Artificial Intelligence April 3, 2026

Periodic Labs: How AI’s Fastest-Rising Startup Hit 1650% Growth

A startup that didn’t even have a name or bank account in early 2025 is now in talks for a $7 billion valuation. Periodic Labs – founded by a former OpenAI VP and a DeepMind materials science lead – emerged from stealth in September 2025 with a $300 million seed round, one of the largest ever recorded. Six months later, the San Francisco-based company is being cited as AI’s fastest-rising startup, with reported growth of 1650%, a figure likely tied to its staggering valuation trajectory from seed to multi-billion-dollar funding discussions.

What makes Periodic Labs different from the flood of AI companies chasing the next chatbot or image generator? It’s building something far more ambitious: AI scientists paired with autonomous physical laboratories that synthesize materials, run experiments, and generate entirely new scientific data. In a landscape where the best frontier models have already consumed an estimated 10 trillion tokens of internet text, Periodic Labs is betting that the next frontier of AI isn’t more data from the web – it’s data from reality itself.

From a Noe Valley Walk to a $300 Million War Chest

The origin story reads like Silicon Valley folklore, except it’s barely a year old. Liam Fedus, who served as VP of Post Training at OpenAI and was part of the small team that created ChatGPT, left the company in March 2025. His departure tweet set off a frenzy among venture capitalists. One investor reportedly wrote a “love letter” to the not-yet-named company. Others sent multi-page pitch documents – not to Fedus, but pitching themselves as worthy backers.

The first call Fedus took was from Peter Deng, a former OpenAI colleague who had joined seed-stage firm Felicis. Over coffee in San Francisco’s Noe Valley neighborhood, Fedus explained his vision. The conversation moved to a walk through the area’s steep hills, where Deng – sweating through a sweater on a day that had turned unexpectedly hot – heard Fedus say something that “literally stopped me in my tracks.” The insight was deceptively simple: everyone talks about doing science with AI, but to actually do science, you need to run real experiments.

Deng committed on the spot. There was just one problem – Felicis’ lawyer pointed out they couldn’t sign a term sheet because the company wasn’t incorporated yet. It didn’t even have a name.

By September 30, 2025, it had both. Periodic Labs announced its $300 million seed round led by Andreessen Horowitz, with participation from Felicis, DST Global, NVentures (Nvidia’s venture capital arm), Accel, and a roster of individual backers including Jeff Bezos, Elad Gil, Eric Schmidt, and Jeff Dean.

The Founding Team Behind the Hype

Periodic Labs’ credibility rests heavily on its founders. Ekin Dogus Cubuk led the materials and chemistry team at Google Brain and DeepMind, where he spearheaded GNoME – an AI tool that discovered over 2 million new crystal structures in 2023. He also co-authored research documenting a fully automated robotic lab that created 41 novel compounds from recipes suggested by language models. Fedus, meanwhile, led the team that created the first trillion-parameter neural network and ran OpenAI’s post-training team, which refines models after initial development.

The broader team is equally stacked. Key hires include Alexandre Passos, a creator of OpenAI’s o1 and o3 models; Eric Toberer, a materials scientist with key superconductor discoveries; Matt Horton, who built two of Microsoft’s generative AI materials science tools; and Dzmitry Bahdanau, co-creator of the neural attention mechanism that underpins modern AI. The founding team collectively contributed to ChatGPT, GNoME, OpenAI’s Operator agent, and MatterGen.

Every week, team members give graduate-level lectures to each other across disciplines. As Cubuk put it, the company believes “tight coupling is extremely important” – everyone needs to understand all parts of what they’re building.

Why the Internet Isn’t Enough for Science

The core thesis driving Periodic Labs is that AI has hit a wall. The internet’s text – roughly 10 trillion tokens – has been fully consumed by frontier models. Better training techniques can squeeze more value from existing data, but there’s a fundamental limit to what can be learned from text alone.

Scientific literature compounds the problem. Published papers overwhelmingly report positive results. Negative experiments – the ones that didn’t work – rarely make it into journals, yet they contain invaluable information about what doesn’t work and why. Formation enthalpy labels, for instance, carry such high noise that training on them doesn’t produce sufficiently predictive models. The epistemic uncertainty that matters in materials science simply can’t be resolved without running an actual experiment.

This is the gap Periodic Labs was built to close.

How Autonomous Labs Change Everything

The company’s approach starts at the quantum mechanical energy scale, where chemistry, materials, and solid-state physics operate. Periodic Labs is building powder synthesis laboratories where robots mix chemical precursors and heat them to discover new superconductors, magnets, and heat shields. These are relatively simple methods, but they generate rich physical data – each experiment can produce gigabytes of unique information that exists nowhere on the internet.

