From Lab to Deployment: Three Computational Trends Shaping the Future of Industry at TechConnect 2026

In early March, the Matlantis US team headed down to Raleigh, NC, for TechConnect World 2026. It’s one of the largest conferences in the country for applied innovation, covering everything from advanced materials and energy to sustainability and AI. While a conference like TMS (which we headed to right afterwards, see our report here is the perfect place to show off the newest computational methods, TechConnect is where those methods actually enter the commercial pipeline. The attendees are from a mix of startups, national labs, academia, and industry, and are the people actively building better batteries, designing improved semiconductor packaging, scaling up new chemistries, and more. From Matlantis’s side, our very own Dr. Taku Watanabe presented on how AI accelerated atomistic simulations with Matlantis are bridging the gap between accuracy and scale.

This application-first mindset shaped our biggest takeaways, and the plenary keynotes set the tone right away. For example, we saw a carbon materials presentation use machine learning interatomic potentials (MLIPs) to explain an experimental puzzle about diamond growth from liquid metal. That’s a huge signal that MLIPs are becoming a mainstream materials tool, not just a niche computational topic. We also heard a refreshingly grounded take on quantum computing; it’s going to complement classical computing, not replace it, meaning classical methods are still very much the primary tool for today's materials simulation. Add in a deep dive on silicon carbide as the critical material for the electrification era, and a strong reminder that advanced computing and semiconductors remain top priorities for federal funding, and it was clear that the underlying message was all about getting things out of the lab and into the real world.  After this, we headed to the technical sessions, where three major trends in applied materials modeling and simulation stood out. While the computational methods themselves were interesting, their scale and application to real-world engineering problems really stood out.

Trend 1: AI-Accelerated Workflows Are Heading for the Factory Floor

We have written a lot lately about the rise of AI in materials discovery. At events like ACS Fall 2025, the conversation was mostly focused on the conceptual shift from traditional computational chemistry toward AI-powered approaches. At TechConnect, however, the big takeaway was just how operational these workflows have become. Companies and labs are actively building real production systems right now instead of just running academic proofs of concept. We saw large-scale prediction platforms screening tens of millions of virtual polymer candidates through successive property filters. These platforms look at things like thermal stability, conductivity, mechanical performance, and gas permeability to identify a handful of synthesizable targets for real applications like PFAS-free fuel cell membranes. We saw self-driving laboratories automating polymer synthesis and testing in closed loops guided by active learning. We saw agentic AI assistants stepping in as conversational tools that researchers can query to predict properties, generate candidates, and visualize results without writing a single line of code. The maturity of these tools was obvious, but a recurring data infrastructure bottleneck also became clear. One student award winner presented an open-source platform for experimental materials data, highlighting a truth the community increasingly recognizes. Without well-structured and accessible data, even the most powerful AI tools will eventually plateau. The applied takeaway here is straightforward: the companies that invest in organizing their proprietary experimental data are the ones that will capture the most value from AI-driven discovery.

Where Matlantis Fits In: At the end of the day, any AI workflow is only as good as the property predictions driving it. Whether a team is building a massive screening platform or setting up an autonomous research loop, they need incredibly fast and accurate atomistic data across diverse chemistries. This is exactly where Matlantis steps in with our ready-to-use simulation engine. By using PFP (PreFerred Potential), our universal MLIP, researchers can generate DFT-quality data at the massive speeds these workflows require. It completely removes the need to spend weeks or months building system-specific potentials from scratch for every new material project.

Trend 2: Materials Simulations and MLIPs Are Being Adopted Across Engineering Domains

In our TMS 2026 recap, we talked about how MLIPs are being industrialized and integrated into multiscale production pipelines. TechConnect revealed the other side of that story, in which applied researchers and engineers who are not computational specialists are now adopting MLIPs. We are seeing them used across domains that never would have appeared in a computational session just a few years ago.

During a plenary keynote on carbon materials, researchers used MLIPs to resolve an experimental question about diamond growth. They were not just showing off a new computational method. They were using it as a practical tool to understand a complex synthesis process. Over in a semiconductor session, machine learning methods enabled the simulation of device heterostructures with over 12,000 atoms. This allowed the team to predict resonance behaviors that simpler models would completely miss. In another session focused on two dimensional materials, teams paired atomic scale simulations with automated microscopy to figure out exactly how to fabricate and control individual nanopores.

