[0:00]In 2025, the world's largest companies reportedly spent around $400 billion on capital expenditures to support the development of artificial intelligence. Adjusted for inflation, that would be nine Manhattan projects or two Apollo programs, all within the space of just one year, and just on the infrastructure alone. Put another way, last year more money was set aside for constructing and fitting out data centers than was spent on building single-family residential homes over the same time. And this number doesn't even include non-public companies like Anthropic or Open AI, which are harder to get reliable financial data on. This number also doesn't include any of the other costs outside of just building and fitting out the facilities themselves, like staffing, energy, security, and, uh, strategic acquisitions like podcasts. These numbers are also only for 2025, and of course, recent announcements suggest that spending this year will once again break new records. Now, the borderline comical numbers being thrown around in the AI industry may not be that surprising to any of you anymore. But it has also been almost four years since this technology has really come onto the scene with the first public release of ChatGPT. In that time, not a single one of these companies has figured out how to turn a profit with this technology, even when using generous financial projections and accounting tricks. The exception to this, of course, has always been NVIDIA, alongside the other hardware suppliers and chip manufacturers upstream of them. The classic analogy that you are probably sick of hearing by now is that all of this may very well be an unsustainable gold rush, but the hardware companies are the ones making reliable profits by selling the pickaxes and the shovels. However, by following the numbers it has raised some questions about where these shovels are actually ending up. At the same time, these companies are promising record levels of new spending on data centers. Reports have indicated that over half of the sites that were supposed to be open this year have either been delayed or outright canceled. Logically, it's very difficult for both of those things to be true at the same time. Even if you completely ignore the question of how their end customers are going to keep on paying for this, there are some other concerning logical paradoxes developing at the same time. These companies can't keep up with demand for new chips, and yet their inventories are growing, they are depreciating their hardware over six years while claiming next year's models will render this year's completely obsolete. And then there's the question of how they plan to power all of this, given the, uh, current state of everything. Nearly half of the U.S. data centers planned for 2026 are reportedly expected to be delayed or canceled. pushback from communities who say they don't want data centers in their backyard. 10 natural gas plants serving this one single data center. says demand kind of being soaked up by by data centers. We never said we were going to invest $100 billion in one round. That never was said. Once we've build the sort of generally intelligent system, basically, we will ask it to figure out a way to generate an investment return for you. Okay, so whether we really like it or not, NVIDIA has become the company holding up the world thanks to its market cap, profit, and reinvestment into feeding the industry that feeds it. As potentially the single most analyzed company in the history of financial markets, I hate to break it to you, but a YouTube video isn't going to unearth something that everybody else was missing all along. However, there are three big questions that are being asked about NVIDIA's place in the current market that are worth understanding, because, well, again, even if you aren't personally invested in this company, any significant change in its price will have implications for the entire economy. So the first big question is, where are all of these chips actually going? In an interview with CNBC last year, Jensen Huang claimed that the company was shipping around 10 gigawatts of GPUs within 2025 alone. Some quick math based on their current product lineup and reported annual sales suggests this might be a slight overestimation. But it is roughly in the right ballpark. Now, this figure was actually taken from an interview where Huang was announcing their partnership with Open AI to invest up to $100 billion in Open AI in order to help them build out over 10 gigawatts of compute themselves, which was again, roughly the equivalent of NVIDIA's entire annual output of GPUs. This, of course, didn't help the accusations of circular dealing. But that's actually not the problem here. The problem was that, according to estimates by Goldman Sachs, there are only around 7.7 gigawatts of AI data centers currently in operation across the entire planet. Now, of course, there are a lot of new data centers under construction that will all need their own complement of fancy NVIDIA GPUs. But there aren't nearly as many as the big announcements might make you think. On the ground research performed by the market intelligence firm Sightline Climate confirmed suspicions first raised by the business journalist Ed Zitron that a lot of data center construction wasn't nearly as far along as the press releases would suggest. Of the 21.5 gigawatts of announced capacity expected to come online before 2027, only 6.3 gigawatts worth of computing infrastructure was actively under construction, and even that makes it sound better than it really is. Under construction could be anything from a data center that is getting its final fit out before going online to a site that has had nothing more than a foundation poured. Now, I will leave a link to all of that investigative research below, as well as some interviews that Zitron has done here on YouTube. Even as we were putting this video together, Bloomberg reported that Oracle and Open AI's flagship Stargate data center in Abilene, Texas, shoved its expansion plans amongst ongoing issues that we will get into later in this video. Now, in the interest of full transparency, across the wider market, some of these numbers are hard to verify because they are coming from a mix of private companies that do not need to provide public accounts