Data centers are the cornerstone of the activities of tech giants, allowing access to many online services. Beyond the economic issue, they raise real ecological and security concerns. Microsoft, Meta and Google have decided to experiment with the use of AI to optimize and ensure the proper functioning of data centers exposed to more recurring failures.
A response to security issues for technicians
Contrary to popular belief, working in a data center is not without risk for the technicians who operate it. In the event of failures, the latter must intervene on electrical equipment which is continuously powered. These incidents, if not anticipated, can quickly turn into a disaster. The sector is still marked by the fire which devastated the OVH data center in Strasbourg in 2021, fortunately causing no injuries.
Cryptocurrency miners resell their graphics cards
In 2015, five employees at an Apple data center in the United States were far less fortunate. They were injured and taken to hospital following a chlorine gas leak. This chemical is used in particular to sterilize the water present in the cooling systems of servers. In order to prevent these accidents, the tech giants are designing artificial intelligence that anticipates failures.
Microsoft has developed an AI system that analyzes data provided by sensors and immediately alerts data center teams to ” prevent or mitigate the impact of security incidents “, explained a spokesperson for the company to TechCrunch. The aim is to minimize human intervention in risky situations.
Such situations can occur when a data center is exposed to a heat wave or even a flood. It is for this reason that Meta is working on an AI capable of simulating the operation of data centers in this type of extreme environmental conditions. The company tries to reduce dangerous operations for its employees.
AI facing the environmental consequences of data centers
The security aspect is not the only concern of companies. According to a survey conducted by the IT Uptime Institute in 2020, a third of data center managers were exposed to major outages during the year. These malfunctions are expensive and increasingly common. A sixth of companies said these outages cost them $1 million. In 2019, only a tenth of them encountered problems of this magnitude. In 2020, during the pandemic, Microsoft experienced several outages of its Azure cloud service as usage increased.
AI could prevent these situations to allow companies to save money. Beyond breakdowns, they can detect excess energy consumption in data centers and optimize their electricity needs. This is what Google has been doing since 2018 with its AI, DeepMind, which would have allowed it to reduce the average electricity consumption of servers by 30%.
These AIs also represent an asset in the face of environmental issues. Data centers account for 1% of global electricity consumption and contributed 0.3% of greenhouse gas emissions in 2020 according to a report by the Environmental Investigation Agency (EIA). Server cooling systems are at the center of ecological issues. Microsoft, for example, uses an AI that reduces energy waste related to cooling by using telemetry data to detect hot spots.
Liquid cooling systems are an alternative to the problem posed by the evacuation of the hot air generated by the servers. Yet a data center consumes up to 19 million liters of water per day to keep it running according to the EIA. Microsoft plans to build between 50 and 100 data centers per year, increasing the ecological impact of these.
The Redmond company could use its AI Planetary Computer, which has the function of responding to ecological challenges, for example by estimating the water consumption of its data centers. If the tech giants are no longer able to cool their data centers, global warming could lead to an increase in the number of outages.