PRAWDZIWE wąskie gardło AI: to nie chipy [Energetyczne wąskie gardło AI]

After 2026, the development of artificial intelligence will be less constrained by access to chips and increasingly by access to energy. Or rather, not just to the energy itself, but to how quickly it can be delivered where it is needed. You could buy all the chips in the world, but if stable power doesn't reach the data center, you won't be able to run a single one of them. In a moment, you will see why energy might become one of the most significant bottlenecks of this decade. And we will look at a company that can provide power in months, while traditional solutions require years. Until recently, the market had practically written it off, but today it is backed by an agreement worth up to $ 25 billion. Welcome to the Rankia Polska channel. If you enjoy our content, leave a like and subscribe to the channel. The partner of this episode is Trade Republic. Currently, you can activate 6%annual interest on uninvested cash up to 200,000 PLN there . And without typical banking complications, this interest rate doesn't just apply for the first three months before dropping to a lower rate. It remains the same the entire time. Details and the link can be found in the video description. Let’s start with the scale of the problem, because it is larger than it might seem. Data centers supporting artificial intelligence are consuming more and more electricity. Globally, their consumption was around 485 terawatt-hours in 2025, and the International Energy Agency estimates it could rise to about 950 TWh by 2030. To have a point of reference, 1 TWh is roughly the amount of energy an average-sized city consumes over an entire year. So we are talking about nearly doubling demand in just five years. And if we look only at the part related to artificial intelligence, energy consumption could even triple. In the United States, the growth could be even more drastic. Data centers consumed about 176 terawatt-hours of energy in 2023. And according to the Lawrence Berkeley National Laboratory, by 2028 they could account for up to 12 %of total electricity consumption in the country. Up to this point, there is nothing particularly surprising here. The fact that artificial intelligence needs huge amounts of energy is discussed practically every week. However, there is one problem that is heard about much less frequently. A data center does not draw power from some abstract, nationwide grid. It must be connected to specific infrastructure on a specific plot, in a specific location, and with specific local constraints. And the biggest problem is increasingly not the amount of energy itself, but the time required to bring it exactly where the data center is being built. Connecting a new project to the power grid in the United States can take from 4 to 7 years. Building a gas-steam power plant takes roughly 4 to 6 years. Today, you have to wait three to four and a half years for a new gas turbine because manufacturers have backlogs of orders, and small modular nuclear reactors, often touted as the future solution for data centers , can require over eight years. Meanwhile, chips purchased in advance sit in warehouses, losing value with each passing month. A billion-dollar data center that lacks power generates zero revenue. It is simply frozen capital. That is why companies building this infrastructure are increasingly asking not just how much a megawatt costs. They are primarily asking when they can get it. As summarized by analysts at OZECO and The Black Line, whose analysis this report is based on. The most valuable megawatt is the one that arrives on time. And there is a company that has turned the speed of energy delivery into its greatest product. That is Bloom Energy. The company manufactures modules that generate power directly at the data center site. Its technology is based on solid oxide fuel cells. In simple terms , the device uses natural gas, biogas, or hydrogen and converts them into electricity through an electrochemical reaction. There is no traditional combustion or moving parts, and efficiency exceeds 50%. However, time is the most important factor. Bloom can install such a system in about two to nine months, instead of waiting several years for grid expansion. In 2025, the company delivered a fully operational system to Oracle in just 55 days. 55 days versus up to seven years of waiting for a grid connection. And in that one number lies the entire advantage of Bloom Energy. The company's own history is also unusual. Bloom Energy was founded in 2001 by KR Sridhar, leveraging experience from technology previously developed for a NASA program aimed at, among other things, producing oxygen on Mars. It sounds impressive, but the road to its current position was very long. For about 15 years, the company watched as one fuel cell competitor after another went bust. Bloom survived. During the hydrogen boom in 2022, the stock price reached about 44 dollars, and later fell sharply. The company debuted on the stock market in 2018 at 15 dollars per share, and just a year later the price approached 3 dollars. Even in February 2025, shares cost about 23 dollars, and the market treated Bloom as a company with no major prospects and little predictability in results. Even analysts complained about a lack of clear direction. The problem was not