As Fortune 500 companies hunt for the next frontier of business growth, artificial intelligence (AI) has taken center stage. Its increased adoption has far-reaching implications for the data center industry.
The last 20 years of rising demand for data centers demand stemmed from storage and computing requirements as well as migration from on-premises to cloud infrastructure. New advancements in software applications and IT transformed client needs, prompting significant growth in data center inventory (Figure 1), according to CBRE Research.
How will AI’s rise affect data center development and demand?
There are numerous unknowns. How will AI affect jobs, infrastructure development, energy use and privacy? Can existing and under-construction data centers support AI’s growth? Will hyperscalers seek facility development in edge markets, where lower-cost power supply and cheaper land are available?
What is AI?
ChatGPT, a chatbot that understands and responds to inputs from users, catapulted AI into the mainstream. It became a viral sensation and the fastest app to reach 100 million users. Taking a step back, what is AI?
AI’s machine learning functions are twofold:
- AI training: Building a model from the input of a dataset
- AI inference: Generating predictions, solutions and actionable results from dataset learnings
The functions do not have to simultaneously work at the same location. Each has its own unique storage, power and computing needs.
In its most basic form, AI can help answer a question or draft an email. Future advanced capabilities will be dramatically more sophisticated.
Use cases
AI’s cultural presence is at an all-time high. But, without the same mass awareness, data center operators have been utilizing AI in the following ways: improving energy efficiency by proactively managing power usage effectiveness (PUE), monitoring a facility’s hardware to extend its usable life by proactively detecting and fixing issues, assisting in planning a data center’s physical space and monitoring temperature and humidity constraints.
Use cases for AI apply not only to data center operators but to users. Customers can deploy AI software from data centers for service chatbots, marketing analytics, data visualization, lead generation for business development, streamlined HR hiring and onboarding processes, self-driving cars, and insurance and fraud detection.
What does this mean for data centers?
The two essential elements of AI machine learning have different data center needs. AI training can be done in a relatively siloed environment. High computing power is necessary but does not require proximity to end users or interconnection to other facilities. A data center in a rural area with lower land costs is an example of this type of facility. In contrast, AI inference requires extremely high performance and low latency for end users and applications to interact with the model in real time. An example of this facility is an edge data center in an urban environment.
In a survey by S&P Global, 84.6 percent of respondents stated their organization’s AI/ML infrastructure spending would increase slightly to significantly. CBRE projects increased demand for data center development in Tier 3 markets, such as Des Moines; Charlotte, N.C.; and Columbus, Ohio.
Power supply constraints continue to be a challenge. AI applications consume significant power. In terms of hardware, AI requires high-performance processors, which need more power than traditional data center processors. In addition to more power, modifications for cooling technology will be required to reduce downtime. Liquid cooling is preferred for high-performance chips due to legacy air-cooled chillers’ limitations. Markets that may be adversely impacted by this liquid cooling need include Phoenix and Southern California, due to their water scarcity. Overall, there is incentive to develop AI-specific data centers in markets with ample power supply, lower energy costs and land prices to handle these complex and high-performance workloads.
AI doesn’t only consume power — it also can mitigate power usage. Reducing emissions is an increasing concern for organizations around the world. According to EY, “With pragmatic usage of AI, companies can save up to 40 percent of the power spent on data center cooling.”
Conclusion
As device software and applications evolve, society’s physical infrastructure must too. AI — like IoT, augmented reality, virtual reality, industrial automation and other new applications —increasingly needs data, reliability, lower latency, computing power and proximity to the end user.
How data centers are specifically tasked and allocated for AI is opaque due to the confidentiality of AI development among most companies. However, the IDC projects worldwide revenue for AI at $154 billion in 2023 and surpassing $300 billion by 2026. This represents a 27 percent compound annual growth rate, more than four times the growth rate of overall IT spending over the same timeframe. The United States is projected to be the largest market for AI, at more than 50 percent of total worldwide spending.
The CBRE report, which originally appeared on CBRE’s website, can be found here.