This article is supported by Legrand.
In a global survey by Google, more than half of the people expect A.I to benefit their daily lives, including jobs, security, and well-being. The success of A.I in meeting these expectations depends on the availability of large amounts of data for training. Data centers provide scalable and secure storage for A.I, using platforms such as High-Performance Computing (HPC), which is capable of handling massive A.I workloads. Large language models like ChatGPT exemplify this capability, processing and analyzing vast amounts of data in real-time to provide contextual results, no matter how expansive the questions the users ask.
The need for big sets of data required by A.I has unavoidably pushed data center players to expand their efforts to provide better services, with significant implications. One of the main concerns is the rise of power consumption, with countries such as the United States predicted to reach 35 GW by 2030, which brings with it another set of environmental problems.
Data Centers are already contributing 2% of total carbon emissions as of 2022 (Greenex DC), meanwhile, increased computing work as a result of A.I integration means that even more emissions are generated.
The earlier mentioned High-Performance Computing (HPC) operates at power densities ranging from 20 kilowatts (kW) to over 60 kW, contrasting sharply with common low-power servers used in things such as web hosting, which only consumes about 100 watts of power. This emphasis on larger computing power will ultimately raise the earth’s temperature, creating extreme fluctuations that overheat data centers.
With the threat of climate change looming over, data center players are pushed to take measures to prevent it from getting worse. Apple and Google aim to achieve carbon neutral by 2030 by reducing and offsetting their carbon emissions. On the other hand, Microsoft has already moved on from carbon neutral to their new goal of being carbon negative, pledging to remove all carbon the company has created since 1975.
One bold solution to the energy and emission problem is by Ian Bitterlin, Britain’s foremost data center expert, in which the world’s internet usage will have to be rationed. However, even he agrees that limiting a large portion of the human population in internet usage would be “political suicide”. A more realistic way of preventing this endless cycle between power consumption and increased emissions is for data centers to be able to manage the temperatures generated, which brings us to cooling solutions.
As the future arrives, simply investing more in conventional air-cooling systems would only increase the risk of damage. In relation, better data center monitoring is critical, preventing overcooling and/or overheating which also have negative impacts on climate change. Inefficient cooling can also reduce equipment life, poor computing performance, and put a larger strain on cooling systems.
Figure 1. ColdLogik RDHx (left), Minkels Nexpand Rack (right)
Source: Legrand (n.d.)
Solutions that can tackle this issue from a more complete perspective, from beginning to end, are crucial, instead of trying to complete a checklist of equipment from customers. The journey of becoming more sustainable starts with the process of building data centers.
Legrand, for example has specific solutions to accomodate the rise of A.I in the data centers.
There are four parts that we are going to discuss in this article:
1) Rack systems
2) Retrofit cooling systems
3) Power distribution
4) Cabling
5) Physical security
The first part we are going to talk about is Minkels Nexpand Rack (Fig. 1), which is able to hold IT equipment up to 1020 kg. Now, this is important because of the physical infrastructure demands of A.I, often needing 200 to 300 kg of weight on a rack. It is highly efficient in terms of timing and costs of installation, while also being reusable for the next project.
This brings us to the next step, which is an efficient cooling method best represented by ColdLogik Rear Door Coolers (Fig. 1). Rising data center power density calls for unique cooling strategies such as liquid cooling, especially when power densification levels reach above 30 kW per rack, causing hotspots in the process.
One reason why liquid cooling is being considered more effective than traditional air-cooling systems in managing A.I workloads is because liquids are thousands of times more efficient per unit volume than air at removing heat. However, air still acts as an important medium for transporting heat outside the hardware. Legrand’s Rear Door Cooler implements the methods of air assisted liquid-cooling to control the room temperature at its most efficient point.
Besides cooling, power is another important aspect; the balance between these two factors is something that needs to be achieved by any data center wanting to be sustainable while also fulfilling the satisfaction of their customers.
Monitoring also plays an important part here, in order to determine when data centers switch between power sources without shutting down power, complying to users’ and regulators’ high standards regarding downtime. The Starline Track Busway (Fig. 2) is providing the scalable part for source of power. It also has the added benefit of faster installation thus lowering labor costs.
Going back to the aforementioned ‘complete perspective’, cabling might not seem obvious in its part of making the data center more efficient and sustainable. In many data centers, cables are usually not removed or changed, for fear of unplugging the wrong cable.
Over the years, the cables can start to pile up and take up space for potential racks. Cablofil Wire Mesh Cable Tray (Fig. 2) offers easy-to-installed, efficient cable management, with improved airflow. The attention to detail goes even further, sharp edges are eliminated through specific welding methods, which makes it easier for installers to change or add new cables.
Figure 2. Starline Track Busway (left), Cablofil Wire Mesh Cable Tray (right)
Source: Legrand (n.d.)
Lastly, the focus on providing physical security can not only make the data center environment more inviting for newcomers, but it also needs to be implemented to protect against new threats.
What was elaborated before in terms of massive amounts of data required for A.I to train on, also means new forms of automated attacks. Methods such as more advanced authentication and encryption are needed to prevent any unauthorized access. Nexpand’s Smart Lock and other Legrand cabinet locking systems are example of solutions, thar are equipped with the latest security protocols.
The final aspect besides the various methods mentioned above in regards to data centers playing a part in reducing its environmental impact, is convincing data center players to invest more in adopting these methods.
Highlighting the different advantages regarding costs and carbon footprint can create a better case for this adoption. This encompasses all data center player’s goals in keeping up with increasing standards of the users while also meeting the corporate sustainability standard.
To close; a net-zero world is still a long journey. Our best-case scenario still lies in these solutions to halt climate change. Investing in energy-efficient hardware will bring forth a greater profit in the long run, creating a better future for the stakeholders in the data center industry.
References:
https://publicpolicy.google/article/global-study-shows-optimism-about-ais-potential/
https://greenexdc.com/why-data-centers-are-affecting-a-carbon-footprint/
https://www.assetspire.co.uk/blog/post/what-is-the-environmental-impact-of-data-centres/
https://www.un.org/en/climatechange/net-zero-coalition
https://dgtlinfra.com/data-centers-artificial-intelligence-ai/
