SRVR ETF: AI-Led Opportunities, But Not A Buy (NYSEARCA:SRVR)

According to an article in Data Center Frontier, the North American data center market is witnessing a shortage of collocation space, driven by the demand for higher AI workloads, which is leading to higher prices. In consequence, my objective with this thesis is to assess the opportunities for the Pacer Data & Infrastructure Real Estate ETF (NYSEARCA:SRVR) which holds some data center REITs as part of its total assets, as I quantify later.

In this case, the price performances of colocation behemoths Equinix (EQIX) and Digital Realty Trust (DLR) as of March this year when the latest stable version of ChatGPT was launched, show that many investors already seem to be betting on AI-led opportunities, but, SRVR seems to have been left behind as depicted in the deep blue chart below.

Therefore, this could be an investment opportunity, but, first, for investors who may not necessarily be tech-savvy, I analyze how Generative AI can translate into higher occupancy rates and revenues.

Generative AI Improving Data Center Usage

One of the best places to start is to study the demand for Nvidia’s (NVDA) GPUs (or graphics processing units) which are the building blocks for ChatGPT’s large language models or LLMs. In its second-quarter financial results of 2024, the company’s CEO, Jensen Huang stressed the upbeat orders for his company’s chips, initially from the large public CSPs or cloud service providers like Microsoft (MSFT) and Amazon (AMZN) which are adding AI capability to their Azure and AWS clouds respectively. He also mentioned orders by social networking providers like Meta Platforms (META) as well as enterprises rushing to deploy their own private AI clouds.

Now, after being procured, these GPUs need to be deployed into servers together with storage systems and this is normally done in a data center environment. Moreover, one of Nvidia’s H100 GPU particular characteristics is that it is more power-hungry than normal CPUs (computing processor units) made by Intel (INTC) while Generative AI also needs a lot of data on which to train its LLMs. These all confirm that there should be a pressing need for both data center power and collocation space.

Pursuing further, it is not feasible to invest in building facilities ground up as this takes time, and, given the current demand for ChatGPT-style applications (as evidenced by upbeat demand for Nvidia’s GPUs), both CSPs and enterprises need to secure whatever spare capacity is available in existing data centers. Thus, as a result of leasing more real estate and providing more power, and at higher rates, the large collocation providers can benefit from windfall gains.

In this respect, as highlighted in red below, a combined 29.25% of the ETF’s top holdings have direct exposure to the data center business, where there are opportunities.

Quantifying the Market Opportunities

Singling out data centers out of the expected total AI-related spending of $5 trillion to 6 trillion, could represent a $300 billion annual business opportunity as per the CEO of DigitalBridge (DBRG), a company that is present across the data center space.

For SRVR, the three holdings I highlighted above managed total sales of $16.5 billion in 2022 as shown below, which amounts to only 5.5% of the $300 billion of opportunities. This means that they can get a piece of the AI pie in 2023 and 2024, and, I have not included Keppel which is located in Singapore.

Examining previous growth, the three U.S. companies have been growing at 6% to 14% yearly during the last five years, according to their income statements. Looking to 2023, things may prove more difficult following a study by the Dell’Oro Group covering the four big data center colocation consumers, namely Amazon, Google (GOOG) (GOOGL), Microsoft, and Meta. According to the study, these may have stopped expanding their footprints in 2023 and are instead optimizing existing leases. This would have meant sluggish growth, but since this study was released in May, it does not include the new prospects from AI.

Now, even if a 10% growth is assumed for 2023, or the same as in 2022, this will not have a significant impact on the ETF’s value since it is composed of only around 24.64% of North American data center REITs (after subtracting Keppel). In this respect, the ETF’s price-to-sales multiple of 3.52x remains below the category average of 6.04x by 42%, but, the opportunities are not sufficient enough to justify an upgrade in valuations.

Longer-Term Opportunities with Fiber and Edge

However, there are longer-term growth opportunities due to the SRVR’s exposure of around 68% to tech infrastructure which is basically about cell towers and fiber, but, no AI-related developments comparable to data centers seem to be occurring at this juncture. The reason could simply be that currently, the focus is mainly on training Generative AI LLMs which requires the parallel processing architectures of Nvidia’s GPUs.

Subsequently, comes the AI inference process whereby the trained model will be utilized by corporations and individuals way more than is being done today, and it is precisely here that other parts of the digital infrastructure, in the form of robust fiber networks, and cell towers become relevant. Depending on the evolution of the usage pattern, like Generative AI being used for robotics applications where low latency is required, there is a possibility that even edge computing facilities normally offered by towerCos like American Tower (AMT) start to see traction.

Therefore, there may be longer-term opportunities, and for those looking for an alternative, there is the Global X Data Center REITs & Digital Infrastructure ETF (VPN) whose U.S. colocation companies (Equinix, Digital Realty and DigitalBridge) account for about 27.5% of its overall weight. I have excluded companies that are not based in North America including GDS Holdings (GDS), a Chinese player as Nvidia cannot export its latest AI chips to that country because of restrictions by the U.S. Department of Commerce.

This ETF has also demonstrated better price performance compared to SRVR both on a three-month and YTD basis and comes at a slightly lower expense ratio of 0.5% than the Pacer ETF as shown below.

However, SRVR has been around longer, comes with a much higher AUM (asset under management), has been increasing dividend payments for the last three years, and offers investors exposure to developed markets only. These are most likely to adopt AI rapidly, according to research by McKinsey.

Rate Dependency and a Hold

Now, it is this exposure to developed markets where central banks have been particularly aggressive in monetary policy in order to address the inflation problem that is one of the main reasons for SRVR’s underperformance relative to VPN which has no such mandate. In this case, higher interest rates have made risk-free alternatives like treasuries issued by the U.S. government as well as CD accounts paying higher interest rates more attractive than dividend income from real estate.

Noteworthily, after some recent economic data covering employment numbers hinted at the possibility of a possible halt in the Federal Reserve’s interest rate hike cycle, SRVR delivered a one-day performance of 1.69%. That was on August 29, and, going forward, it is difficult to predict what the U.S. Central Bank will do as it also has a 2% inflation target, a point which was reiterated by its Chairman at the recent Jackson Hole meeting. Therefore, investor enthusiasm may be temporary as CPI figures get released and imply that rates have to be hiked again.

In conclusion, this thesis has shown that there is likely to be more real estate demand as Nvidia’s GPUs start to be deployed in data centers. There could also be additional business for towerCos and fiberCos in the next phase of AI development, but, in the meanwhile, a lot also depends on the Fed, and, at the current stock price of $28.94, SRVR is a hold.