Joe Zidle: A Real-World Investment in the Age of AI

Bubble speculation is something of a sport in Wall Street circles. After all, to be first in identifying any form of excess can be wildly profitable. Being the next Jeremy Grantham, often acknowledged for predicting the 1999 tech bubble, or Michael Burry, he of the housing bubble prophecy in 2007 that earned him recognition in books and film, has significant appeal. But their accurate predictions are clear exceptions in bubble speculation history. So, when I see more and more reports questioning whether there’s a generative AI bubble, I view them with a skeptical eye.

When analyzing a sector for potential excess in the market, we have to look at its foundation, deeper than recent performance and hype. AI’s foundation is data and the infrastructure required to use it, which is one reason that leads me to believe that this isn’t a bubble. It’s a developing long-term growth opportunity across a potentially transformative technology value chain.

It’s early days of AI’s growth story At Blackstone, we recently assembled a network of chief technology officers to discuss, among other topics, AI’s impact and its potential. A common observation from these CTOs is that many companies are in the research phase with AI and have yet to fully exploit its capabilities for productivity. I believe these discussions leave little doubt that generative AI will bring about monumental transformations, but it will do so in ways that we cannot yet fully understand.

Yes, AI-related stocks have surged significantly this year. But as Deutsche Bank Research Strategist Jim Reid humorously noted in his “Flippant Friday” post, if this were the trajectory of AI stocks with only 50% public awareness, where will these stocks land when the entire population becomes aware of AI?

Where we are currently is the beginning of businesses and the wider economy integrating AI. With as many, if not more, questions than answers currently, what we can do is look to identify the dynamics that figure to be integral to AI’s implementation and growth trajectory.

More data means more storage needed In some respects, generative AI reminds me of the earliest days of e-commerce growth. It was clear that Amazon would fundamentally disrupt the retail industry, and that sent investors hunting for the infrastructure that would enable its success. Not so obvious initially was that warehouses would be integral to Amazon’s rise. To fulfill its promise, Amazon needed last-mile distribution, which required a massive amount of warehouse capacity near population centers. Urban in-fill warehouses were transformed.

Today, we can look at AI and data centers the same way. Generative AI models rely on robust data processing capabilities and extensive storage capacities, which should drive demand for physical data centers that can handle the intensive computational requirements of generative AI applications. Also, data centers enable distributed training and inference, efficiently coordinating operations across multiple machines or multiple other data centers. The demand for data centers has increased dramatically in recent years with storage and computing requirements shifting to the cloud. According to CBRE estimates shown in Figure 1, data center inventory has nearly tripled since 2015.^[1]


Figure 1: Data Center Primary Market Total Inventory (MW)

_{Source: CBRE Research, CBRE Data Center Solutions, H2 2022}


This data center growth really should come as no surprise considering the sheer mass of data that the world now produces each day. Figure 2 shows some mind-blowing examples, like that more than 230 million emails and 16 million texts are sent every minute of every day. Each minute also yields 5.9 million Google searches, 1.7 million pieces of content shared on Facebook, and 347,000 Tweets shared on Twitter.


Figure 2: Amount of Data Generated Each Minute of the Day


_{Source: Domo, data compiled by Goldman Sachs Global Investment Research}

Opportunities in generative AI’s physical presence Now add AI to the mix. It’s natural to think of AI as an esoteric, abstract concept; a form of software that lives in the ether. In reality, real world data centers anchor AI, just as they do data storage and cloud-based applications that run on physical servers. The physical needs of AI will diverge in certain ways from existing applications and could lead to the development of data centers tailored for the new technology. Meta, for example, recently shared details on its own plans to redesign data centers projects to support its AI development.