Amidst soaring demands driven by the rapid advancement of AI, tech firms, data center developers, and utility companies are increasingly alarmed about potential electricity shortages in the U.S.
However, a recently released paper indicates that the situation may not be as severe as it seems, provided that data centers and other major electricity consumers slightly reduce their usage.
By restricting their power draw from the grid to 90% of its maximum for a few hours annually, new users could potentially free up 76 gigawatts of capacity across the United States. This figure exceeds the total global consumption of data centers, as reported by Goldman Sachs. To contextualize this number, it accounts for about 10% of the peak electricity demand in the U.S.
If data centers were to implement even greater limitations on their usage, they could release even more capacity.
Such initiatives are not entirely unprecedented.
For many years, utilities have encouraged large electricity consumers, including shopping centers, universities, and factories, to scale back their usage during peak demand periods, particularly on sweltering summer days. These consumers often reduce air conditioning or shut down energy-intensive machines momentarily, which earns them credits on their utility bills.
Data centers have generally remained passive in such practices, prioritizing uninterrupted service and performance for their clients. The study posits that data centers are well-suited to participate in demand-response programs due to their inherent flexibility.
The study highlights several methods by which data centers can reduce their power consumption. One approach is through temporal flexibility, which involves rescheduling computing tasks to times when electricity demand is lower. For instance, tasks like AI model training could be easily postponed to accommodate short-term reductions.
Another approach is spatial flexibility, where companies transfer computational tasks to areas experiencing low demand. Data center operators can also consolidate workloads and power down some servers to reduce consumption.
For mission-critical operations that cannot be delayed or relocated, data center operators might rely on alternative energy sources to compensate for any curtailment. Battery systems are particularly effective since even moderately sized setups can deliver power for several hours almost immediately.
Some companies have already engaged in pilot programs of similar nature.
For example, Google has leveraged its carbon-aware computing platform, initially designed to minimize emissions, to support demand-response mechanisms. Enel X has collaborated with data centers to utilize batteries from uninterruptible power supplies (UPS) to help stabilize the grid. Additionally, PG&E is offering expedited connections to the grid for data centers that agree to take part in demand-response initiatives.
While these adjustments will not entirely remove the need for new energy sources, they could transform a potentially severe situation—where half of the new AI servers lack adequate power—into a more manageable issue.
Compiled by Techarena.au.
Fanpage: TechArena.au
Watch more about AI – Artificial Intelligence
