Don’t be Afraid of AI

AI has arrived.  I’ve heard rumours that it’s already learned the whole of internet, twice.   Eventually, we won’t talk about AI as a separate entity but it will become integrated with everything we touch and use.  A study by Amazon Web Services (AWS) researchers suggests 57% of content published online is already AI-generated or translated using an AI algorithm (via Forbes).  Ultimately, it seems likely that we may need AI to fight AI, with banking checks powered by AI to check against AI fraud, as an obvious example.  

There is much consumer fear around the introduction of machine learning.  Yes, it is undoubtedly an exciting development and offers huge potential in areas such as medical advancement, however, the practicalities of integration, for some, is truly terrifying.  Partly this fear is driven by the unknown.

One thing we do know is that AI requires significantly more computing power and therefore energy.  For example, it has been estimated that a voice AI internet search takes up ten times more computing power than a normal query and this has knock on effects in terms of energy use and data.

To quantify this further, the International Energy Agency has stated that data centres currently use about 1% of the global electricity demand.  However, McKinsey estimated Why invest in the data center economy | McKinsey that by 2030, data centres’ power consumption will almost double and is expected to reach 35 gigawatts of power consumption annually, up from 17 gigawatts in 2022. 

However, we must not fear AI.  It’s here to stay.  Instead we need to adapt and learn to live with it, and quickly.  This means that everyone must play their part from the legislators introducing new laws about issues such as power consumption and even copyright, to those supplying the infrastructure within our burgeoning data centre industry.

We all need to be flexible enough to cope with whatever the AI future will throw at us.  However, a little like climate change, we must think of the future but we also need to protect ourselves in the here and now, because change is already happening and is putting pressure on Grid stability as well as infrastructure design changes to accommodate AI. Protecting Critical Power Infrastructure with a UPS is more vital than ever.

From a data centre perspective, protecting power is critical and currently, 99% of the power required for data centres is unrelated to AI’s power-hungry model, for now.  In five years’ time, data centres will need to scale their UPS solutions and manage their energy storage far better than they do today.  So how do we deal with the present day and keep an eye on AI’s rapidly evolving future?

Flexibility is Key

Data centres are already gearing up to match the increased demand for data.  However, no-one can quantify the anticipated size of that demand.  I believe underutilized data centres do offer an opportunity to absorb some AI.   However, data centres need to embrace this.    Currently, everyone is still learning, and AI is being kept in separate data halls but this will change as it becomes more integrated into everything we do.

We don’t know what a data centre will look like in 30 years and so flexibility is key.  Facilities need to look for solutions which will allow them to scale as needs change but also minimise the cost of investment while the future size of the load is unclear.

Active Management

It has been estimated that AI learning requires three times the electricity needed by normal servers and this is set to have a greater impact in the future.  Facilities will need to be better at managing energy consumption to become more efficient, so our electrical infrastructure can cope with the dramatically increased energy demands of AI.   

Currently, in Ireland for example, data centres are now responsible for a colossal 18% of all electricity use (Central Statistics Office, May 23).  In Dublin, new datacentre builds have been banned so they can ensure there is still sufficient power for homes.  We have no choice but to manage energy better and look at alternatives.

Simple measures that can be implemented include locating UPS batteries away from any heat source in their own dedicated environment, resulting in a reduction in cooling requirements.  Or, consideration given to using alternative battery technology which require less space and can run optimally at much higher temperatures.  

Right sizing the UPS is also important.  For example: you may have multiple UPS supporting a much smaller load than intended at design stage, so switching off UPS modules that are not required will have a positive effect on efficiency use.  There is a nervousness around switching off systems. However, as long as the required resilience level is maintained the positives outweigh the negatives.

To enhance efficiency, some datacentres with a resilience level of N+N and above manage their UPS systems by utilising both economy (eco) mode and true online double conversion (inverter) mode. It may be the case that the A feed (N) will be fed via the inverter and the B feed will run in eco mode. The latest technology offers 97.1% efficiency in normal operation, switching to eco will increase to 99% allowing the data centre to save circa 2% in losses.  Over time, this adds up and for anyone running legacy UPS systems with efficiencies as low as 85%, the savings will be much greater.  Sometimes this can be perceived as a risk as eco mode is effectively raw mains, but if the mains goes out of tolerance the transfer back to inverter is instantaneous and seamless.  This is a good option for datacentres with a minimum resilience level of N+N. 

As we can see, there are numerous ways to make an immediate impact on reducing power usage.  Getting the right advice and learning how to harvest the information from the UPS means that informed decisions can be made.  The overall result is savings on energy and operating costs. 

Harnessing and storing Energy

In the future we believe UPS will offer a far greater opportunity to help facilities manage energy better.  With its associated batteries, a UPS can potentially be used as a micro-grid or energy hub to offset energy use at times of peak demand.   It could also be used as an interface to accept alternative sources of renewable energy which could contribute back to the grid.  Energy companies realise how much power data centres use and so are likely to introduce incentives for facilities who can contribute to supporting the grid. 

Renewable energy will be useful to re-charge battery banks and this energy used to take the peak off the grid energy demand.  We have worked on numerous projects which allow UPS to take advantage of such peak shaving functionality.  There is also an opportunity where the UPS could enable energy to be sent back to the grid simultaneously as accepting energy from renewables while supporting the load.  The grid, batteries and renewables will need to work together to satisfy and offset future demand.

AI offers the potential for great technological advancement.  AI may be good for business and create jobs in the data centre market.  However, we need to collaborate now to solve the immediate challenges it brings in relation to energy use and management.  One day we can ask the AI to solve these issues, but we are not there yet.  In the meantime, don’t be afraid of AI.  Protect your critical power now and be ready for the future with a UPS system with a long design life, which is flexible, scalable and adaptable to integrate with different energy management protocols and future ready to accept alternative energy sources.

For further information about Centiel’s flexible, future-ready UPS products, please visit: www.centiel.com