that the technology did not work. The market simply did not yet have the problem that this technology solved. When there is plenty of time, speed is not of such value. When time becomes a constraint, suddenly you can pay a great deal for it. In the second quarter of 2026, Bloom exceeded one billion dollars in revenue in a single quarter for the first time. The company reached over a billion dollars in sales , which is 166%more than the year before. A company that many investors had practically written off just a short while ago has just posted the best quarter in its history. At this stage, the story might look nearly perfect, but a fast-growing company doesn't always mean a good business. And one of the things that helps distinguish a temporary boom from something more lasting is who is willing to put up major capital for its growth. In October 2025, Bloom signed an agreement with Brookfield, one of the world's largest infrastructure asset managers, worth up to $ 5 billion . Eight months later, in June 2026, the framework of this cooperation was expanded to $ 25 billion. And here, one must be careful with interpreting that number. It does not mean $ 25 billion in revenue for Bloom Energy, nor guaranteed orders of that value. It is about a financing structure for energy projects related to the development of artificial intelligence. Brookfield provides the capital and the ability to scale projects. Thanks to this, the end customer doesn't have to finance the entire installation upfront from their own funds. According to the report mentioned earlier, this is what could transform Bloom from an equipment manufacturer into a full-fledged infrastructure platform that major capital can invest in. And perhaps this is where the most important advantage of the entire model lies. It is not just about the fuel cell technology itself. Bloom can offer in one package the technology, project preparation, installation, fuel access, service, and financing—and that's just a specific delivery timeline. Competitors might try to copy the chemistry behind the cells themselves. It is much harder to create an entire system where all these elements work together. And major capital is willing to pay for speed for a very simple reason. Time costs money. Every month of delay for a data center means another month of financing costs, depreciation, and the loss of part of the economic life of the computing equipment. That is why a solution that allows a data center to be launched sooner can justify a higher price, even if the energy itself isn't the cheapest over a 10-year horizon. Therefore, the client is not just buying electricity. They are primarily buying the certainty that the data center will be launched on time. If, after all this, you are interested in the possibility of investing in Bloom Energy or other American companies, you can do so through Trade Republic. You can find the link in the description. But before we look only at the growth potential, we must also see what could go wrong. And the report itself does not hide these risks. The first is a very high concentration of customers. In the second quarter of 2026, one customer accounted for about 73%of Bloom Energy's revenue. That is a massive dependency. If the relationship with such a large client changes, the impact on the entire company's results could be very severe. The second risk is competition. Bloom does not operate in a vacuum. It competes with gas engines, turbines, existing power plants, energy storage, and in the future, also with nuclear power. Depending on the region, access to gas, and local regulations, it may turn out that in another location, a competing solution will be faster or more cost-effective. The third risk is less obvious. Bloom is profiting today partly because the traditional power grid is developing too slowly. If regulators and grid operators speed up the connection process for new projects, some of that advantage will significantly diminish. Moreover, the very success of companies like Bloom increases pressure on authorities to solve this problem. And there is a fourth risk remaining, perhaps the most important in the long term. Bloom solves the time problem, but not necessarily the long-term economics of the entire project. If a client installs Bloom cells only to launch a data center a few years early, and reverts to backup power after obtaining a standard grid connection, the real market for this technology may be smaller than the current narrative suggests. This whole story—rising demand, the speed premium, and a deal worth up to 25 billion dollars—is based on one more fundamental assumption. Revenues generated by artificial intelligence must continue to grow. If demand for AI slows down, the biggest operators will start cutting back on data center investments , and then some of the projects currently in the forecasts might vanish before anyone even orders a single device for them. That is why, if we want to check if this energy boom has a solid foundation, one of the first places worth looking at is the results of a company that is currently one of the best barometers for all of artificial intelligence: Nvidia. In this video, I analyze their latest results and check what they truly say about whether the current investment frenzy still has room to grow. That is all for today. Thanks for watching and see you later.