99% of Wall Street Doesn't Understand This I'M GONNA EXPLAIN

I'm gonna explain the finance of AI starting with the layer that's often the most overlooked and yet the most critical. And we're going to do it in less than three minutes. Here we go. Nvidia CEO Jensen Wang called the AI ecosystem a five-layer cake. Layers five and four are the ones that are the sexiest and most visible. With OpenAI and Claude arguably the two poster children. But in order to build the models, they need layers one to three, the energy, the chips and the infrastructure. This is what Jensen calls the AI factories. Now it's worth noting that each of the big hyperscalers, defined as a company that operates computing infrastructure at a massive scale as its core business, have different parts of the five layers. With Google the only hyperscaler with all five layers in house, at least for now. Regardless, in order to run the models you need compute. The compute is what we're producing in these AI factories. Now the way to think about how these work is it's basically like an assembly line. You have some kind of raw material that comes in, fuel. It's then converted into energy, which then feeds into the grid, which then goes into a data center, which by itself is just a powered shell. The shell is useless without the chips and then out of the data center comes compute. But the part of the system that is the limiting factor is actually the energy. So where does energy for these data centers come from? Well, historically many pulled from the grid. And while we call it the grid, it's not just one. In the US, we actually have three. The Western interconnection, the Eastern interconnection and Texas. ERCOT has its own. Regardless, with 66 different balancing authorities trying to keep the lights on, operating under different rights in different states, some regulated, others not, it's a tangled interconnected system. But the grid's most important job is matching supply and demand in real time. And the best way to visualize that is through the generation stack. Take every power plant in a region and stack them up from cheapest to run to most expensive. At the bottom we have nuclear and you put coal in there because you can't really modulate it. You have solar and wind which are renewable, free to run once built. The majority of the demand however will likely be satisfied with natural gas fired power plants where the cost to run is a function of the price of gas. Now that's dependent on geography. The closer you are to a pipeline, the cheaper you are. The closer you are to a pipeline, the cheaper you are. Which might only run a few hundred hours a year. Grid operators stack supply against total demand and the price for electricity must be high enough for the least efficient and most expensive plant to run. The less efficient plants stay offline and the cost of the marginal plant is what sets the price for everyone. But here's the thing. In residential areas, demand isn't static. It changes hour by hour where you typically see peak demand from 2:00 to 9:00 PM every day. Data center loads are actually static. So they basically run all the time, 24/7. So when a data center draws on the grid, it's not only pushing prices up. It's pushing into a grid that is already maxed out. Last week the DOE declared New York City to be in a grid emergency as a heat wave pushed demand to record highs. This explains why the new laws in many states, as if you want to build a data center, you must bring your own electrons. Follow for Part 2. We will get into some insane tax benefits AI data centers are getting and so much more.