Fireside Chat with Ekin Dogus Cubuk, Co-CEO of Periodic Labs | RAISE Summit 2026

Problem Statement: The Data Limitation in Scientific AI

• Current Large Language Models (LLMs) rely on static, scraped internet data, which lacks the context required for complex physical sciences (e.g., engineering, semiconductors).
• Scientific experiments inherently lose context; a single mole of chemicals contains ~10²³ atoms, far exceeding storage capacity, and negative results are rarely published.
• Unlike coding or mathematics where full context is available, physical discovery requires iteration (often 100+ trials) rather than a single static dataset.
• High variability in experiments (location, operator) makes human execution less consistent than automated systems.

Company Founders and Background

• Doge (Founder): PhD in Physics (2010); built ML tools for solid-state physics and materials science; former researcher at Google Brain and DeepMind (GNOME paper author).
• Liam Fieders (Co-founder): Undergraduate Physics background; former Google Brain researcher; co-created ChatGPT at OpenAI.
• Strategic Synthesis: The company leverages Doge's expertise in physical systems/simulations, Fieders' background in LLMs/RL, and recent breakthroughs in robotics.

Core Technology and Infrastructure

• Methodology: Combines LLMs, Reinforcement Learning (RL), simulations, and high-throughput robotics to create an automated "lab that iterates with the universe."
• Physical Setup: A new laboratory in the Bay Area featuring robotic arms for chemical experiments and sample transport, integrated with traditional scientific tools (microscopy, diffraction) controlled by AI.
• Primary Advantage: Robots provide higher consistency than humans and enable significantly higher throughput; the company claims to have one of the highest experimental throughputs ever attempted for this specific domain.
• Current Limitations: Robots lack fine motor skills for delicate cleaning (e.g., cleaning crucibles), a task still requiring human intervention.

Strategic Focus: Semiconductors and Materials

• Target Sector: Initial focus is on semiconductors and future materials due to the industry's massive impact on Moore's Law and the feasibility of simulating atomic-level systems.
• Feedback Loop: Aims to create a recursive improvement cycle: AI improves chips → better chips enable faster/better AI → AI further improves chips.
• Evolution from Biology: Unlike drug discovery (which faced historical skepticism), current AI models (e.g., graph neural networks since ~2016) can now model general atomic interactions across the periodic table, not just specific force fields.
• Human-AI Division of Labor: AI handles tool execution, procedure, and data analysis; humans handle hypothesis generation, creativity, and determining "truly novel" research directions.

Operational Targets and Timeline

• Goal: Reach a throughput of 1,000 experiments per day by the end of the summer.
• Immediate Focus: Improving semiconductor process engineering (deposition, etching, lithography) and solid-state chemistry/powder synthesis.
• Experimental Method: "Baking" approach: combining precursors in specific ratios and heating them at specific temperatures to observe reactions.
• Scope Constraints: The lab avoids radioactive or highly toxic elements (e.g., uranium), focusing on inorganic synthesis that can be safely iterated upon.
• Key Bottleneck Identified: Automated characterization (e.g., X-ray diffraction, SEM) is more difficult than the physical execution of experiments.

Business Model and Market Strategy

• Short-Term Revenue: Focus on process engineering improvements for semiconductor manufacturers rather than pure IP sales.
• Long-Term Vision: Transform material discovery into a profitable business by generating valuable Intellectual Property (IP) once predictive models reach sufficient accuracy.
• Market Context: Materials discovery value is difficult to capture currently (similar to the pre-Genentech era for drug discovery), but profitability is expected to rise with model accuracy.
• Educational Impact: Aims to reverse the decline in physics/chemistry majors by making solid-state sciences exciting through the integration of cutting-edge AI and robotics.