I will start with the downsides and move on to the upsides regarding a few companies that are indispensable to AI data centers.

Chips require cooling to operate efficiently. We are all aware of this, but recent developments over the past few months have significantly complicated the competitive landscape in this area. If you follow AI infrastructure, you are likely already familiar with what I am about to say. You know that Amazon ($AMZN) and Alphabet ($GOOGL) have developed and begun using their own cooling systems; that is the first point. Another point is that SpaceX ($SPCX) and Planet Labs ($PL) are preparing data center infrastructure in space. Their clients include xAI, Google, and Nvidia ($NVDA); that is the second point. The years 2027 and 2028 are being discussed in this context—I cannot say if they will succeed, but their goal is to leverage the cold temperatures of space, which is a primary objective alongside energy considerations.

Let’s turn to the data center power capacity expected to be established in the US over the next two to three years; I have calculated a general average based on reports from BofA, Goldman Sachs, and VanEck. There is a projected data center capacity of 50–60 GW in 2025, with an expected addition of 12–17 GW in 2026 and another 15–20 GW in 2027. Goldman Sachs projects that, in a scenario of aggressive growth, capacity additions could reach 36 GW, though there are unfortunately concerns regarding energy supply and installation logistics. For 2028, a capacity increase of over 20 GW is anticipated.

Goldman also has a report specifically addressing Data Center CapEx. Total data center CapEx between 2026 and 2031 is projected at approximately $2.147 trillion; if we estimate the cooling segment's share at roughly 15–20%, we arrive at a figure between $322 billion and $429 billion—a total that includes air cooling, liquid cooling, CDUs, and thermal management equipment. That 15–20% represents a significant share, especially given the increasing need for liquid cooling with next-generation chips.

So, who is competing for this 15–20% slice? The list goes on and on: Vertiv ($VRT), nVent ($NVT), Eaton ($ETN), Modine ($MOD), and others. Among them, Vertiv stands out as the leader in liquid cooling; they are also active in the 800V DC space. In its latest earnings report, Vertiv saw its backlog rise from $15 billion to $18 billion. Of course, companies in this sector don't typically boast high margins—so the ratio of total revenue to cash flow might not seem impressive—but remember, this is standard for the industry. Vertiv’s growth is expected to peak within the coming years: 37% in 2026, 30% in 2027, 21% in 2028, and 15% in 2029. These are certainly not bad figures; the company is set to maintain growth above 20% for roughly another three years. While the PEG ratio stands at 0.5x today, it rises to 0.7x for 2027 and 2028—a natural progression as the growth rate slows. In other words, the stock is currently cheaper than it is projected to be in 2027 and 2028.

So, would I buy the stock? If I were to add a new stock to my portfolio, the potential return would have to justify the risk involved. The same applies to Vertiv; I haven't performed my own valuation yet, but to be honest, I might open a small position around the $210 level, depending on the circumstances—though the stock might never reach that price, or I might be focused on something else if it does. I wouldn't be too upset if it doesn't happen; I like the company—it has a solid "moat"—but I’m not absolutely crazy about it.