and large public companies that can mix in their air related operations with the rest of the business on their financials. But the point is that unless every data center on Earth was replacing its graphics cards every 14 months, NVIDIA's current rate of production would be oversupplying the real market that actually exists right now. Which sounds bad, but it gets worse. Even if we generously assume that every data center verifiably under construction right now goes online within the next eight months and every existing data center updates their current hardware this year, it's not quite as simple as just adding these two numbers together to make 14 gigawatts of total demand. There is no hard and fast industry standard. But since tech companies have started using megawatts and gigawatts to measure the size of their data centers, that is typically included the entire input of the center, not just energy going exclusively into the GPUs. In addition to this, there is also the networking, cooling, storage, and processing overhead. According to the International Energy Agency, a typical data center normally dedicates around 46 to 65% of its energy to just the compute. So, even if we again take the generous high-end estimate and apply it to the generous estimate of every data center under construction going online this year, and then generously assume every existing AI data center will upgrade its hardware, then generously assume that every single last one of these facilities are going to use NVIDIA GPUs, there's just isn't enough of a gap to cram 10 gigawatts worth of GPUs into. So, either Jensen Huang was overestimating how many cards they actually produced last year, or they are ending up somewhere the analysts can't see. And oh yeah, that's just the first problem. So it's time to learn how money works to find out where all of this money and all of these computer chips are actually ending up. This video is sponsored by Monarch Money. Monarch Money is a private ad-free personal finance platform that brings all your accounts into one clean dashboard. It automatically tracks your spending and net worth and helps you stay financially aligned without juggling spreadsheets. Instead of just showing where your money went, Monarch makes it easy to see the full financial picture, make smart adjustments, and make progress towards your goals with confidence. Plus, with bank-level encryption and read-only access, it's secure, and thousands of users have given it a 4.9-star rating. I started using Monarch for my monthly budget check-ins, and it's been a game changer. All my accounts, cards, loans, and investments sync from over 13,000 institutions into one beautiful view on web or mobile, saving me time hopping between apps. I love the flexible budgeting that separates fixed and flexible spending with customizable rules and categories. And the insights explain spending changes or unusual transactions in plain English, helping me stay aware without constant alerts. Seeing my net worth trend and goal timelines for saving or paying down debt makes progress feel real and motivating without the stress. Try Monarch yourself. Go to the link in the description and use code HMW50 for 50% off your first year.
[8:35]Okay, so one thing that may start to sound a little bit confusing when seeing the headlines and the press releases about these new data center developments is why they are measured in gigawatts in the first place. A watt is a measure of power, and when it comes to computers, not all watts are created equal. If you are building a gaming computer, its power draw would probably not be the first spec you focus on, because, well, it just makes no sense. My current Mac mini desktop draws about 15 watts, but in almost every computing task, it would absolutely destroy my 150-watt 15-year-old gaming computer that I just don't have the willpower to upgrade for, um, unrelated reasons. Now, I am not bringing this up to rant about the cost of RAM, but instead, because it relates to the second major problem facing NVIDIA and the expectation that it will keep on delivering more chips every year. And that is this power limitation. According to industry analysis, the biggest bottleneck facing new data centers today is not necessarily in the advanced computer chips to run their models, but rather in getting the electrical infrastructure to support it. Power has become the real-world constraint. So that's how these projects are getting compared. In the same way we might compare FPS in Cyberpunk 2077. In a typical data center, the transformers, power supplies, generators, and connecting wires typically take up less than 10% of the total build cost, but nothing can happen until those components are in place. The price of things like a Proloc transformer have more than doubled over the last four years, and supply struggles to keep up with this new demand on top of supplying regular energy grids for regular households and businesses. It hasn't helped that a lot of these technical components have typically come from China, South Korea, Mexico, and Canada, making tariff disruptions harder to navigate. What this has meant is that since the supply of everything that goes into these data centers has been so limited, companies are buying everything they can, as soon as they can, even if it takes a long time to actually be put into use. If NVIDIA says the latest batch of H200s is ready to purchase, these companies are compelled to buy them even if they don't actually have space to put them yet, because if they don't, then they will have to go to the back of the line again. The same goes for other components like cooling or energy supplies. Even if they don't have anywhere to use them right now, they are motivated to buy them just in case. In industrial planning, this is called the bullwhip effect, and it can help to explain the extreme levels of spending at the same time as very few projects seem to be making much progress. Now, this is a widespread issue across the world, but to really highlight this problem, there are actually two new fully fit-out data centers just up the road from NVIDIA's headquarters in Santa Clara that are sitting completely offline because they are waiting for local utilities to catch up. At the moment, NVIDIA is benefiting greatly from this. They can almost guarantee that