Periodic Labs

Investment memo — hiring, valuation, commercialization, technical milestones & competitive landscape

Data scraped March 25, 2026 · Research updated March 26, 2026 · Sources: Bloomberg, TechCrunch, a16z, DOE, Cognitive Revolution, Ashby, LinkedIn, X

Built with Claude Code + Codex

The Thesis

Periodic Labs is building an autonomous materials discovery loop. AI models predict novel materials, robotic labs synthesize them, instruments characterize results, and the data trains better models. Their hiring maps directly onto each stage of this pipeline.

The Discovery Pipeline

The 18 roles cover the full loop, and a few of them span multiple stages. The counts below are overlapping stage touches rather than a strict partition of the open roles.

Stage 1

AI Predicts

3 roles

→

Stage 2

Lab Builds

5 roles

→

Stage 3

Robots Run

4 roles

→

Stage 4

Instruments Measure

3 roles

→

Stage 5

Model Learns

5 roles

Research Domains

Six distinct domains emerge from the listings, revealing exactly where Periodic Labs is placing bets.

Thin-Film Materials & Novel Synthesis

5 roles · heaviest investment

This is the core science bet. They're hiring across the full thin-film lifecycle: a senior scientist to lead PVD synthesis (sputtering, PLD, MBE), a research associate to run deposition and nanofabrication, a characterization scientist to measure properties, a condensed matter theorist to model quantum materials, and a simulation scientist for multiphysics digital twins.

Signal: The emphasis on thin films, PVD, and novel compounds — combined with superconductivity and magnetism in the theory role — strongly suggests they're searching for new superconductors or quantum materials. The condensed matter theory role explicitly lists superconductivity and magnetism as bonus qualifications. The thin-film roles mention "novel materials predicted by AI models" and "previously unrealized compounds."

Research Scientist, Thin Films

Research Associate, Thin Films

Research Scientist, Materials Characterization

Research Scientist, Condensed Matter Theory

Multiphysics Simulation Scientist

Autonomous Lab Infrastructure

4 roles

They're building a fully automated physical lab from scratch — not just buying equipment, but designing the mechanical systems, control architectures, and robotic workcells. The automation engineer links instruments to data systems, the controls engineer architects PLC/SCADA, the mechanical engineer designs fixtures and enclosures, and the systems engineer manages vendors and installation.

Signal: The level of in-house engineering here is unusual. Most materials labs outsource automation — Periodic is building it themselves, suggesting they need tighter integration between the lab and AI systems than off-the-shelf solutions allow. The MES role (below) confirms this: they need custom scheduling and data provenance across dozens of instruments.

Frontier LLM Training for Science

3 roles

Not just fine-tuning — they're training frontier-scale models from scratch on scientific data. Midtraining focuses on synthetic data generation, distillation, and continual learning at trillion-token scale. Posttraining builds RL environments where models learn to design real experiments. The distributed training engineer runs the 5,000+ GPU clusters underneath.

Signal: The posttraining role description is remarkable: models that "generate hypotheses, design experiments that run in an actual lab, operate sophisticated scientific equipment." This isn't a copilot — they're building an AI scientist that runs physical experiments autonomously. The RL reward signals come from real lab results.

Research Engineer, Midtraining

Research Engineer, Posttraining

Distributed Training Engineer

Lab Software & Data Systems

2 roles

The software backbone: a Manufacturing Execution System (MES) managing scheduling, workflow orchestration, and data provenance across dozens of instruments (furnaces, dispensers, diffractometers). Plus a research engineer who owns the bridge between scientists' experimental designs and the automation platform.

Signal: The MES role mentions event sourcing, DAG orchestration, and resource locking — this is serious systems engineering applied to lab automation, not a simple LIMS. They need every experiment to have full lineage tracking so the AI can learn from every result.

Software Engineer (MES)

Research Engineer, Lab Automation

Supercompute & Infrastructure

3 roles

The compute layer: managing 5,000+ GPU clusters with Kubernetes/SLURM, building high-performance LLM inference (vLLM, SGLang, TensorRT-LLM), and securing the whole stack end-to-end. The inference engineer role specifically mentions integrating inference into RL workloads — the models need real-time access during training loops.

