Big Ideas Lab
Today’s toughest industrial challenges are being solved without ever touching the factory floor.
3,314 words · 9 speakers
A scientist looks out into a crowd of some of the nation's most influential manufacturers. Their voices lower as he walks to the center of the stage.
The cool metal of the podium presses against his palms. Bright lights blaze down, washing the room in white, blurring faces into shadowed silhouettes. His heart hammers in his chest, a steady drum of anticipation. But he knows the material. He's conducted the research, he's run the simulations. He begins his presentation.
I had done this beautiful elegant molecular detailed simulation and I was talking about the parameters that went into it.
Everything went exactly as planned. The models worked, the results were clear.
And at the end, they're like, "Okay, any questions?"
The audience lights lift, and he sees their expressions for the first time.
There was just this dead silence.
The scientist awkwardly steps offstage. Not in defeat, in confusion. What just happened? A few moments later, the gap between the world of private industry and scientific inquiry became painfully clear.
And all of a sudden, I hear this guy, and he's like, "Son, I'll just be honest with you. I have no idea what you're talking about."
It's like, "You're talking about atoms and molecules? I have no idea how this relates to anything."
For decades, this gap has slowed progress across American industry. Scientists could model the world in extraordinary detail. Manufacturers knew the real-world problems costing them time, energy, and money. But connecting those two worlds was its own challenge. Now, that divide is starting to break down. Glass manufacturing plants the size of football fields are running more efficiently. Electric vehicle batteries are lasting longer. And paint that once corroded too quickly is being redesigned at the molecular level.
This is the work of the High Performance Computing for Energy Innovation Program, turning simulations into real-world solutions.
Welcome to the Big Ideas Lab, your exploration inside Lawrence Livermore National Laboratory. Hear untold stories, meet boundary-pushing pioneers, and get unparalleled access inside the gates. From national security challenges to computing revolutions, discover the innovations that are shaping tomorrow, today.
Looking for a career that challenges and inspires? Lawrence Livermore National Laboratory is hiring for a Power Grid Dynamics Engineer and a Transmission Grid Engineer, along with many other roles in science, technology, engineering, and beyond. At the lab, every role contributes to groundbreaking projects in national security, advanced computing, and scientific research, all within a collaborative, mission-driven environment. Discover open positions at llnl.gov/careers, where big ideas come to life.
Every innovation begins with a question: Can this be better? Across factories and industrial plants, every decision has a cost: time, energy, money. A single misstep can ripple for years. The High Performance Computing for Energy Innovation Program, or HPC for EI, uses some of the world's most powerful supercomputers to answer that question for private industry. By connecting industrial leaders with computational physicists at national laboratories, HPC for EI turns simulations into solutions, helping companies optimize energy, reduce waste, and stay competitive in a global economy.
Aaron Fisher directs the initiative.
What we do in this program is we find people in heavy industry that uses a lot of energy, uses a lot of resources in the United States, and we match them with computational physicists across the national laboratories. We help them move the needle on their energy needs and move the needle on their resource needs and make the businesses more competitive.
Resources you use every day, like AC in your car.
There's a company that has been trying to design better heat exchangers for car air conditioners. One of our researchers here at Lawrence Livermore did a project with them to improve the design of their air conditioning. They used a technology called topology optimization. So they basically let an AI system have at the design of this heat exchanger. And it was something like 20 to 30% more efficient at moving heat through the heat exchanger than your sort of classic designs. They estimated that if they got that thing into a car, on a hot day, you'd save like 10% of your gas cost for running the air conditioner.
These partnerships between private industry and national labs are funded by the U.S. Department of Energy. As of 2026, DOE National Labs have collaborated with industry on over 200 projects through the HPC for EI program.
One of the things the program has been focused on is reducing energy need and energy intensity in industry. If you can go after an industry and shave a couple percent off of the industry, you'll have a huge impact.
Heavy industry consumes nearly a third of the nation's energy. Even small improvements can ripple across the economy, driving efficiency and keeping American industries competitive on the global stage. Brandon Wood is a project lead within the program. He oversees projects that put this computing power directly into the hands of industry.
We're building a US manufacturing base that is smarter, more agile, more efficient.
Since launching, HPC for EI has scaled rapidly in both size and scope.
The bottom line is we're using computing to solve real-world problems.
We've launched about 200 projects. We've got about 20, 25 in-flight right now. The projects last for about a year. Each project gets a $400,000 budget, which covers the staff at the national laboratory. We work with 12 national laboratories in total. I think we've worked with over 100 different companies at this point. We've worked with really small companies, the startups trying to get a new product off the ground. We've worked with huge, huge companies that actually already do some computational physics, and we help bring them to the next level.
You can see the results of these computations in your own home.
They were modeling the fibers in paper towels, figuring out how to pack more fibers in and make it more observant.
Or in your garage.
There was one that the program worked with. It was literally a project about watching paint dry.
Not the kind of work you'd expect.
