Big Tech's Biggest Gamble Is Finally Falling Apart

[00:00:00]Something very significant happened that did not get nearly enough mainstream attention as it should have. S&P Global Ratings, which is one of the three major credit rating agencies in the world, the people whose entire job is to assess whether companies can pay their debts, downgraded Oracle's long-term credit rating from triple B to triple B minus.

[00:00:21]That's the lowest rung of what's called investment grade. In institutional finance, a triple B minus rating is the absolute thin line separating a respectable corporate citizen from the financial equivalent of selling knock-off electronics out of the back of a rented transit van. S&P is essentially telling the markets that Oracle is one bad Tuesday away from paying his bondholders in expired Amazon gift cards. And the reason S&P gave was not Oracle's legacy database business or a bad quarter or some sick management scandal as is so common these days, but just one other company, you guessed it, OpenAI.

[00:01:02]S&P explicitly identified OpenAI as a key credit risk for Oracle, noting that OpenAI accounts for roughly half of Oracle's $638 billion in remaining contractual obligations. When S&P explicitly names a non-profitable startup as your single biggest credit risk, they are dropping the polite corporate code. They're effectively telling the market that Oracle has decided to co-sign a multi-billion dollar luxury car loan for an unemployed friend who promises they're about to hit it big on the stock market. But look, maybe we're being unfair, right?

[00:01:37]Personally, I have complete faith in this arrangement. After all, the company is literally named Oracle. You don't name your entire corporate empire after a mystical entity that can seamlessly peer into the future unless you actually know what the hell is going to happen in 3 years, right? Clearly, Larry Ellison, their CEO, knows something that the rest of us mere mortals don't, or you know, they just really needed a press release to juice the stock price. One of the two.

[00:02:04]In this video, I'm going to explain what Oracle and OpenAI's deal actually is and why the downgrade happened, what it means if things actually get worse, why OpenAI is still a company with genuine potential despite everything, and why the word bubble, which everyone seems so certain of, is actually a beautiful optical illusion hiding the real disaster underneath.

[00:02:27]To understand why S&P is nervous, let's first understand what Oracle has committed to.

[00:02:33]Oracle is one of the oldest enterprise software companies in the world, founded in 1977, built on database management systems, cloud infrastructure, and enterprise licensing.

[00:02:44]For decades, it was the plumbing of corporate IT, but Oracle's core business has been flat or declining for nearly a decade now. The database licensing revenue that built the companies was plateauing, cloud computing was growing, but Oracle was a distant competitor behind Amazon Web Services and Microsoft Azure. Larry Ellison, founder, chairman, and the company's largest shareholder with roughly 40% of Oracle stock, needed a growth story, and he found one in OpenAI.

[00:03:15]In early 2025, Oracle became the primary infrastructure builder for Project Stargate, which is OpenAI's plan to construct the largest AI data center network in history. The scale is difficult to overstate, approximately 7.1 gigawatts of data center capacity, costing an estimated 340 to 400 billion dollars to build, funded [snorts] by a roughly 300 billion dollars contract over 5 years, in which OpenAI pays Oracle for compute capacity as it comes online. The logic behind the bet is very straightforward. If AI compute demands grows the way the industry predicts, whoever owns the physical infrastructure, the data centers, the power connections, the cooling systems, the GPU clusters, becomes the landlord of the entire AI economy. It's kind of like owning the railway tracks during the Industrial Revolution. You don't really need to know which train company wins, you just need to own the tracks.

[00:04:12]It's reasonable. Oracle saw Stargate as its chance to leapfrog AWS and Azure in a single move. And Ellison, who is by the way this is a man who races yachts competitively and bought an entire Hawaiian island, said yes. And to be fair to Larry, if you already own an entire island, standard investments like low yield treasury bonds just don't really give you the same dopamine hit anymore. Building a $300 billion supercluster for an unprofitable startup is the corporate executive version of just going to Vegas. The problem is the gap between what Oracle has to spend and what Oracle gets back. Oracle has to commit the capital up front to build the data centers. OpenAI pays over time as capacity becomes available. S&P now forecasts Oracle's capital expenditure for fiscal 2027 at 90 to 95 bill dollars, up from a prior estimate of 60 billion. The projected free cash flow deficit has widened to negative 42 billion dollars against an earlier projection of negative 24 bill. Oracle currently carries approximately $165 billion in total debt obligations. And the revenue to service all of this depends on one company, OpenAI, which lost 20.9 billion dollars operationally in 2025, has never generated a profit, has postponed its IPO to 2027, and is watching its market share decline below 50% for the first time. Well, that might have to do something with a a CEO named Ben Saltzman, but we'll get into that.