Signal: 5,000+ GPUs is large for a materials science company but NOT frontier AI lab scale — Anthropic, OpenAI, and DeepMind operate 50,000–100,000+ GPU clusters and spend billions annually on compute. Periodic's scale is more comparable to well-funded startups like Mistral or Character AI. With NVIDIA (NVentures) as an investor and Manik Singhal hired from Crusoe (GPU cloud), they likely have favorable pricing or rental arrangements rather than owning the hardware. The $300M raise at $7B valuation is significant but can't fund true frontier-scale compute — they're betting they can get more scientific signal per FLOP than general-purpose labs.

Supercompute Infrastructure Engineer

LLM Inference Engineer

Lead Security & IT Engineer

Product & Scientist Tooling

1 role · earliest stage

The newest domain — described as "one of the first product engineering hires." Building chat interfaces, dashboards, and experiment planning tools that put frontier models in the hands of bench scientists. Also responsible for telemetry and metrics that feed back into model training reward functions.

Signal: This single hire signals a transition from pure research to usable tools. The mention of "internal and customer settings" suggests they may be planning to commercialize their models or platform, not just use them in-house.

Product Engineer

Headcount by Domain

Thin Films & Synthesis

Autonomous Lab Infra

Frontier LLM Training

Supercompute & Infra

Lab Software & MES

Product & Tooling

What This Tells Us

They're hunting for new superconductors and quantum materials. The team confirms this: Minyong Han (MIT PhD MBE intermetallic thin films, Stanford PLD complex oxides), Grace Pan (Harvard PhD, 3D quantum materials ultra-thin limit), Killian Sheriff (MIT PhD, metallic alloys), Dominik Kufel (Harvard, "AI for Quantum Materials"), Christopher Rom (NREL, nitride thin films), Eric Toberer (thermoelectric materials professor). The job listings mention superconductivity and magnetism as qualifications.
The team was pre-built before the announcement. 15 of the 39 current employees joined before the Sep 2025 public launch, including 7 in May 2025 alone. The stealth phase ran Apr–Aug 2025 with a fully operational early team in place before anyone knew the company existed.
They pulled from exactly 3 talent pools. Google/DeepMind (6 people including both co-founders), OpenAI (4 people — Fedus literally brought Pandey, who "followed our VP of Post-training"), and national labs + Tesla for the physical science side. Meta/FAIR contributed 3 more.
The RL loop is physical, not simulated. Costa Huang (CleanRL creator, "exploiting physical rewards"), Vincent Moens (TorchRL creator, "building RL infra"), Rishabh Agarwal (NeurIPS best paper for offline RL). Three of the world's top RL practitioners, all working on the same thing: reward signals from real lab experiments.
Dzmitry Bahdanau is a sleeper hire. The co-inventor of the neural attention mechanism — the foundation of every transformer and LLM — is listed simply as "Team member." His LinkedIn bio: "Move bits to construct new things from atoms." This is arguably the most important hire in the company after the co-founders.
Muratahan Aykol is a key early hire. Google DeepMind Staff Research Scientist, previously at Toyota Research Institute and Rivian. Joined May 2025 during the stealth phase. His headline: "AI for Science." His materials-AI background at DeepMind directly complements Cubuk's GNoME work.
The lab is real, not theoretical. Naveen Menon ran cathode manufacturing pilot lines at Tesla. Sam Cross was Sr Principal Engineer Lab Automation at Lila Sciences and Samsung. Nicholas Bergantz has 20 years in robotics. Jun Feng was Director of Process Engineering at Applied Materials running PECVD semiconductor lines. They hired people who have actually built and operated physical labs at scale.
Stephan Hoyer (Mar 2026) suggests a climate/weather angle. The most recent hire was Google's Senior Staff SWE who led AI weather/climate models and created Xarray. This hints at expanding beyond materials science into Earth systems modeling.

The Team (LinkedIn Deep Dive)

39 current employees identified. All 39 LinkedIn profiles fully scraped with complete career timelines. 1 known departure (Michael Zhang).

