 4: They have the sheer unprecedented scale to safely finance these customers, through a tight capital environment. Host 5: So synthesizing this, the true bottleneck bottleneck we are hitting isn't a lack of enterprise customers wanting to use AI. AI. Host 4: Right. Host 5: The The bottleneck is the suffocating cost of working capital required to bridge the gap between buying the chips and renting renting them out. Host 4: Precisely. The research firm TS Lombard actually ran a worst case scenario stress test on this exact dynamic. Host 5: Oh, what did they find? Host 4: They modeled catastrophic 20 trillion dollar equity crash coupled with 125 to 190 billion dollars in direct credit losses on AI related debt. Host 5: Which sounds incredibly apocalyptic for the market, like 2008 levels. Host 4: It sounds devastating. Yeah. But their conclusion is what really matters here. They expect that blast radius would land almost entirely on tech stockholders. Host 5: Oh, really? Host 4: Yeah. It would be incredibly painful for venture capital and tech equity. Absolutely. But it would likely be isolated from the broader broader global economy. Host 5: So it's like a contained financial financial explosion within the tech sector, not a macroeconomic contagion like the housing crisis. Host 4: Exactly. But if the hardware middlemen are suffocating under the cost of capital, that exact same financial pressure blasts straight up the pipeline to the software layer. Host 5: Which makes sense because the AI labs themselves are are in an absolute death match for cash and market share right now. Host 4: The dynamic is basically identical. Capital is incredibly expensive and these frontier models cost hundreds of millions of dollars just to train. Host 4: Let alone the continuous continuous cost of running inference for users. Host 5: So as a result, the labs are fighting tooth and nail to secure secure the recurring revenue from paying enterprise enterprise customers. Host 4: Yeah, and we are seeing how aggressive that fight has become. Look at the data from the from the HR software company Rippling regarding their internal coding work. Host 5: Oh, this was fascinating. In just a five month window, Anthropic's share of Rippling's coding work plummeted from from 57% down to just just 18%. Host 4: A massive drop. Host 5: And in that exact same window, OpenAI surged surged from 16% up to 52%. Host 4: And when Rippling CEO was asked about that massive shift, he cited one primary factor, and that was cost. Host 5: OpenAI is simply cheaper for for them to use right now. Host 4: Exactly. Host 5: But like, let's look at the the mechanics of why they are cheaper. Is OpenAI intentionally undercutting Anthropic on inference pricing right now? Host 4: You mean effectively operating at a deliberate loss on every token generated? Host 5: Yes. to lock in their own private funding round at that reported 1.25 trillion dollar valuation. Because if you have more runway, you can afford to bleed cash to capture market share. Host 4: It's a classic tech playbook. Host 5: Are they deliberately trying to starve Anthropic of momentum and revenue just weeks before before Anthropic's IPO roadshow? Host 4: Which is a huge question. And we should absolutely note here that Anthropic roadshow timing is reported not confirmed. Host 5: Right, reported, not confirmed. But the strategic timing of this price war is war is impossible to ignore. Host 4: It really highlights a severe structural vulnerability for Anthropic. Yipit data estimates that Anthropic is currently relying on just its top 1% of customers to to generate 50% of its total total spending. Host 5: Wait, half their revenue comes from just 1% of their users? Host 4: Yes. That is an incredibly intense level of customer concentration. If a competitor like OpenAI targets that specific 1% with loss leading pricing, the revenue impact is immediate and drastic. Host 5: And amidst all this intense margin pressure, two key engineering leaders for Claude Code, Tariq Shihapar and Anthony Morris, they just departed the company. Host 4: Right before they potentially have to pitch institutional investors on an IPO. Host 5: Yeah, the timing is rough. Host 4: Now, neither gave a specific reason for leaving, which, you know, isn't unusual in tech. But when you zoom out, and combine executive departures predatory price cutting from your biggest rival and a heavy reliance on a tiny tiny fraction of your user base Host 5: It creates a very turbulent high risk narrative heading into a public offering Host 4: It certainly does Host 5: But operating margins and aggressive competitors aren't the only liability stacking up off balance sheet for these labs. Host 4: Now the legal and legal and regulatory risks are skyrocketing and this is becoming frankly an existential issue for anyone