Home Tech The data center is built like this: Use AI to create the...

The data center is built like this: Use AI to create the “engine room” of the Internet


Source丨Forbes Author 丨 Martin Giles Compiler丨Technology Walker Equinix may not be well-known, but their services support the trillion-dollar global network business system. This company, headquartered in Redwood City, California, has a market value of US$72 billion, with total revenue of US$6 billion last year, and operates a comprehensive infrastructure network of more than 220 data centers in 26 countries or regions. Equinix customers use a large number of servers (with a total area of ​​more than 2.4 million square meters) provided by these data centers to store and process massive amounts of data necessary for online business. In addition, customers can also use the nearly 400,000 optical fiber links managed by Equinix to transmit data between their own servers and servers of other enterprises, and enjoy smooth and convenient interconnection support. Milind Wagle became the CIO of Equinix in September 2016. In his opinion, “The company has become the engine room of the digital economy. After all, the Internet must have a physical foundation to form and maintain operation, and this foundation is the data center.” Picture: Milind Wagle, CIO of Equinix The infrastructure operated by Equinix, as well as the service resources provided by competitors such as Digital Realty Trust and CyrusOne, will not directly appear in the eyes of customers. But the failure of these facilities is likely to cause serious damage: earlier this month, another network backbone company, Fastlyl, suffered an accident that affected multiple websites involving customers such as Amazon, The New York Times and Forbes. Although the incident itself has received high levels of attention, there are still many serious problems that people cannot feel at all, and technology executives such as Wagle have been looking for ways to enhance network resilience. “Foresight” in the data center Like other competitors, Equinix also needs to measure when the server in the data center may reach the upper limit of usage, when it needs to purchase land, and plan for the construction of new data centers and how much electricity the new facilities may require. Wagle explained, “Accurately predicting the power consumption and space occupancy of a data center is a very important task.” According to Equinix data, a data center that can accommodate thousands of servers often requires an investment of hundreds of millions of dollars. The cost depends on the specific location, local electricity price, selected equipment and hardware configuration. Data center companies also need to seriously consider designing long-term power purchase contracts. This is because each data center consumes several megawatts of power, and any calculation errors may bring huge costs. An Equinix spokesperson also shared an example of a Silicon Valley data center contract that contains a monthly fine of $223,000, emphasizing that the center’s expected power consumption should not be less than the committed use of 3 megawatts. For a long time, Equinix has been reviewing historical trends through the planning process, manual spreadsheets and dedicated planners, and combining information from customer channels to make clear estimates of future needs. But with the development of the business, this work has become more and more complicated. To this end, the Wagle team began to seek new applications of AI technology, hoping to develop algorithms to assist Equinix planners to improve their work efficiency and enhance their forecasting level. But the early results gave everyone a blow. Wagle recalled, “At first, everyone was not very interested, and most employees were skeptical about it, and our model was not 100% accurate.” But during the exploration, he also realized the great potential of AI technology and advocated for the team. Allow more time. He firmly believes that algorithms and a large amount of data from customer channels and historical usage patterns are the key to the company’s upgrade of the planning process. Wagle’s IT team found that an important reason why its AI model did not perform well in predicting demand in the early stage was that some of the input information itself was not accurate enough. To this end, he began to persuade colleagues to participate together, helping him to broadly improve the quality of data throughout the enterprise. Many observers believe that this incident is also an important node for Wagle to win the trust and support of opponents. Brian Lillie, who worked with Web domain registrar Verisign and Wagle, said, “His will is very firm. As long as he sees things right, he can always burst out strong willpower and enthusiasm for work, and must pull opponents out. Come to my side.” It was based on his understanding that Lillie decided to invite Wagle to join Equinix and take up the position of CIO. “AI Exploration” in the Data Center Wagle also said that Equinix has now fully launched a new idea of ​​using AI to promote facility planning. “This brings us a very valuable modeling tool, we can predict how long the data center capacity will be completely exhausted, in order to make better supply chain decisions.” These models also help Equinix more Targeted optimization of its own power consumption, as far as possible to improve the environmental sustainability of the business. His team has also been trying to introduce AI technology in other fields. In terms of sales, they built new models to identify and attract new customers. This set of models absorbs about 100 different input indicators, including the target company’s financial status and expected growth trajectory. They also use AI technology to promote workflow automation, including working with internal finance teams to develop algorithms, reducing the amount of manual approval work that needs to be processed every year to about one-third. Picture: An Equinix data center in Amsterdam. The establishment of new Equinix data centers everywhere also reflects the tremendous benefits that Wagle’s efforts have brought to the enterprise itself-since this year, Equinix’s stock price has risen by 17%, which greatly exceeds the overall increase of the Nasdaq index. The company said that 2021 has become one of the most active years of business expansion in its history, and it is currently building 36 infrastructure projects in 19 countries or regions. Their plan also includes the construction of more “super large-scale” giant data centers, sufficient to support the data and processing needs of Internet giants such as Amazon and Google. On June 14 this year, Equinix announced that it has reached a US$3.9 billion deal with Singapore’s sovereign wealth fund GIC for the continued expansion of facility construction. Equinix is ​​also continuing to expand its fiber optic cable network to maintain efficient data flow between corporate customers. Timothy Horan, an analyst at Oppenheimer, an investment bank, believes that “many customers choose Equinix because they can more easily communicate with other companies here.” Horan compares Equinix to the central station of the Internet. Wagle and his team plan to further promote the full application of AI technology to ensure that these key interconnection systems and the data centers that support the network system always maintain stable and reliable operation