as soon as they get their chips from TSMC, they can immediately sell them off again because there will be a buyer ready for it, even if it does end up sitting in a warehouse for a few months. It's a dangerous game that only takes a small demand correction to cause a massive oversupply, but for now, it's been incredibly lucrative. Except that that assumption is starting to strain now. NVIDIA ended its financial year in January, and when their annual reporting was released, they had again had a record year in terms of sales in the top line and profit in the bottom line. But some other numbers started to catch people's eyes, primarily that their inventory had more than doubled from the year before and quadrupled from 2024. If NVIDIA really is struggling to keep up with demand, there really shouldn't be any reason why they are sitting on so much of their own product. This is either suggesting that it's getting harder to move these chips, or more realistically, that NVIDIA themselves are struggling with the same supply chain problems upstream of them, and they are just trying to get as much supply as possible with the confidence that they will eventually be able to sell it to someone. Ironically, when this anomaly was initially highlighted, it was actually an LLM market algorithm that found the disconnect first, alongside the trend of their customers taking a longer time to actually pay for their receivables. So yeah, the power issue was already a major problem. But today there is also the energy issue on top of that. When these facilities are fully operational, they draw a lot of power, and power over time is energy, and energy has already become very expensive over the last few years, thanks largely to these very same data centers. But the war in Iran has, well, I would say poured fuel on the fire, but it's kind of done the opposite of that. Higher energy prices significantly cut into the viability of running these centers and the cash burn rates associated with them. So far, most data centers get their energy directly from the local energy grid, which means now they are going to be paying more money and waiting even longer for capacity to come online. Other newer centers have gotten around this by using their own natural gas turbine generators. But well, natural gas has just doubled in price, doubling the largest line item in their ongoing expenses. It's not great, but it all gets much worse when you consider these issues in the context of the third major problem facing NVIDIA right now. And that is how long these chips are expected to last. Now, not to yank my own crank here, but about eight months ago, in our video on Amazon's AI spending, we highlighted the problem of capital expenditure going towards cutting edge chips, which by their nature, depreciate extremely quickly as newer better models come out. Well, four months later, Michael Burry, of Big Short fame, as well as a handful of other investors put out pieces effectively saying the same thing. The industry standard amongst the big tech companies is to depreciate these GPUs over six years when in reality, they would be lucky to stay operationally viable for three years. This just basically means that they count 1/6 of the purchase price against income to offset taxes, but also more accurately report the true annual profitability of the business. By stretching out depreciation for longer than reasonable service life of these cards makes expenses look better than they really should be for the companies that make up a majority of NVIDIA's demand. In a somewhat confusing rebuttal, NVIDIA themselves actually responded to this criticism by defending their own accounting practices, which wasn't the accusation in the first place. NVIDIA doesn't actually have that many hard assets at all. It is just a chip designer with most of the real hardware being manufactured by their suppliers, primarily TSMC. So nobody really cared about their depreciation schedules. The point was, if companies like Microsoft, Oracle, and Meta had more honest accounting, it would make their profits look worse, which may reduce investor hype around AI investment and ultimately the demand these companies have for even more GPUs coming down the pike. The reason this critique was leveled at NVIDIA specifically is because if Microsoft is pressured into reevaluating this accounting standard, they might lose a little bit of profitability on paper, but they still have a major business outside of this AI stuff. But if they order fewer GPUs, well, that is NVIDIA's entire business. Now, both the ongoing supply bottlenecks and high energy prices make this problem much worse. If NVIDIA is consistently releasing a new model of flagship AI GPUs every year that make everything that's come before it obsolete, then it could become much harder to justify purchasing pallets of current day chips in advance in the hope that they will be ready when a data center eventually comes online, because by that time, they might already be irrelevant. On the other end of the spectrum, the depreciation of chips already in service becomes worse, the higher energy prices get. If energy prices are very low, it can remain worth it to run older, less efficient hardware, even if it uses much more energy to achieve the same thing. But as energy prices have increased, those margins can be squeezed to the point where it costs more to run a server in energy inputs alone than it could be rented out for. At that point, multimillion dollar racks of last year's cutting edge hardware effectively become e-waste. Now, the market can stay irrational longer than any naysayers can stay solvent, and it is worth realizing that even if the business case makes no sense, it can stay in motion for as long as investors are happy to put money into it. But well, even that might be starting to change. For the last four years, private credit companies like Blue Owl and BlackRock's credit arm have been major financing partners on some of the biggest data center projects, but they are now facing their own industry-wide problems, which is going to make it much harder for them to maintain the supply of easy financing. Go watch this video next to find out what private credit actually is and why it is all gone so badly wrong. And don't forget to like and subscribe to keep on learning how money works.
