The AI Data Center Buildout Is Accelerating, and Nuclear Could Be the Most Underowned Piece of the Puzzle

Data center buildout is continuing as firms seek additional AI compute, including some taking on debt. Timing a peak or potential AI bubble is becoming harder, and semiconductor investors are advised to exercise caution, especially those without significant gains yet. The outlook is supported by Google I/O themes such as Omni, Genie, and Generative UI, which could change Search and increase compute needs. Token consumption is expected to follow a parabolic curve as inference demand grows. AI features, platforms, and agentic applications may be transformative and token-heavy. Shifting users from blue-link pages to interactive mini-app widgets and background agents requires substantially more compute, even when generated outputs are not ultimately used.

"The data center buildout doesn't seem like it's about to slow down. As more firms look to step up and supply AI compute, some of which (unlike the cash-rich hyperscalers) are risking their shirts by going into debt, it feels like timing the peak in the AI boom or the beginning of an AI bubble (if there really is one) will get harder to do. In any case, I think the semiconductor names warrant extra caution on the part of investors, especially for those who arrived a bit later, without the significant gains to show quite yet."

"Indeed, buying into parabolic movers might guarantee nothing more than volatility. In any case, the foot remains on the AI gas, and as token consumption moves along a parabolic curve as well, the bull case suggests that the latest upswing in semiconductor stocks isn't a sign of a bubbly move that requires a vicious burst."

"When you consider the Alphabet ( NASDAQ:GOOG | GOOG Price Prediction) Google I/O presentation and all of the mind-blowing applications that are just around the corner from hitting prime time, from Google Omni to Genie and Generative UI, which could forever change the business of Search, it's hard not to think that compute demands are just going to keep moving along that exponential curve, rather than a linear one, as inference demand soars."

"If Google is going to move its userbase from serving up a page of blue links to generating widgets, so-called "mini-apps," interactive UI with AI on the spot, whilst running AI agents in the background, there's going to have to be a lot more compute. And the jarring thing is that much of the stuff that's generated might not even be used by whoever entered the prompt."

Read at 24/7 Wall St.