[00:05:48]In essence, Oracle's primary plan to secure its financial future relies entirely on the cash flow of a tenant that burns through $20 a year and has the revenue stability of a teenage crypto influencer. It is a landlord-tenant relationship where the landlord is building a luxury skyscraper for a tenant whose currently monthly income consists entirely of IOUs written on the backs of napkin. But hey, we have to assume the Oracle crystal ball is working flawlessly here and that they've looked deep into the cosmic timeline to see the exact moment OpenAI suddenly becomes a stable, profitable utility company. Either that or the crystal ball was actually just a reflection of Larry's yacht monitor and everyone in the finance department was just too terrified to tell him. I don't know.

[00:06:36]S&P put it directly, if OpenAI were unable to meet its payment obligations, Oracle would be left with long-term data center rental agreements that could neither be easily terminated nor transferred to other customers on comparable terms. And OpenAI's ability to service its contracts, S&P noted, depends on the AI boom continuing, the models remaining market leading, and the company continuing to raise external capital, which is not considered certain.

[00:07:04]Now, what happens if Oracle gets downgraded one more time? If it drops from triple B minus to double B plus, it enters what the market calls junk status. At that point, Oracle becomes what's known as a fallen angel, which is a company that was investment grade and has fallen below. This matters practically a huge deal because many investment funds, like pension funds, insurance companies, institutional portfolios, are required by their own rules to hold only investment grade bonds. If Oracle becomes junk, those funds would be legally obligated to sell their Oracle bonds regardless of whether they want to. That force selling crashes the bond price, which raises Oracle's borrowing costs significantly, potentially doubling them, at precisely the moment when the company needs to borrow tens of billions more to keep building data centers. It's basically the financial equivalent of a Wile E.

[00:08:00]Coyote cartoon. You're sprinting off the edge of a cliff, actively building the bridge underneath your feet using borrowed wood, while the bank is calling you to say they are increasing rate on the nails.

[00:08:12]Oracle is trying to stay ahead of this, of course. It completed a $5 billion mandatory convertible preferred stock issuance in February, and has announced a plan $20 billion equity raise later this year. But a separate bondholder lawsuit alleges Oracle misled investors about the scale of its planned debt issuance around the time the OpenAI contract was announced. Oracle shares are down 28% year-to-date, trading near their 52-week low. It's a lot of drama, S&P also made a comparison that should worry Oracle shareholders. Unlike AWS, Google Cloud, and Microsoft Azure, which all have internal workloads to absorb excess capacity and deep financial reserves, Oracle is building capacity almost exclusively for external customers. If demand does not materialize, the others have a fallback.

[00:09:02]Oracle doesn't, which, if we're being completely real, sounds terrifying. It's the kind of high-stakes corporate tightrope walk that gives corporate treasurers stress dreams. This is a company that has taken enormous risk.

[00:09:14]The risk is real and documented, not by analysts on podcast, but by the credit rating agency whose job it is to assess exactly this kind of thing. I'm not going to tell you Oracle is doomed, but I am going to tell you that the numbers are genuinely precarious, and the trajectory requires things to go right in ways that the people assessing the situation professionally are not currently confident about. But hey, what do the credit rating agencies know, right? S&P is looking at spreadsheets.

[00:09:42]Oracle is looking at the literal fabric of time. Surely the ancient prophets of database software didn't accidentally sign a contract that could drag them into junk status. That would imply a profound lack of foresight from a company whose name is literally a synonym for foresight. Their username checks out. Now here is where I think most of the commentary gets lazy because it's oh so easy to look at Oracle situation, look at OpenAI's losses and conclude that the whole thing is about to collapse. But that conclusion requires you to ignore several things that are actually true. OpenAI's revenue tripled from 3.7 billion dollars in 2024 to 13.07 billion dollars in 2025. That growth rate, while decelerating, is still extraordinary by any standard.

[00:10:28]ChatGPT remains the single largest AI assistant by market share even having fallen below 50%. The models themselves, personal feelings about leadership aside, remain competitive across a range of use cases. The GPT-4 family was, in my personal view, a superb piece of work. Some of their more recent releases have been less impressive to me and I've been vocal about that in previous videos, but the underlying research capability is genuine. I've covered OpenAI's trust failures at length in this channel, the silent model rerouting, the sycophancy, the NDAs, the leadership decisions that have cost the company talent and public goodwill. I stand by all of that, but personal dislike for leadership is not the same as a death sentence for a company.

[00:11:13]Companies survive bad leadership. They survive trust crisis. They survive market share erosion. What they don't survive is running out of money. And OpenAI, for all of its losses, still has approximately 50 billion dollars in assets and access to the deepest pool of venture and institutional capital in technology history. Could Open AI fail?

