Key insights from the 2026 MaRDA Virtual Annual Meeting on data, AI, and the future of materials science

The sixth annual Materials Research Data Alliance (MaRDA) Virtual Meeting brought together researchers, industry leaders, and government representatives to explore how data, artificial intelligence, and shared research infrastructure are reshaping the future of materials science. With more than 460 registrants, the three-day meeting returned to a central question: How do we build the shared data, infrastructure, and workforce capacity needed to move faster, and move breakthroughs into real-world use?

While conversations spanned policy, research, and implementation, a consistent theme emerged: AI’s impact will be limited without purposeful data foundations. Attendees repeatedly pointed to the need for FAIR data practices, interoperable infrastructure, and trustworthy workflows—alongside the rise of automation and self-driving labs that are making materials R&D more continuous, scalable, and model-informed.

Key Takeaways:

• From discovery to deployment: The field is shifting from “find a new material” to “deliver a complete material solution,” including processing pathways, validation, and manufacturability.
• AI needs infrastructure, not just models: Progress depends on shared tooling, repositories, metadata practices, and data governance—not only better algorithms.
• Autonomous labs are becoming real R&D engines: Robotics + closed-loop experimentation + real-time analytics are transitioning from pilots to a new mode of discovery at scale.
• Industry adoption is accelerating, but trust is the bottleneck: Organizations want AI that is auditable, testable, and reliable in regulated, high-stakes environments.
• Data readiness starts upstream: The push is toward making datasets machine-usable at creation, rather than retrofitting structure after the fact.
• Workforce skills are expanding: Materials researchers are increasingly expected to understand data management, software engineering practices, and AI-enabled workflows alongside domain expertise.

Day 1: National Priorities, Shared Infrastructure, and the “Last Mile”

The meeting opened by framing materials innovation as a data-and-AI-enabled ecosystem problem, not just a set of individual research breakthroughs. Discussions emphasized how AI and automated experimentation are changing the pace of discovery, while also surfacing a persistent constraint: translation from lab results to deployable systems remains difficult.

A major thread centered on interagency coordination and national research priorities, including the need for aligned infrastructure investments—such as integrated platforms that connect computing, experimental facilities, AI systems, and data resources. Attendees also discussed why a single, unified materials data system may be unrealistic, but agreed that successful models for data sharing and FAIR practices should be identified, replicated, and scaled.

Another key conversation focused on the reality that many valuable datasets still sit in legacy formats (including lab notebooks and disconnected internal systems). This limits reuse and makes it harder for AI systems to learn effectively. Participants also debated the role of standards: while “perfect” universal standards may be unattainable, many emphasized that some shared structure—schemas, metadata conventions, and provenance capture—is essential for interoperability, reproducibility, and meaningful AI use.

Industry-oriented discussions reinforced a point repeated throughout the day: AI may accelerate discovery, but the “last mile” to marketable products—validation, manufacturing constraints, supply chain considerations, regulatory realities, and integration into systems—often determines whether innovation succeeds.

Day 2: AI in Manufacturing and the Emergence of Autonomous Experimentation

Day two highlighted the growing push to apply AI not only in discovery, but across product development and manufacturing. Discussions explored where industry sees value today and what gaps are slowing adoption.

A recurring theme was scalability: organizations need AI tools that work across broad portfolios and varied conditions, rather than one-off models for narrow use cases. Attendees also emphasized that AI recommendations must earn trust through traceability, testing, and validation, especially when changes affect manufacturing lines, regulated environments, or safety-critical systems.

The meeting then shifted into the future of labs: self-driving laboratories and autonomous experimentation. Attendees discussed how AI-guided closed-loop systems—combining robotics, real-time characterization, and automated decision-making—can dramatically speed iteration cycles, improve experimental consistency, and generate richer datasets. However, discussions also stressed that autonomy is not about replacing traditional labs; rather, it is about creating complementary capacity and enabling experimentation at scales that would otherwise be impractical.

A forward-looking takeaway from this day: the most impactful autonomy will come not from one “super lab,” but from interoperable, modular systems where tools can connect, communicate, and share workflows across institutions.

Day 3: Workforce at AI Speed and Building “AI-Ready” Data

Day three focused on the foundations that determine whether AI-enabled materials science can scale responsibly: people and data.

In workforce discussions, attendees explored how education is changing to keep pace with AI-enabled R&D. Conversations emphasized a shift from teaching coding as a standalone skill toward teaching software engineering habits in the domain, including documentation, reproducibility, workflow management, and responsible use of AI tools. Participants also discussed how low-cost automation tools and project-based learning can expand access to hands-on experience with modern experimental methods, while still ensuring that students build durable fundamentals in materials science.

The conference concluded with a deep dive into materials data repositories and AI-ready data. Discussions drew an important distinction:

• FAIR data is foundational (findable, accessible, interoperable, reusable).
• AI-ready data often requires additional structure, harmonization, and machine-readable metadata so that models and AI agents can use it reliably.

Several practical messages surfaced:

• Make data AI-usable at the start, not after a project ends.
• Include provenance and metadata so results are interpretable and reproducible.
• “Negative” or non-ideal outcomes can be just as valuable for learning as successes.
• Incentives for data sharing remain a challenge; participants discussed better mechanisms for credit and attribution for dataset contributions.

Across three days, the meeting made clear that the next phase of materials innovation will be shaped less by isolated breakthroughs and more by shared infrastructure, interoperable workflows, and a workforce fluent in both materials science and data-driven methods.

The direction is clear: materials research is moving toward an ecosystem where AI models, autonomous labs, and repositories operate together, turning experiments into reusable knowledge faster, and helping the field advance from discovery to deployment with greater speed, trust, and impact.

You can review the MaRDA2026 recordings and agenda here.