The compute power of artificial intelligence is revolutionizing nearly everything it touches, yet the tremendous demand and limited supply have put increasing pressure on energy sources to keep up the pace. Semiconductor fabs and other manufacturing businesses, in particular, must find innovative solutions to optimize their energy use and management. Power may be the hard barrier that will stop all progress if we can't find viable solutions. At SEMICON West, Mark Bidinger, president of commercial and industrial segments at Schneider Electric, shared his insights into this issue and potential solutions.
Marcy LaRont: Mark, as we look at what will be needed to power AI, semiconductor fabs, and other electronics technology, what can we expect going forward in the context of electronics?
Bidinger: Going forward, there will be a significant gap between supply and demand. In just the semiconductor industry alone, by 2030, we will face a 4-gigawatt (GW) gap between supply and demand, and what we produce must be used much more effectively and efficiently than ever before. As a frame of reference, 4 GWs is 4 million kilowatts (kW). The average home in the U.S. uses 1.2 kW of power per day. 4 GW could theoretically power 3.3 million homes for one hour.
First, though, I believe we still have a significant amount of energy savings runway, especially looking at brownfield institutions. Roughly 50% of their energy consumption is in operating the fab facility itself, and the other 50% is used in the processes. There are probably some pretty easy ways to achieve a 10–20% energy savings, especially if we run that challenge through an AI/ML (machine learning) process and see what it comes up with. We've been doing that with some clients. We take some of those solutions and scale them.
The process of looking for energy savings beyond that will be more complex. It is a challenging ecosystem to manage, but there is still considerable room for energy savings when considering the entire picture.
Energy resiliency and reliability in the past had more to do with preventing facility downtime. In the future, resilient, reliable, green, and cost-effective energy sources will be more interactive with utility grids. They will be the energy backbone of businesses and communities, giving the fab plant or campus more control.
LaRont: Does that mean that facilities must be more reliant on themselves for their power needs and not solely reliant on external sources?
Bidinger: In energy, we say “independency,” which means we are bringing microgrids and other things into the equation. Schneider has been recognized as the number one integrator of microgrids in the world. We're not making the fuel cells and other technologies, but we're bringing it together in one envelope.
We are currently working with a facility where there are 35 different AI systems overlaid onto it between the building and manufacturing processes, finding and implementing where energy can be conserved or used more efficiently. It has produced quite a lot of cumulative energy savings. That is just one customer and one building, and this is very scalable going forward, not only for this client but globally across all industries.
LaRont: Mark, what are microgrids, and what do they offer in terms of better power management solutions?
Bidinger: A microgrid is an aggregate of more than one energy source to leverage or augment what you're getting from the utility. In many cases, the microgrid will give you access to incremental power if needed. However, it also provides you with the ability to interact with the utility as needed. You're becoming a kind of “prosumer,” if you will, which is very much how most industries need to think about managing their power going forward.
LaRont: That sounds like more of a partnership mentality.
Bidinger: Yes, it is more of a partnership relationship.
LaRont: Speaking of collaboration, please explain Schneider's Catalyze program and its purpose in the broader business picture?
Bidinger: Catalyze consists of nine end users (companies/entities). Schneider is helping aggregate their consumption of green, reliable, cost-effective energy. We're not breaking down any secret sauce, but as an industry, we're helping aggregate their need for power and how we can help manage power and energy consumption going forward.
LaRont: Why are partnerships like this key to solving these bigger issues industry is grappling with?
Bidinger: There are myriad reasons why these collaborations are important. At first, of course, everything is financially driven. If you look at the amount of energy that this industry is consuming, there are more eyes on it than ever before. How high-use businesses consume and manage energy efficiency will be scrutinized in the context of the community’s greater energy needs. However, as an industry, as long as we remain proactive and continuously learn and apply those applications as we go, we will also continue to evolve.
LaRont: In terms of power, we're facing some very real hard stops that may affect innovation and progress. Power availability may be the biggest barrier that we face. In the next five to 10 years, what will significantly address that challenge?
Bidinger: Clearly, the utilities are ramping up, but I don't think they can ramp up at the speed various industries require. In 2026, AI alone will consume over 1,000 terawatt-hours (TWh) of energy, roughly equal to the annual electricity consumption of India.
Access to energy will be part of the equation, but it will be more complex than just “access” as we have come to think about it—it will have to be “reliable access” to energy.
There’s also the question of how we use it as an economic vehicle for or within businesses and communities. I am not an economist, but the power and energy consumption, management, and access model of the near future will need to be a much more interactive part of communities going forward.
LaRont: What advice would you give to industry companies?
Bidinger: It is no longer about using a three-phase uninterruptible power supply (UPS) to keep a fab line up 24/7, but rather about using that same digital infrastructure with the ability to create some elasticity when needed, to go back to the grid.
LaRont: Do you believe that AI will be just as valuable in helping solve this power challenge as the technology is in consuming it?
Bidinger: I think AI with machine learning is phenomenal, and there is a lot of potential there to help create solutions for power access and efficiency.
LaRont: Mark, that has been really interesting. I appreciate your time.
Bidinger: Thank you, Marcy.