[00:11:34]Yes. Could it fail to pay Oracle?

[00:11:36]Totally. Oracle's own annual report apparently concedes this possibility.

[00:11:40]Could a change of leadership, a renewed focus on the models that actually work, and the market share it still holds be enough to turn the trajectory? Also, yes. But I want to resist, and what I think we should all be trying to resist, is the temptation to treat this as a binary outcome. It is really not Open AI thrives or Open AI dies. There is a vast spectrum of possibilities between those two endpoints, and most of the interesting analysis, along with the eventual truth, lives somewhere in the middle on that distribution. Oracle's bet might partially pay off. Open AI might grow enough to service some of its contracts, but maybe not all of them.

[00:12:18]The data center capacity might be repurposed. The industry might consolidate. The technology might advance in ways that generate revenue nobody currently forecasts, which brings me to the thing everyone keeps saying. I need to talk about this word because it's everywhere, and it's starting to bother me. AI is a bubble. The AI bubble is about to pop. We're in the biggest bubble since dot com. I hear it in videos. I read it in articles. I see it in my comments every single day, and I really understand the appeal. There is something deeply satisfying about predicting that the arrogant will be humbled, that the inflated will deflate, that the people who've been insufferable about AI for 4 years will finally get what's coming to them. I really get the schadenfreude, genuinely.

[00:12:59]But what is a bubble, exactly? I speak with finance professionals on a daily basis, probably more than is good for me. These are the kinds of people who evaluate risk and price assets for a living, and the answers are always a lot less confident than the headlines suggest. They don't typically use the word bubble in professional conversations. They talk about overvaluation, underpricing, fundamentals, risk-adjusted returns, concentration risk. Bubble is a the for headlines and pub conversations. So, of course, like any nerd, I went to the academic literature. What I found is that two Nobel laureates in economics cannot agree on whether the concept even exists. Eugene Fama, the father of the efficient market hypothesis, has said, "I don't even know what a bubble means.

[00:13:43]These words have become popular. I don't think they have any meaning." His position is that markets reflect available information. Sometimes participants are wrong, but being wrong is not the same as being irrational.

[00:13:56]Prices incorporate uncertainty, and uncertainty, [clears throat] of course, includes the possibility of being spectacularly incorrect.

[00:14:04]Robert Shiller, who coined the term irrational exuberance, argues the very opposite, that bubbles are real, driven by narrative psychology, and identifiable in their broad contours.

[00:14:16]But, even Shiller acknowledges that you cannot reliably identify a bubble in real time. You can only confirm it retrospectively after the correction.

[00:14:25]And these two positions are not as incompatible as they sound. Humans are contradictory creatures by nature. A decision that contains hope and emotion is not automatically rational. Venture investing, which is the entire practice of funding companies that don't yet have products, revenue, or customers, is literally the act of paying for something that doesn't exist based on a probability weight assessment of whether it might. It's not a bubble. That's how innovation has been funded for centuries. The question is always whether the probability assessment was reasonable, and you can only find it out exposed.

[00:15:02]The economist Hyman Minsky offered a framework that I think is more useful than the word bubble. He described three stages of finance: hedge finance, where you can pay both interest and principal from income; speculative finance, where you can pay interest but need to refinance the principal; and Ponzi finance, where you can't pay either, and you need the asset to keep appreciating just to stay solvent. The AI industry, by this framework, is arguably in the speculative phase. Companies can service some obligations, but need continuous growth and external capital to sustain the structure. Whether it tips into positive territory depends on whether revenue catches up to the commitments.

[00:15:42]That's a useful, specific, and answerable question. Is it a bubble?

[00:15:46]It's not.

[00:15:47]And here is where my own position lands, and I want to be very transparent about it. Yes, AI valuations look high relative to current revenue. That is measurable and not seriously disputed.

[00:15:58]But something is only definitely a bubble in retrospect. Right now, what I see is a probability distribution. There is a scenario, and it's not a negligible one, maybe 20%, where a breakthrough architecture, a new methodology, an unexpected application of AI generates revenue that justifies these valuations or exceeds them. The Putnam fellowship scores arriving 7 years ahead of expert predictions, the Erdős conjecture being disproved by an AI model, the 100 times energy efficiency gains from neuro-symbolic approaches, these breakthroughs prove that AI capabilities can arrive decades earlier than anyone forecasts. The option value of that possibility is not zero, and pricing that option value into current valuations is not insanity. It's just how markets work. Consider the housing market. Is housing a bubble? Prices have been unsustainably high for the last 50 years. At what point does a bubble become just what things cost? Consider Amazon in 1999. The stock dropped 93% after the dot-com crash. Pure bubble, right? Except if you held through the crash, you eventually made 200 times your money. The internet was the future.

