We’re no stranger to supply chain issues in the post-Covid world. But one particular bottleneck that analysts are looking at could have an impact on our rapid advancement in AI infrastructure over the next few quarters.
Experts are warning that we’re at the brink of a critical shortage in the supply of transformers for electrical power grids.
Citing unprecedented strain, an analyst at Rystad Energy has estimated we’ll be seeing this problem until Q4 of 2026: U.S. agencies in renewable energy discovery have also been sounding the alarm bells.
“The United States is currently experiencing unprecedented imbalance between supply and demand for transformers—not the shape-shifting robots, but the crucial devices used on the power grid,” write spokespersons at the U.S. National Renewable Energy Laboratory, invoking the other use of the term “Transformers” for the legacy brand of children’s toys. “Almost every kilowatt-hour of electricity flows through a distribution transformer. Similar to how a traffic cop manages the flow of vehicles on a road, distribution transformers manage the flow of electricity along the power grid by changing high-voltage electricity from transmission lines into low-voltage electricity before it reaches consumers.”
Reporting from the Financial Times shows a current producer of the transformers is “overwhelmed,” and not able to meet demand, with wait times for the components ramping up several hundred percent, suggesting that utilities will wait years for delivery.
What Does a Transformer Do?
A transformer is essentially a ‘repeater’ for electrical power, made to even out the delivery of electricity in distributed systems.
The transformer steps up or steps down the voltage to make it appropriate for its route to endpoint delivery.
That means these components will be very much in demand as innovators scale data centers to handle the power demand of large neural networks.
We’re also transitioning to a renewable energy grid, which could create additional strain.
“Today, the grid is undergoing a forceful, unprecedented shift, propelled by the triple-impact push for decarbonization, the necessity to replace aging infrastructure, and the urgent expansion required due to vigorous demand, including from new loads like data centers and electrification of the industry, commercial, transportation, and residential sectors,” wrote Sonal Patel at Power this spring. “If the current transformer boom could be boiled down to two key factors, they would cover a critical need for reliability and resilience.”
IT Semantics: Transformers, and Transformers
Here’s something that’s important to keep in mind, too, as you’re reading about this problem.
Don’t confuse these types of electrical hardware with another very different piece of infrastructure for neural networks that is also, confusingly, called ‘transformers.’
The new kind of transformer, which is responsible for certain kinds of logic operations in AI, is a system that directs the attention of the neural network to enable capacities like problem-solving and reasoning.
A company called Etched made history this year by etching this design right onto a chip, with the proprietary Sohu design, in the race toward more specialized GPU and logic components for computer power.
So the two different types of transformers belong to two different power worlds: raw electrical power or compute.
Vendor Demand
What about AWS and Azure? These dominant vendor giants have big power needs. How are they going to source energy?
Breaking news suggests at least one of these mammoth companies has a plan.
“AWS announced it has signed an agreement with Dominion Energy, Virginia’s utility company, to explore the development of a small modular nuclear reactor, or SMR, near Dominion’s existing North Anna nuclear power station,” wrote Diana Olick for CNBC Oct. 16. “Nuclear reactors produce no carbon emissions.”
That might stem demand a bit, but presumably, all of that power will still be flowing through the grid…
In some ways, you could think about the lack of electrical transformers as similar to the microprocessor shortage we saw in past years. The primary maker in Taiwan was supplying so much of global demand, and was maxing out, and other issues were cropping up, too.
Eventually, the U.S. administration, arguably motivated by a form of IT protectionism, proposed and passed a $50 billion Chips Act for U.S. production, and we seem to have seen some easing of this particular supply chain issue.
But when you’re looking at price and delivery, anytime something like this comes up, we’ll need accelerated production to help manage and improve our electrical grid for the future!
Source: www.forbes.com…