trying to navigate this space Host 5: Right let's look at look at what is happening in Texas. Attorney General Ken Paxton is suing OpenAI and Sam Altman personally. Host 4: Now regardless of where anyone stands on Texas politics politics or Ken Paxton's individual claims claims we have to objectively look look at the structural threat this specific lawsuit poses to the entire software industry. Host 5: Exactly. Exactly. We aren't here to endorse the claims, but to impartially analyze the legal mechanisms being deployed Host 4: Right. So Paxton is suing under the Texas Deceptive Trade Practices Act and the state's SCOPE Act. Host 5: Yeah. And the 67 page petition alleges that ChatGPT was marketed as safe without adequate age checks or parental controls. It actually cites cites at least eight violent incidents involving users as well as the July hugging face breach involving OpenAI models. Host 4: And OpenAI for their part responded by saying they are committed to working with Texas on AI policy for the whole industry not just one company. Host 5: But the true underlying threat in this lawsuit isn't just about consumer protection in the traditional sense of you know fines or warning labels. Host 5: It's the structural remedy the state of Texas is demanding. Host 4: This is the part that fundamentally alters alters how tech companies operate. Host 5: Right. They want an injunction barring OpenAI from releasing any any new ChatGPT models in Texas without prior approval from a court appointed independent evaluator. Host 4: Just think about the mechanics of how software deployment actually works. Tech relies on rapid continuous deployment. Host 5: Pushing updates constantly. Host 4: Exactly. If a single state can legally force an AI lab to submit its core product to a judge's appointee for testing and approval before it's allowed allowed to launch to the public. Host 5: That is effectively a state level licensing system for frontier AI models built entirely through the back door of consumer protection law. Host 4: It would completely bottleneck the deployment cycle. Host 5: It's an incredibly adversarial approach. But then then you look across the pond and you see a completely different, almost contradictory strategy happening in the UK. Host 4: Yeah, the UK just announced a massive AI funding package over 600 million pounds largely focused on security rather than than aggressive restriction. Host 4: restriction Host 5: Right. About two thirds of that funding is earmarked specifically for security initiatives. Host 4: And the biggest single chunk is 165 million pounds dedicated to a national AI biosecurity initiative. Host 5: Which includes building a a brand new 100 million pound AI biosecurity lab to test these models for for dangerous capabilities. Host 4: Okay, but this is where the economics of AI create a glaring contradiction. The UK government is setting up this watchdog biosecurity lab to police AI bio risks. Host 5: Yeah. Host 4: but that very same lab is being supported by up to 200 million pounds in free free model credits donated by OpenAI and Anthropic. Host 5: Right, the very companies they are supposed supposed to be testing. It creates a fascinating conflict of interest. Host 4: It's like an umpire in a baseball game game whose salary is being paid directly by the pitcher. How structurally independent can a government watchdog actually be when it's absolute absolute biggest operational input, 200 million pounds worth of compute and API access is a handout from the exact same companies it is mandated to police. Host 5: It's a huge paradox. Texas is trying to build a legal wall and the UK is basically moving into an apartment funded by the labs. Host 4: it just highlights the sheer practical difficulty governments face right now. The technology is so computationally expensive to run and the technical expertise required to evaluate it is so highly concentrated that regulators literally cannot afford to cannot afford to stress test the models without heavy subsidies from the model builders themselves. Host 5: Which creates a regulatory gray area that institutional investors are definitely going to flag in future future risk prospectuses. Host 4: Oh, speaking of prospectuses. Host 6: yes. Host 7: Let's talk about the absolute cherry on top top of Anthropic's potential IPO IPO risk factors. Host 6: Ah, yes. the king Elvis Presley Host 7: the copyright lawsuits are relentless and they're starting to target the core training data Host 6: Yeah, Authentic Brands Group, which is the entity that controls Elvis Presley's catalog, is officially suing Anthropic. They claim claim the Claude model was trained on 274 Presley songs without any permission or licensing agreement. Host 7: And the math on that gets scary fast. At a statutory cap of $150,000 per willfully