Current headcount

39 employees

Funding

$300M Seed (a16z led) · ~$7B new round in discussion (Bloomberg, Mar 2026)

Previous employers

Google/DeepMind (6), OpenAI (4), Meta/FAIR (3), Tesla (3), Microsoft Research (1), xAI (1), SpaceX (1)

PhDs on team

MIT (5), Harvard (3), Stanford (1), Cambridge (2), Northwestern (2), Colorado State (1), MILA (2), UCSB (1)

Key investors

a16z, Felicis, DST Global, NVentures (NVIDIA), Accel, Coatue, Lightspeed, Jeff Bezos, Eric Schmidt, Jeff Dean

Advisory board

Carolyn Bertozzi (Nobel laureate), Mercouri Kanatzidis, Steven Kivelson, Zhi-Xun Shen, Chris Wolverton. Advisor: Tal Broda (ex-VP Compute, OpenAI)

Hiring Sequence — Exact Dates from LinkedIn

All 39 start dates confirmed from LinkedIn profile scraping. Company incorporated ~Apr 2025, announced Sep 2025.

Apr 2025

Incorporation + first 2 hires

Emirhan Kurtulus (HyperbeeAI, Stanford NLP) and Wei Chen (TikTok Tech Lead, Twitter, Brookhaven National Lab) join as employees #1 and #2. Both co-founders have left OpenAI and Google by March.

May 2025

Stealth team of 8 assembles

Alexandre Passos (scikit-learn, Google 9yr, OpenAI), Costa Huang (CleanRL, AI2, Hugging Face), Matthew Horton (Microsoft Research MatterGen, Berkeley Lab Materials Project), Xiang Fu (Meta FAIR, MIT PhD), Eric Toberer (Colorado School of Mines professor), Naveen Menon (Tesla 5yr, cathode materials), Elsa Cong (EA, ex-Visa). Also: Muratahan Aykol (DeepMind Staff Research Scientist, Toyota Research Institute, Rivian).

Jun 2025

First interns + key AI hires

Killian Sheriff — "Intern No. 1, joined two weeks after incorporation." MIT PhD Materials Science. Dominik Kufel — "Intern No. 2," Harvard, AI for Quantum Materials. Rohan Pandey — OpenAI MTS ("followed our VP of Post-training onto the founding team"). Reiichiro Nakano — OpenAI 5yr. Vincent Moens — Meta 4yr, TorchRL creator, founding team, building RL infra.

Jul 2025

Google researcher joins

Gowoon Cheon — Google Research Scientist 2yr + SWE 2yr, Stanford PhD Applied Physics, KAIST. Crystal structure search with GCNs.

Sep 2025 — Public launch

$300M announced, 10 more hires

Co-founders Fedus (OpenAI VP, ChatGPT) and Cubuk (DeepMind, GNoME) officially announce. Joined by: Kate Lauterbach (DeepMind Program Lead → Google 10yr, founding team), Sam Cross (Lila Sciences, Samsung, MIT PhD — founding team), Dzmitry Bahdanau (neural attention co-inventor, ServiceNow 5yr), Rishabh Agarwal (DeepMind 7yr, NeurIPS best paper), Xander Dunn (Jasnah, Ava Labs, Apple), Janosh Riebesell (Radical AI, Cambridge PhD), Aryan Suri (Tesla 3yr), Peiwen Ren (intern, Northwestern/QuesTek).

Oct–Dec 2025

Materials science wave

Daniel Chica (Oct — Columbia postdoc, Northwestern PhD Chemistry), Christopher Rom (Oct — NREL 3yr, nitride thin films), Christie Koay (Dec — Princeton postdoc, Columbia PhD), Minyong Han (Dec — Stanford postdoc PLD, MIT PhD MBE), Hilary Johnson (Dec — Lawrence Livermore 3.5yr, MIT PhD Precision Machine Design). Killian Sheriff promoted from intern to MTS.