Component Function Key Advantage
AI Agents Read literature, run quantum simulations, predict material properties, direct robotic experiments Faster hypothesis generation than human researchers
Robotic Hardware Powder synthesis – mix precursors in precise ratios, heat, characterize properties Enables thousands of experiments daily vs. human limits
Data Loop Collect GB-scale experimental results including negative outcomes, feed back to AI Creates proprietary datasets that don’t exist anywhere else
RL Environment Nature itself verifies predictions – synthesize a material and definitively know if predictions were correct Grounded in physical reality, not internet text

The key insight is that nature becomes the reinforcement learning environment. When the AI predicts a material’s properties and the lab synthesizes it, there’s a definitive answer about whether the prediction was right. This mirrors the pattern that has driven AI’s fastest progress – domains like math and code where results are verifiable.

There’s a powerful flywheel at work: better AI proposes better experiments, better experiments generate better data, and better data improves the AI. If this loop works at scale, it could become one of the most important architectures for scientific discovery in the coming decade.

Real Customers, Real Problems

Periodic Labs isn’t just a research project. The company is already deploying solutions with industry partners across semiconductors, space, and defense – sectors representing trillions in annual R&D spending. One concrete example: the company is helping a semiconductor manufacturer struggling with heat dissipation on its chips. Periodic is training custom AI agents for their engineers and researchers to interpret experimental data and iterate faster on solutions.

The go-to-market strategy follows a “land and expand” model – solve critical problems with clear, measurable evaluations, demonstrate what’s possible when you optimize against physical reality rather than internet text, and then scale. The target market is enormous: advanced manufacturing, materials science, semiconductors, energy, and aerospace collectively represent roughly $15 trillion of global GDP.

The Growth Trajectory: From Seed to $7 Billion

The 1650% growth figure associated with Periodic Labs appears most directly tied to its valuation trajectory. The company raised its $300 million seed round in September 2025. By early 2026 – roughly six months later – it entered discussions for hundreds of millions more at a $7 billion valuation. No public data on exact revenue figures, headcount growth numbers, or lab output metrics has been disclosed, as the company has prioritized rapid iteration over public disclosures.

What’s clear is the velocity. The timeline from incorporation to $7 billion valuation talks spans approximately six months – an acceleration that reflects both the quality of the founding team and the enormous investor appetite for AI companies that bridge the digital-physical divide.

Milestone Date Details
Fedus departs OpenAI March 2025 Departure tweet triggers VC frenzy
Felicis commits first check Spring 2025 Company not yet incorporated or named
Stealth exit and seed round September 30, 2025 $300M seed led by a16z
Team and lab buildout Q4 2025 26+ elite hires from OpenAI, DeepMind, Microsoft
New funding discussions Early 2026 Targeting $7B valuation with new investors including Coatue, Khosla Ventures, Lightspeed

What Periodic Labs Is Chasing

The company’s ambitions are deliberately enormous. Its primary targets span some of the most consequential unsolved problems in physical science.

Andreessen Horowitz partner Anjney Midha, who led the firm’s investment, frames the opportunity starkly: if Moore’s Law is slowing, this is how we restart it. The bottleneck has been the iteration speed of human-led experimentation. Periodic Labs aims to remove that constraint entirely.

The company has also established a Scientific Advisory Board featuring professors from Stanford and Northwestern, and launched an Academic Grant Program to support pioneering research outside its walls.

Risks, Open Questions, and What Comes Next

For all the momentum, Periodic Labs faces real challenges. Building and scaling physical laboratories is capital-intensive in ways that pure software companies never encounter. Verifying experimental results at scale introduces complexity that no amount of compute can shortcut. And the company is operating in a space where timelines for genuine scientific breakthroughs are measured in years, not quarters – a tension with the velocity expectations of venture-backed growth.

There’s also the question of what “1650% growth” actually measures. Without public revenue figures or standardized metrics, the number likely reflects valuation growth from seed to current funding discussions. That’s impressive, but it’s a different kind of growth than recurring revenue or customer expansion. The company has been transparent about prioritizing speed over disclosure, which is typical for a startup at this stage but leaves outside observers with limited visibility into operational progress.

What’s undeniable is the strategic positioning. As AI’s text-data era reaches its limits, Periodic Labs has planted itself at the intersection of artificial intelligence and physical reality – a space that most AI companies have avoided because it’s hard, expensive, and slow. If the autonomous lab flywheel works as designed, the company won’t just be AI’s fastest-rising startup. It could redefine how humanity discovers the materials that power everything from smartphones to fusion reactors.

Sources