When our own team presented our universal MLIP, PFP, the audience's questions were incredibly revealing. Instead of asking about the underlying model architecture, which are questions we would have received a few ago, they wanted to know about validation strategies across material families, how the models apply to systems like amorphous electrolytes, and how to choose the right level of theory for their specific engineering problems. That shift in the Q&A really tells the whole story. When users stop asking if a technology works and start asking how to apply it to their specific system, you get the sense that it has crossed the threshold from a neat research tool into essential engineering infrastructure.

Where Matlantis Fits In. As MLIPs transition into standard infrastructure, the focus is shifting away from deep model building expertise and moving toward accessibility and reliability. We designed the Matlantis platform specifically for this shift. It allows researchers across any domain to run large scale simulations without needing to maintain specialized hardware or develop their own custom potentials. Because we use a single universal potential that works across the entire periodic table, an engineer designing battery electrolytes and an engineer modeling semiconductor interfaces can use the exact same tool. It completely lowers the barrier to entry for the diverse applied communities we spent time with at TechConnect.

Trend 3: The Biggest Engineering Challenges Are Converging on Multi Component Interfacial Simulation

While the first two trends outline how the computational toolkit is maturing and who is actually using it, the third trend focuses on the problems people are trying to solve. Across all the technical sessions, we were struck by how consistently the most urgent engineering challenges share almost the exact same computational profile. Regardless of the field and application, many researchers are dealing with multi-component systems and chemically heterogeneous interfaces under conditions that require large scale atomistic treatment.

Over in the semiconductor sessions, the conversation was largely about system level integration more so than device layer physics. Presenters emphasized that packaging and thermal management are rapidly becoming the primary performance limiting factors. Current thermal simulations often rely on interface property assumptions that diverge from what is actually being manufactured. Teams are developing novel materials for stress relief in 3D integration and creating next generation electro-optic modulators. Designing and screening these new technologies requires the ability to simulate realistic multi-material stacks at the atomic scale.

When looking at sustainability and critical materials, the challenges all come down to handling chemical complexity at a massive scale. Lithium demand is projected to surge over the next ten years. Because of this, direct extraction technologies like advanced membranes and ion exchange materials require intensive computational screening across vast compositional spaces. Meanwhile, PFAS remediation is advancing toward deployable solutions such as photocatalytic metal organic frameworks. Simulating the complex guest-host interactions and catalytic mechanisms of these frameworks is essential for rational design. The biorefining community is also working hard to integrate renewable feedstocks into existing petroleum infrastructure. This creates tough catalyst optimization problems that span very diverse reaction chemistries.

In grid scale energy storage, the field is diversifying decisively beyond lithium. Interestingly, discussions around zinc-based chemistries made up many of the battery sessions. We saw everything from insertion and conversion cathodes to unique electrolyte additives designed to suppress dendrites and corrosion. Several companies are actually already manufacturing these systems and deploying them at real sites. Thermal storage, organic flow batteries, and long duration solutions really rounded out the picture. Across all of these diverse technologies, the simulation needs are strikingly similar. Researchers need to predict intercalation mechanisms, screen electrolyte formulations, and model degradation pathways in highly complex aqueous environments.

Where Matlantis Fits In.  What unites these three domains is that the underlying problems are inherently multi-component and interfacial. This is exactly where classical force fields break down and traditional DFT simply cannot scale. Universal potentials like PFP bridge this gap perfectly. They enable researchers to model realistic packaging stacks, complex contaminant interactions, or zinc anodes in multi-salt electrolytes without having to develop a custom potential for every single new combination. As these fields race to move from the lab into commercial deployment, having a simulation tool that keeps pace with the speed of experimental iteration becomes a massive competitive advantage.

Looking Ahead

If our time at TMS showed us how computational toolkits are maturing, TechConnect showed us where real world demand is heading. AI-driven discovery is almost operational and we are seeing MLIPs being adopted by frontline engineers instead of just specialized modelers. Furthermore, the toughest problems across semiconductors, energy storage, and sustainability are all pointing toward the need for large-scale multi-component atomistic simulations. That is exactly what universal MLIPs were built for. Researchers want to know how to make AI and atomistic simulations work for their unique systems at their specific scale and on their tight timelines. We built Matlantis to be exactly the partner they need to get those solutions out the door.