Jan–Mar 2026

Continued scaling — hardware, semiconductor, newest hires

Nicholas Bergantz (Jan — robotics leader 20yr), Manik Singhal (Jan — Crusoe, d-Matrix GPU architect), Jordan S. (Jan — OpenAI robotics tech, Tesla), Dennis van der Staay (2026 — Meta 6yr, MIT EE), Mia Johansson (Feb — SpaceX 3yr), Jun Feng (Feb — Applied Materials 7yr Director, semiconductor PECVD), Florian Göltl (Feb — xAI 1yr, U of Arizona professor), Peiwen Ren promoted intern→MTS, Grace Pan (Jan — visiting scientist, UC Berkeley, Harvard PhD Physics). Most recent: Stephan Hoyer (Mar 2026 — Google 10yr Senior Staff SWE, Xarray creator).

Where They're Recruiting From

Exact previous employers from 39 scraped LinkedIn profiles. Two distinct talent pools that rarely overlap.

AI / ML / Software (22 people)

Google / DeepMind (6) — Cubuk (co-founder, 8yr), Lauterbach (Program Lead), Cheon (Research Scientist), Agarwal (Staff Research Scientist via Brain), Kurtulus (intern under Cubuk), Hoyer (10yr Sr Staff SWE)
OpenAI (4) — Fedus (co-founder, VP Post-Training), Passos (MTS), Nakano (5yr MTS), Pandey (MTS Resident, "followed our VP")
Meta / FAIR (3) — Moens (4yr, TorchRL), Fu (FAIR Research Scientist), van der Staay (6yr Research Scientist)
xAI (1) — Göltl (STEM Data Team Lead)
AI2 + Hugging Face (1) — Costa Huang (CleanRL creator)
SpaceX (1) — Johansson (SWE 3yr)
TikTok + Twitter (1) — Wei Chen (Tech Lead Ads ML)
ServiceNow Research (1) — Bahdanau (Research Lead 5yr, attention mechanism inventor)
Other AI (4) — Dunn (Jasnah/code gen), Riebesell (Radical AI), Singhal (Crusoe/d-Matrix GPU), Bergantz (robotics 20yr)

Materials Science / Hardware (14 people)

Tesla (3) — Menon (Staff Production Engineer, cathode materials), Suri (Data Engineer), Jordan S. (technician)
National Labs (3) — Horton (Berkeley Lab Materials Project 7yr), Rom (NREL 3yr, nitride thin films), Johnson (Lawrence Livermore 3.5yr)
Microsoft Research (1) — Horton (MatterGen, MatterSim)
Applied Materials (1) — Feng (Director Process Engineering, semiconductor PECVD 7yr)
Samsung (1) — Cross (Senior Staff Engineer)
Postdocs (4) — Koay (Princeton/Columbia), Han (Stanford/MIT), Chica (Columbia/Penn State), Pan (UC Berkeley/Harvard)
Professors (1) — Toberer (Colorado School of Mines, 15yr)
Key signal: The materials team can actually BUILD a lab and RUN experiments — not just model them. Tesla battery experience, national lab thin-film work, semiconductor process engineering.

Previous Employers — Visualized

Curated from the same 39 scraped LinkedIn profiles, using the primary prior-employer buckets referenced in the narrative above.

AI / ML / Software

Materials / Hardware / Science

Hires per Month

Full Team Roster — Ranked by Hire Order

39 current employees, all scraped from LinkedIn with full career timelines. Sorted by confirmed start date (Apr 2025 → Mar 2026). 1 known departure: Michael Zhang.

At a Glance

Open Roles

Atoms (Lab)

Bits (AI/Infra)

PhD Required

Remote OK

All 18 Roles

Financing & Valuation

Periodic Labs emerged from stealth on September 30, 2025 with a $300M seed / founding round at a $1.3B valuation, led by a16z with Felicis cutting the first check. Six months later, Bloomberg reported (March 25, 2026) that Periodic is discussing a new raise at ~$7B valuation — implying ~5.4x value accretion in six months.

Seed / Founding Round — Sep 2025

$300M AT $1.3B

Lead: Andreessen Horowitz (a16z). First check: Felicis Ventures.
Institutional: DST Global, NVentures (Nvidia), Accel, General Catalyst, Radical Ventures, Coatue, Lightspeed.
Angels: Jeff Bezos, Eric Schmidt, Jeff Dean (Google Chief Scientist), Elad Gil.
One of the largest seed rounds in history. VCs engaged in a bidding war — one investor reportedly wrote a "love letter" to secure allocation. Natasha Mascarenhas at TechCrunch scooped the fundraise before it closed.