infringed work, work, that's a theoretical maximum liability of about 41 million dollars. Host 6: Which which stacks on top of a previous suit from the Music publisher reservoir, who is seeking about 151 million dollars over 853 different songs. Host 7: Right. And you might look at those numbers and think, well, well 421 million dollars is a rounding error for a company trying to hit a multi billion dollar valuation. Host 6: But it's Elvis. Juries know that Juries know that name. The public knows that name. It's a massive PR nightmare hitting right at the moment when you're trying to convince risk averse institutional investors that your underlying underlying foundational data pipeline is legally sound. Host 7: And the outcome of all these music related suits Elvis Reservoir Universal Music Group they are currently hinging on a deeply important pending fair use ruling from Judge Yumi Lee in San Jose Host 6: oh interesting Host 7: yeah that specific ruling regarding whether training AI models on copyrighted song lyrics constitutes fair use that will likely set the precedent that dictates the the outcome of all these cascading cases Host 6: So taking this massive web of debt, aggressive pricing and legal threats, what should you actually be watching for in the immediate future? Host 8: There are four key indicators on the immediate horizon horizon that will signal where this is heading. Host 6: Okay, what's the first one? Host 8: First next week's Q3 earnings from the hyperscalers Alphabet Microsoft and Amazon Amazon you want to look directly at their capex guidance Host 6: Because that is the holy grail metric right now Host 8: Exactly Are they still committing tens of billions of dollars to infrastructure build outs or are they they quietly pulling back? Host 6: What's the second indicator? Host 8: Second, we need to to watch whether Nvidia's investment in the Australian Neocloud Firmus actually closes. and And again, to be precise, this investment is reported, not confirmed. confirmed. Host 6: Right. But if it does close... Host 8: It signals that Nvidia is still willing to aggressively inject capital to keep their localized ecosystems moving even as credit as credit tightens. Host 6: Got it. Third? Host 8: CoreWeave's next earnings report. We need to see if that interest expense trend trend worsens. Host 6: Right. If their operating income can't start catching up to their massive debt obligations, they basically become a canary in the coal mine for the entire neocloud model. Host 7: Precisely. And finally, as we discussed, that pending fair use ruling from Judge Yumi Lee in San Jose. which will instantly clarify Anthropic's liabilities heading into any potential public offering. Host 6: Because despite all this data, there is still so much structural risk that we just do not know. Host 7: We absolutely have to be clear about the unknowns. For instance, we do not know exactly how much of Nvidia's roughly 500 billion dollars in off balance sheet commitments actually come due in the near term. Host 6: And we don't know if a major neocloud going bankrupt would flood the market with cheap hardware crashing secondary GPU prices enough to trigger NVIDIA's own resale guarantees. Host 7: Exactly. We don't even know the exact monetary demand in that Elvis complaint because the actual requested figure wasn't made public in the filing. Just the statutory maximums. Host 6: Right. Host 7: And perhaps most importantly for the tech ecosystem, we don't know if the big cloud providers Amazon, Microsoft, Google have enough latent enterprise demand to easily absorb all this extra computing capacity if these smaller debt burdened neoclouds go belly up. Host 6: Which leaves you with this final provocative thought to mull over. We started off comparing Nvidia to a car dealership guaranteeing your loan Host 8: Right Host 7: Right Host 6: but when you zoom out and look at the entire board with NVIDIA backstopping the hardware debt to prop up their own sales, and the AI labs subsidizing government regulators regulators with free compute just so they can be monitored, and the capital markets systematically squeezing out the middle tier startups, Host 8: it paints a very very specific picture. Host 6: It does. Are we actually witnessing the birth of a decentralized democratic tech boom, or are just three or four mega corporations quietly underwriting the entire global AI ecosystem entirely off their own balance sheets? Host 8: That is the trillion dollar question. We'll see you back here tomorrow at the same time time. And please check the show notes for all sources. That's Super Intelligence News and Report for Sunday, October 25th. Sources for every story are in the show notes. If this helped you make sense of AI today, follow the show and share it with a friend who's curious. See you Monday.