DataBank is a leading provider of enterprise-class data center, cloud and interconnection services.
When you look back at seismic events that have happened in your career, how does this AI revolution compare to previous events like the dot-com boom?
As Mark Twain famously said, history doesn't repeat itself, but it rhymes. My career started in the mid-90s at the advent of the commercial Internet when the Internet did not exist as a business function, and look how far we have come today. The current phase brought by ChatGPT and generative AI feels reminiscent of the early days of the Internet. In the mid-90s, dial-up was a big thing, and Netscape's browser was an aha moment, signaling the Internet's transformative potential. ChatGPT is akin to that moment.
In the 90s, the focus was on gaining access to the Internet, leading to a massive demand for telecommunications and service providers. Today, generative AI's key inputs are GPUs, making companies like NVIDIA extremely valuable, and the need for data centers to house these GPUs. As a data center operator, it feels similar to the 90s' ISP boom. Data centers are now the enabling technology for adopting generative AI.
How do you balance meeting the growing demand for data centers while maintaining reliability and not compromising on quality?
The data center business is fundamentally a real estate business. We procure land, bring power to it, and construct facilities to industry standards. Our product is reliable and predictable, which is what our customers want as they push the forefront of new technologies. Since 2016, we have expanded from six data centers in three markets to much larger facilities. For example, our DFW3 facility in Dallas, initially considered large at 10 megawatts, is now dwarfed by deals as large as 320 megawatts, like Microsoft's recent lease in Atlanta.
The challenge now is managing the massive demand, which has led to bottlenecks, particularly in power constraints. For instance, in Ashburn, Dominion Power had to pause new projects due to transmission capacity issues. Similar bottlenecks have occurred in other markets like Santa Clara and Phoenix. Supply chain issues also pose challenges, with longer lead times for essential equipment. We have had to plan far ahead, ordering significant capacity for projects slated for completion in 2026, to ensure we meet future demand.
Are you seeing a shift towards building data centers in colder climates due to temperature and water constraints in places like Texas?
There is a common misconception about the need for water in modern data centers. About 10-15 years ago, evaporative cooling was common, but now we have moved to closed-loop systems that use minimal water. Water scarcity is not a significant issue for us. As for heat, modern data center designs and construction standards have improved efficiency significantly. We use a metric called PUE (Percentage Utilization Effectiveness) to measure efficiency, and with today's technology, we can achieve low PUEs even in warmer climates like Texas.
While colder climates like the Nordics or Quebec, with abundant hydropower, can be appealing, most data centers are built near population centers with good fiber infrastructure. Texas, despite its heat, remains attractive due to its population size and energy resources. Efficient cooling systems and strategic insulation allow us to operate effectively in these regions. So, while there are niche examples of colder climates being advantageous, the majority of our data centers remain in key hubs like Phoenix, Dallas, and Atlanta, where demand and infrastructure are robust.
What keeps you up at night in the context of DataBank?
We have had a very successful run over the last seven years, with great business, customers, and relationships. Reflecting on why the internet bubble resulted in a crash, it was due to an oversupply that could not be absorbed quickly in 2000-2001. Eventually, all that fiber got absorbed, but it took longer than expected. One of my concerns is whether customers will be able to monetize their investments in AI to build real business value, or if there will be a pause similar to the internet era where they need time to figure out how to use the technology effectively.
The good news is that our business is massively diversified, with over 2500 high-credit-quality customers. We feel very comfortable with our position. Unlike the unbridled investment in capacity during the 2000 era, investors today are more rational, demanding proof of demand before investing. My concerns revolve around potential oversupply and the adoption of technology that brings business value, but I believe these will solve themselves over the long term. It's more of a short-term or intermediate-term dynamic.
With likely trillions of dollars being poured into achieving artificial general intelligence (AGI) over the coming decades, what do you see in your crystal ball for the future of the industry?
We closely follow the trends you mentioned, informed by our relationships with a diverse range of customers. This gives us insight into how the business community is investing in technology. Ultimately, we look at our business, investor base, and reasonable investment levels considering inherent business risks. Our focus is on funding these investments, delivering great products, and ensuring we can monetize them effectively.
While there are many concerns, from climate change to geopolitical tensions, we can't worry about everything. We have a solid strategy and feel confident about our growth targets, which align well with macro demand trends. Regarding the broader AI landscape, we are still in the early stages. As highlighted in "AI Superpowers" by Kai-Fu Lee, we are just beginning to see general-purpose AI applications in fields like drug discovery. The next few years will be critical as more industries figure out how to integrate this sophisticated technology into their operations, driving data center demand.
If a startup wanted to pivot towards data centers, what low-hanging fruit could they target for VC investment?
In the data center space, the physical construction of facilities is not a startup-type business; it is an established sector with many players. However, there are opportunities in data center technologies especially in enhancing efficiencies. Innovations that can drive down PUE (Power Usage Effectiveness) would be valuable. For example, building data centers at a price point that customers want while achieving a lower PUE would be highly beneficial.Other promising areas include ancillary services, monitoring services, and efficiency improvements. These are more suited for startups and offer potential for significant impact and VC interest.
How has the synergy between government, industry, and academia in the US benefited Data Bank?
In the U.S., the political landscape is highly polarized, which presents challenges. However, there are two areas where bipartisan collaboration could be beneficial. First, transitioning the electrical grid to be renewable and capable of delivering more capacity is crucial. A government plan to accelerate projects in transmission, generation, and distribution would support the development of data centers. Second, regulation is lagging behind technological advancements. Current legislation on content moderation and ownership is outdated. With new technologies posing privacy and security concerns, it is essential for the government to create a level playing field with clear rules.