Reported New Round — Mar 2026

~$7B VALUATION

Bloomberg (Mar 25, 2026): Periodic is "in discussions with investors about raising at least hundreds of millions of dollars at a valuation of about $7 billion." Terms are still evolving. Bloomberg notes the company has "achieved early commercial traction with semiconductor industry customers and is already generating revenue."

Valuation risk: $1.3B → $7B in ~6 months implies ~5.4x step-up. For a company that is less than a year old with nascent revenue, this pushes valuation risk substantially higher. At $7B, Periodic needs to become a generational platform company — not just a well-funded research lab. The bar for commercial execution at this price is very high.

Commercialization — Hard Evidence

This is not "they may commercialize later." Multiple independent sources confirm Periodic is already deploying solutions with paying customers in semiconductor, space, and defense.

Semiconductor Customer (Named Use Case)

LIVE

Periodic's own website (periodic.com) states they are helping a semiconductor manufacturer address chip heat dissipation by training custom agents on experimental data. This is a concrete, named use case — not a future aspiration.

Space & Defense Customers

a16z CONFIRMED

a16z's investment thesis independently confirms: "Current customers span space, defense, and semiconductor sectors." Specific applications include "heat dissipation problem-solving and automation of simulations."

Revenue Signal

BLOOMBERG

Bloomberg/TFN (Mar 2026): The company "has achieved early commercial traction with semiconductor industry customers and is already generating revenue, distinguishing it from peers remaining in research phases."

Signal: Three independent sources (periodic.com, a16z, Bloomberg) confirm real customers and revenue. The Product Engineer role referencing "internal and customer settings" corroborates commercialization is underway, not hypothetical.

Business Model & Capital Intensity

Periodic should be treated as both an AI-infra company and a lab-capex company. The near-term business is contract/custom AI for physical R&D; the long-term play is licensing or direct economics from discoveries.

Near-Term: Contract AI for Physical R&D

GTM NOW

In a November 2025 Latitude Media interview, Cubuk described the near-term business as providing Periodic's models to other firms doing physical R&D — effectively custom AI + experimental data as a service. Land-and-expand across semiconductors, defense, space, and advanced manufacturing.

Long-Term: Licensing & Discovery Economics

FUTURE

Cubuk left open a longer-term path toward licensing models or capturing direct economics from discoveries. If Periodic discovers a commercially viable superconductor or novel semiconductor material, the IP value dwarfs the software revenue. This is the "blue sky" scenario that justifies the $7B valuation — but it requires actual breakthroughs.

Capital Structure: GPUs + Lab

DUAL BURN

Cubuk stated the raise size was driven by both GPU cost and lab cost, with GPUs potentially a large share. The company operates 5,000+ GPU clusters (job listings) plus a physical lab in Menlo Park with PVD chambers, cryogenics, robotic workcells, and dozens of instruments. With NVentures (Nvidia) as an investor and Manik Singhal hired from Crusoe (GPU cloud), they likely have favorable compute pricing.

Capital risk: Unlike pure-software AI companies, Periodic carries two heavy cost centers: GPU compute for frontier model training AND physical lab capex/opex for autonomous experiments. The $300M seed funds both; at $7B, the next round presumably extends this runway. But if the closed loop takes longer than expected to compound proprietary data, the dual burn rate becomes a serious concern.

DOE / Public-Sector Adjacency

In December 2025, the U.S. Department of Energy announced that Periodic Labs was one of 24 organizations collaborating on the Genesis Mission — a national effort to use AI to accelerate discovery science, strengthen national security, and drive energy innovation.

Genesis Mission — 24 Collaborators

DOE / WHITE HOUSE

The full list: Accenture, AMD, Anthropic, Armada, AWS, Cerebras, CoreWeave, Dell, DrivenData, Google, Groq, HPE, IBM, Intel, Microsoft, NVIDIA, OpenAI, Oracle, Periodic Labs, Palantir, Project Prometheus, Radical AI, xAI, XPRIZE.