[00:17:09]Most of the companies riding the hype are not, but the one that survived transformed everything. So, is AI overvalued? Probably relative to today's revenue, yeah. Is it a bubble? Nobody knows. Not the analysts, not the Nobel laureates, not the finance professionals, and certainly not anyone on the internet who seem very confident about it. Anyone telling you with certainty that it is a bubble or isn't a bubble is selling you confidence they do not have unless they're Oracle, who of course can see the future. The honest position is uncomfortable. It doesn't give you a team to join or prediction to make. It just says the risks are real, the potential is real, and the gap between them is where the actual story lives. But, let's just strip away the bubble debate for just a second. There is a behavioral signal right now that I think is more telling than any credit rating or Nobel laureate. According to the Wall Street Journal, multiple data center owners are actively trying to sell majority stakes in operating companies worth tens of billions of dollars to, wait for it, private equity firms. These are the people who build the infrastructure. They are closer to the physical reality of AI compute than any analyst, any journalist, or any YouTuber. They know exactly how much power these facilities draw, how far behind schedule the construction is, and how dependent the revenue is on a handful of companies that are not yet profitable, and they're trying to get out. If Jensen Huang is right that we're at the beginning of the cycle, that AI infrastructure demand is about to explode, then selling your data center right now is like selling beachfront property the week before somebody announces a new motorway to the coast.

[00:18:49]You just don't do that. You hold. The fact that they're selling right now tells you either they don't believe demand is coming, or they believe demand is coming, but they cannot survive long enough to see it. Either way, it's not really the kind of behavior of people who think they're at the beginning of anything. And companies can be right about the future and still go bankrupt waiting for it to arrive, by By way.

[00:19:10]That's exactly what happened to dozens of fiber optic companies during the dot com crash. They built the cables that the internet eventually needed and they went bust five years before the demand caught up. Somebody else bought the cables at a discount and made fortunes.

[00:19:25]The infrastructure was right, but the timing was wrong, which means the absolute worst case scenario for Larry Ellison is that he spends $90 billion building the physical temples of the AR revolution only to go bankrupt right before the god arrives, leaving some random liquidation fund to buy the world's largest GPU cluster for $40 and a pack of biscuits, using it exclusively to host a massive, highly optimized Minecraft server. To be very honest, I'll keep my eye out for that one because I'll throw in $40, biscuits, and a dirty martini along with those no sunglasses for Larry to hide his tears.

[00:20:01]Both outcomes are possible, neither is certain, and the honest thing to do is hold both in your head simultaneously rather than collapsing into whichever narrative feels more satisfying. Because if Oracle fails, that's not just Larry Ellison losing money. That's tens of thousands of employees. That's pension funds holding Oracle bonds. That's communities built around construction projects. Wanting it to fail because the CEOs are annoying is understandable as an emotion. It's irresponsible as a position. Sorry that I had to say that, but I generally think it's true. The honest position is uncomfortable because it doesn't give you a side. The valuations are high. The risks are real.

[00:20:42]The technology is extraordinary. The leadership in many cases is poor. The potential is genuine. The gap between spending and revenue is alarming. We are all in a real world, messy, complex cluster F of contradiction and nobody, not the credit rating agencies, not the Nobel laureates, not the finance professionals I speak with regularly, and certainly not some computer scientist with a YouTube channel, can tell you with certainty what happens next. What I can tell you is that the word bubble has become a substitute for thinking. It lets you skip the hard work of evaluating each company, each technology, each decision on its merits, and just say it's all just going to pop.

[00:21:21]It's not analysis, my friends, it's a mood, and moods don't build anything as beautiful as they are. Oracle is in a very precarious position, that's a fact.

[00:21:30]OpenAI's ability to pay its bills is uncertain, also fact. The AI industry capital commitments exceed its current revenue by an extraordinary margin, fact. But the technology is also advancing at a pace that the people building it did not predict, and the possibility that those advances generate the revenue to close that gap is not zero, fact. Bears eat beets. Yikes, wrong video again. Hold all of those truths at once. Please try to resist the urge to collapse them into a prediction.

[00:22:01]The gap between what AI costs today and what it might generate tomorrow is where the real story lives, and anyone claiming they know how it ends is telling you more about their temperament than about the future. If you want to understand what OpenAI's leaked financials actually reveal about the company at the center of all of this, the $21 billion operating loss, the $5.7 billion marketing spend, and the trust crisis that's costing them market share, I covered that in detail in this video that I'm linking here on your screen.

[00:22:30]Thanks so much for watching this one.

[00:22:32]I'll see you on the next one.