What they were doing was painting cars. If you put a thinner layer of paint on the car, you don't have to spend as much energy drying it, and you use less paint.
The results speak for themselves.
So it's two years, literally simulating paint drying, but they actually managed to make an improvement. It was like 30% energy reduction.
The supercomputers across all these projects have logged more than 916 million core hours, enough to keep a single laptop running for thousands of years, tackling problems that no one could solve by hand.
It's the computers and all the support staff that run the computers and all the people that keep the software going on all the computers, and then all the scientists that know how to use them, and decades of code development on supercomputers like this. Right? So it's a lot of technology and expertise to leverage.
Even with these breakthroughs, connecting the worlds of science and industry requires more than technical knowledge.
I think it does take a special part of the brain to be able to bridge that fundamental and applied. The expertise is really critical. And that expertise isn't just the ability to run simulations. It's the ability to translate real problems into simulation language. That is the hardest problem is the complexity of real world operation in industrial environments. They're messy, they're dynamic, they're unpredictable. This is not the way we run laboratories, but it's the way industry operates. So we have to bridge that gap.
To manage that complexity, the program follows a clear step by step process to turn intricate industrial challenges into solvable simulations.
The first thing is actually really important, which is figure out which question to ask. And that may sound trivial because you say, well, the problem was provided by the company. But a problem that's provided from an industrial sort of market driven standpoint, how that translates into the types of simulations that we do is its own problem. That's its own science question. And so a lot of times there's kind of that initial management of expectations. This is what we can do. This is what your problem is. Let's turn that into a real question. And that determines that drives what type of simulation we actually do. In many cases, that's actually harder than running the simulation. Simulation may take time and computing power, but asking the right question is really, really important.
Once the right question is defined, the simulation can begin.
We distilled it down to these are the things we can actually compute. Set it up in the computer, run the simulations, and then comes the analysis part. So you get the data out. Sometimes you can translate that very quickly to the original problem of interest. A lot of times, because it's high performance computing, which means almost every single time you're generating a lot of data. And so distilling also all of that data down to the parts that really matter.
The computer runs the scenarios, but it's the scientists who interpret their results.
Once they've done that, they have a tool that they can really use to study that process. And then they can start doing things like changing what they want to do. What if we use three burners here instead of two? Or what if we tried a different set of temperatures? They can start asking those what if questions to see how the process evolves without having to do it.
Sometimes the most important discovery a simulation makes is realizing you're solving the wrong problem.
That insight can redirect years of research, saving time, money, and resources while pointing the way to genuine innovation. Inside the automotive industry, one project focused on the future of solid-state batteries for electric vehicles.
They're trying to optimize it. And they were focused on one line of inquiry with their optimization, but they weren't sure if that was the right one. And so we've done a lot of battery simulations and battery material simulations in my group. So we took this on and we started simulating the properties of these systems. And we actually found that that line of inquiry was wrong. It turned out that the direction that they were going was maybe going to solve one of those problems will make the other problem much worse.
A digital time machine, seeing decades of wasted effort before it even happens, and redirecting research choices to the path that actually leads to results.
And so they took that and they'd been implementing it, going in a new direction. We're actually building on that with a second phase of that project.
The simulation runs inside a computer, but the systems they represent exist out in the real world, inside factories, inside furnaces, inside machines that operate on a massive scale. So how can a limited number of lab hands give insight to a seemingly endless number of industry problems? They create new hands. Digital hands.
It will no longer be only human.
Lawrence Livermore National Laboratory is hiring. If you're passionate about tackling real-world challenges in science, engineering, business, or skilled trades, there's a place for you at the lab. Right now, positions are open for a Senior Guidance, Navigation, and Controls Engineer and a Senior Business Operations Specialist. These are just a few of the more than 100 exciting roles available. At Lawrence Livermore, you'll work on projects that matter, from national security to cutting-edge scientific advancements. Join a team that values innovation, collaboration, and professional growth.
Explore opportunities at llnl.gov/careers, where your next career move could make history.
Okay, where were we?
Okay, let's make the battery more efficient using this method.
Interpreting request: increase efficiency.
We're trying to train what they call agentic AI systems.
Reframing goal: maximize usable energy over life cycle within thermal and material limits.
An AI agent is an AI system that can take a goal and work toward it on its own.
Actually, let's try this.
Running constraint simulations. Optimal pathway identified. Life cycle efficiency improved by 11% without instability. An agentic AI inflection point.
The world is now awakened to the agentic AI inflection. We call it agentic AI. It can perceive and understand the context of the circumstance.
I think that every business in the future, just like they have an email address and a website, I think every business is going to have an AI agent.
I think every customer I talk to, agentic is the destination they're trying to get to.
Instead of a scientist guiding every step, the agent can be sent to explore different possibilities, test variables, and learn from the results as it goes.
And the goal there, we could hand them an agent and the code that could make it easier for companies to maybe do some of this work themselves.
