PLEASE read my friend Ben Kritz’s column on this topic last Tuesday. I agree with what he wrote and will approach the topic from a complementary angle. Data centers are not of one type. I do not claim to be an expert, but I did gain some knowledge as an independent director of the Nasdaq-listed SPAC Apex Treasury Corp. which recently signed a merger agreement with TECfusions, valuing the latter at $4 billion. Let me quote from their press release of July 22, 2026:
“TECfusions, a rapidly growing developer and operator of AI-ready data centers and power infrastructure, signs a business combination agreement with Apex Treasury Corp.
“TECfusions is positioned where accelerating AI data center demand meets a scarcity of power-secured capacity. Company ethos centered on the fusion of ‘Technology, Environment, Community’ focuses on delivering AI-ready capacity while advancing environmental redevelopment and local community outcomes. Unique adaptive reuse strategy designed to convert legacy industrial sites into AI-ready, power-enabled data center infrastructure, accelerating deployment in power-constrained markets.”
The reason I quote this is that the presentations relating to this are the main source of my knowledge on artificial intelligence (AI)-focused data centers. Basically, there are three types of data centers, the existing traditional ones, hyperscalers like those of Amazon, Google and Microsoft, which are similar in purpose to the traditional but of massive scale, and the AI-ready ones which are being purpose-built and where much of the expected exponential growth and at least in the US much of the issues and controversy on power, water use and related issues are concentrated over. PLDT is planning to list Vitro REIT, which is good. FYI, Singapore already has two listed data center Reits — Keppel DC Reit and NTT DC Reit. They are traditional ones with hyperscalers, too.
To use an inexact analogy, traditional data centers are like airplanes, hyperscalers like wide-body jets and AI-focused data centers are like rocket ships. I found this article “AI Data Center vs Traditional Data Center — what is the difference?” from RCR Wireless News from March 27, 2025, by Juan Pedro Tomas to be a useful intro. He starts with, “AI data centers are designed to support complex AI workloads, while traditional data centers focus on general computing tasks. But what exactly sets them apart? Let’s take a closer look at the key differences.”
He lists the key distinctions as follows: 1) purpose and workload; 2) hardware and infrastructure; 3) cooling and power consumption; 4) networking and data processing; 5. storage and data management; 6) scalability and flexibility; and 7) cost considerations. Let me leave it to you if you are interested in reading the article. It is available without a paywall. What is key once you understand that there are basically two types of data centers, are the nature and characteristics of both, and their potential and limitations are different. Also key is understanding where data resides has changed over the last 10 years.
Data used to be stored on-prem, or where you were located. The employers I worked for had in effect small data centers in the office or at least at main regional offices. Then came “cloud” providers like Amazon Web Services and others who would manage your data remotely for you. Hence, the “cloud.” Small users stored the data using cloud services and heavy users or those where security was very important used their own on-prem. Eventually, for heavy data users like financial institutions and others as the cloud becomes more reliable, and the volume and complexity of data become more profound, it becomes a hybrid of on-prem and cloud.
The next step was where the data servers of these cloud service providers were located. One used to have the US, China and regional cloud locations for the continents, mostly where there was critical mass of use, security and reliable supporting infrastructure. Singapore played that role for many in Southeast Asia. Eventually, most countries sensibly required at least critical data to be housed in the country. Our Konektadong Pinoy Law requires that as do most countries of decent size. That added impetus to data centers being built in the country to comply.
AI has led to a different type of data center which is a contrast to the traditional and hyperscaler data centers. One will clearly understand that if you read the article I cited. To summarize, the traditional data center is growing, but a mature though critical business. It is really like a REIT for tech. Their dividend yield ranges from 5-7.5 percent, which implies a PE ratio of 15-20x inclusive of the tax benefits, and the nearly full dividend payout ratio. Vitro REIT, which is well run and of good scale (34 megawatts [MWs] of combined power use), seems to fall primarily here.
An AI data center is like a tech company that has REIT and infrastructure-like characteristics. It caters to the very specific and high-performance technical requirements for AI computing. It is very high cost and scale as besides more expensive hardware and specialized equipment, it requires much more power, and unlike the air-cooled traditional data centers, requires liquid cooling given the much higher temperatures, so needs ample water resources for this. More noise, too. Much more complicated than traditional data centers. That is why the successful listed AI data centers have much higher multiples given the higher value add and specialization.
They charge much higher amounts and require much larger scale (they cost tens of billions of USD per location). Some AI data centers will use a few thousand MWs per location (Philippines total peak demand is about 14k MW).
The two biggest listed data center companies in the US which cater to both AI and traditional are Equinix and Digital Realty Trust. Their market caps are around $100 billion and $67 billion, and trade at PE ratios of 60 and dividend yields of around 2 percent.
The M&A and financing deals for pure but reputable AI data center plays are at around that valuation level. PLDT’s Vitro REIT is of scale and welcome. Then take a look at the requirements for AI-focused data centers. Both are valid but different.
The traditional/hyperscaler like Vitro is a REIT with a growing but mature business. The AI centered is like an exponential growth tech company with REIT features and complex with very high cost and scale requirements, initially I expect most to be built in the US and China, where AI development is centered.
The author is an independent director of the state-run Maharlika Investment Corp.
