CyrusOne Intelliscale™ Design and AI Data Center FAQ

Today's Artificial Intelligence (AI) data centers require careful design considerations to ensure optimal performance, scalability, energy efficiency, and reliability. CyrusOne's Intelliscale™ Design is engineered to meet the growing demands of AI.

As an innovative and new offering based on the changing needs of the market, people have many questions about AI Data Centers and CyrusOne’s Intelliscale™ Design.

This FAQ is designed to help answer some of the most common questions.

General Questions:

Q1) What is an AI data center?

A1) An AI data center is a facility that is purpose built to support AI’s large volume of data and heavy compute requirements, coupled with its unprecedented energy demand and need for cooling. To meet the computational and deep learning needs of AI computing, an AI data center houses servers that are equipped with high-performance hardware, including GPUs (Graphics Processing Units) and TPUs (Tensor Processing Units), which are optimized for parallel processing and matrix operations essential for AI model training and inference. In addition, they provide optical networking and more efficient storage to support the scale that AI models require. An AI data center is also more compact taking up 25 percent less space and delivering more watts per square foot than traditional data centers.

Q2) What are the primary differences between a traditional data center and a data center built for AI?

A2) The primary differences are in the amount and type of compute power delivered in a more confined space. This requires more power, cooling, and infrastructure than traditional data centers, and typically a larger capital investment.

Q3) Can current data centers be retrofitted to support AI applications?

A3) Yes, retrofitting a traditional data center is possible especially given the CyrusOne Intelliscale™ design incorporates CyrusOne’s standard kit.

Q4) Do AI Data Centers need to be built in availability zones (AZ)?  (Answer what is AZ?)

A4) No, one of the long-term advantages of AI data centers is that they do not have to be built in the availability zone. Availability zones are isolated data centers in a region or area in which public cloud services originate and operate. This ensures that customers have stable connections and increased speed, plus powerful computing power close to them. An availability zone consists of multiple data centers, all of which are equipped with independent power, cooling and networking infrastructure and connect to each other over redundant, high-speed, low latency, network links.

Q5) With the growth of AI, can data center capacity keep up with the demand?

A5) Tirias Research forecasts that given current growth generative AI data center server infrastructure plus operating costs will exceed $76 billion by 2028. A similar report issued in late 2022, prior to the explosion of generative AI estimated the entire market would be at $48 billion by 2030. Data center construction can keep up with the demand if the power, water and other infrastructure needs can keep up and be expanded.  One of the advantages of AI data center not having to be in the availability zone (AZ) is that construction can be in more rural areas near power supplies and other resources rather than in urban areas.

Q6) What challenges does AI bring to data center providers? (Compute power/storage)

A6) Given the computing power and energy required by AI, the biggest challenge is developing near enough to large power supply and then designing and delivering adequate cooling. Because an AI data center does not have to be in an availability zone, finding locations in more rural areas near power supplies will help overcome that challenge. CyrusOne has also been a leader in designing environmentally conscious energy and cooling systems and continues to innovate in ways that will allow for the efficient cooling of these facilities while becoming more and more environmentally friendly. Similarly, water supply and power generation are typically near each other so adequate water should not be a long-term c­­hallenge. ­

Q7) What does “ultra-high density” mean?

A7) A common definition of data center density refers to the amount of power delivered per square foot. As technology improves the amount of the amount of power delivered per square foot has continued to grow. What was high density five years ago, is now considered low or medium density. In fact, in 2010 data center density was reported to have “hit a wall” at 100 to 200 watts per square foot. Today’s traditional data centers provide between 150 and 300 watts per square foot.   A high-density data center like CyrusOne’s Intelliscale™ Design delivers more than 2,000 watts per square foot - an astounding 600-to-1,200 percent more watts per square foot than typical data center power density.

Q8) How can facilities that use exponentially more power and resources continue to be environmentally conscious?

A8) The data center industry as a whole and CyrusOne in particular, have been on the forefront of environmental, sustainability and governance (ESG) efforts. CyrusOne’s innovations have accelerated the reaching of their ESG goal of being carbon free by 2030. Those energy efficiencies will become even more effective as technology, experience and ingenuity come together to meet future needs. Similarly, CyrusOne has deployed a unique cooling system that greatly reduces and, in some cases, eliminates water use as they strive for a net water positive portfolio.   The ultra-high density of an AI data center also means that the footprint of the facility will be at least 25 percent less than traditional data centers using fewer materials, reducing demand for labor, and being a good neighbor in the communities where they are built.