When agentic AI systems are built on top of decades of scientific research from national labs, the implications quickly expand.
AI has been involved rapidly in the past few years, and the capability of AI is increasing at a speed that we can't even imagine.
Yiping Hu is a project lead for HPC for EI, exploring the intersection of artificial intelligence and mechanical engineering.
The HPC projects that I've been involved in, the goal is to help the company to accelerate their simulations.
AI tools aren't just speeding up routine tasks. They're giving scientists the ability to rethink how complex manufacturing problems are solved. Opening doors to discoveries that were once out of reach. And in heavy industry systems, that power isn't just convenient, it's critical. Nowhere is that more clear than in glass manufacturing. The tanks that hold the molten material can be enormous. At that scale, even the smallest problem can cascade into massive waste, affecting a material the nation depends on.
The melting tank itself, they wanted to know what's happening inside of it. Because you can't like see through it, and it's really large. So they want to know the flow field inside of this large melting tank. And they want to know, because they have several points that they can heat up temperature or heat up the tank. And different combinations of these temperatures would have a huge effect on the flow field. And how the flow field looks like would have a large effect on the final quality of the glass. Currently, like what they do is they run a big simulation. As you can imagine, given how large that melting tank is, the simulation could take weeks or months to run for one set of parameters.
That take too many times. And obviously, it costs a lot of money and then also produce a lot of carbon dioxide.
Instead of running large-scale simulations over and over again, the Lawrence Livermore HPC for EI team built an artificial intelligence model to predict the results.
We helped them develop, we call it target model or reduce order model. Basically is a machine learning based model that can do these type of predictions. So given these control parameters, you fit it in through this machine learning model and then it will instantly gives you the predicted flow field of this large melting tank.
What once required hundreds of simulations can now happen in seconds.
At the beginning, they will provide us with five, six data points. And then we will use that to train the model and then test it out. And also feed this data into the model that we already developed before, either from the model we use for other projects or just like the model that we tried on some other toy example simulations. But we do try to use these data and try to use the model that we have and see, for example, which model works the best.
The result is a breakthrough, not just in speed, but in real world impact.
It definitely saved them over like million dollars per year and also huge CO2 reduction. If they further advance the model that we provided and integrate it in their manufacturing pipeline, I think that will make an even more huge impact later on.
The tools that accelerate industrial innovation don't just stay on the factory floor. They also strengthen the technologies and materials our nation relies on every day.
We've got these national security laboratories that are participating in this program. And these national security laboratories have put together simulation codes to solve national security problems.
These are the same codes used to model critical national security systems.
It lets us stress test our codes in a way that we haven't been able to before. And in some cases, we've improved our codes through these projects. Those answers aren't quite right. We need to improve this and add this bit of physics. And then that folds back into the national security mission that now they have a better code to solve their problems. And this has happened numerous times in the national security laboratories.
Beyond codes and simulations, the program is also tackling the materials that keep our industries and our nation running.
In addition, we've had a lot of projects that are sort of focused on producing critical materials. For instance, we just started a project that is focused on producing magnesium. Here in America, we don't produce a lot of our magnesium in America. It's an onshoring opportunity. We're starting to look more at things like that as well.
Every success strengthens American manufacturing, making factories smarter, more agile, and more competitive on the global stage. The world's hardest problems rarely behave like lab experiments. They're messy, interconnected, industrial. But with the right tools and the right collaborations, complexity becomes something we can explore.
I actually think the biggest metric for success, it's the duration. It's the longer lasting relationship of trust and understanding of what HPC can provide beyond that one problem that you focused on for 12 to 24 months. We started the conversation, we built a relationship, we said this is what computing can do for you. We solved some piece of that problem to the point where they thought this is a new tool. This is a new collaboration. This is something we can leverage and we built on that.
Something we can understand and something we can solve. At its core, HPC for EI is developing more than just technology. It's developing expertise.
I would say it's really navigating complexity that is the most compelling in my mind. And so I think where we're going to see the biggest benefits is not actually just in making things faster. I think it's making us able to tackle more complex problems.
And that is where the greatest potential lies, not from solving the same problems faster, but from solving problems we once thought were impossible.
Thank you for tuning in to Big Ideas Lab. If you loved what you heard, please let us know by leaving a rating and review. And if you haven't already, don't forget to hit the follow or subscribe button in your podcast app to keep up with our latest episode. Thanks for listening. Lawrence Livermore National Laboratory is hiring. If you're passionate about tackling real-world challenges in science, engineering, business, or skilled trades, there's a place for you at the lab. Right now, positions are open for a Senior Guidance, Navigation, and Controls Engineer and a Senior Business Operations Specialist.
These are just a few of the more than 100 exciting roles available. At Lawrence Livermore, you'll work on projects that matter, from national security to cutting-edge scientific advancements. Join a team that values innovation, collaboration, and professional growth. Explore opportunities at llnl.gov/careers, where your next career move could make history.
Page updated October 31, 2026 · transcript and summary are machine-generated