Signal: Periodic is the only materials-science startup on a list otherwise composed of hyperscalers, chip companies, and foundation-model labs. This is a meaningful ecosystem signal: Periodic is already inside the emerging federal AI-for-science stack. The Genesis Mission aims to "double the productivity and impact of American science and engineering within a decade" using AI — Periodic's autonomous lab thesis fits this exactly. This does not prove revenue, but it signals federal credibility and potential government contract pipeline in defense, energy, and critical materials.

Critical Milestone: Is the Closed Loop Live?

The single most important technical question for a March 2026 assessment: has Periodic transitioned from assisted lab workflows to genuine robotic closed-loop operation?

October 2025: Lab Up, Robots Not Yet Running

KNOWN STATE

TechCrunch (Oct 20, 2025): "Periodic has already set up its lab, was working with experimental data and testing predictions, but the robots were not yet up and running." Cubuk said "they will take a bit to train."

March 2026: Key Diligence Question

UNKNOWN

Five months have passed. The January 2026 hiring burst — Nicholas Bergantz (20yr robotics), Jun Feng (Applied Materials semiconductor lines), Jordan S. (OpenAI robotics tech) — suggests active lab buildout. But whether the system has moved from human-assisted experiments to genuine autonomous synthesis → characterization → model feedback is not publicly confirmed.

Why this matters: The entire thesis depends on the closed loop compounding proprietary data faster than peers. If robots are running autonomously, the data flywheel is spinning and the moat is real. If the loop is still human-assisted, Periodic's data advantage over DeepMind/Microsoft is limited to what human scientists can manually produce — which is dramatically slower and doesn't justify the $7B valuation premium over competitors. This is the #1 diligence question.

Technical Validation & Risks

The underlying thesis — AI can predict materials and autonomous labs can synthesize them — is directionally validated by external benchmarks. But the risk case has also sharpened.

External Validation

BULL CASE

DeepMind GNoME (Nature, 2023): 2.2M new crystal structures predicted, including 380,000 stable candidates. Cubuk co-authored this work.

Berkeley A-Lab (Nature, 2023): 41 new inorganic compounds synthesized out of 58 attempts over 17 days — 71% success rate — fully autonomous.

Microsoft MatterGen: More than doubled the share of stable, unique, and novel materials vs prior generative models. Periodic hired Matt Horton, a creator of MatterGen.

These results support the plausibility of model-guided discovery + autonomous synthesis working in tandem.

The Risk Case

BEAR CASE

A-Lab critique: A ChemRxiv reanalysis (Leeman et al., 2024) concluded that A-Lab "failed to make new materials" — directly contradicting the Nature paper's claims. The experimental and computational methodology was flagged as having serious problems. This is the strongest published counterpoint to autonomous-lab optimism.

Data quality: Reviews in npj Computational Materials (2024) highlight persistent issues: noisy datasets, missing metadata, poor reproducibility, and lack of standardized formats across labs. Published literature has noise floors too high for effective training (per Fedus, Cognitive Revolution podcast).

Autonomy gap: A 2024 community survey found most discovery-focused labs sit at low-to-mid autonomy (L1–L2), with only a few entering L3. Automation often carries large upfront time-and-money costs. Moving from "AI suggests, human runs" to "AI runs everything" is a harder engineering problem than the model research alone.

The real question: Not whether the concept is interesting — it clearly is — but whether Periodic can build a clean enough, integrated enough, fast enough loop to compound proprietary data faster than peers. The data quality problems that plague the field are exactly what Periodic claims to solve with its own labs. If they can't produce cleaner data than published literature, the moat collapses.

Competitive Landscape

Over $1.3B+ has flowed into AI materials discovery startups in the past two years. The field has moved significantly since most earlier comparisons were drawn.

Lila Sciences

$550M RAISED · $1.3B VAL

$350M Series A (Oct 2025), total funding $550M, $1.3B+ valuation. Backed by Nvidia. Building "AI Science Factories" and already onboarding first customers in energy, semiconductors, and drug development. Scaling to a 235,500 sq ft facility in Cambridge, expanding to SF and London. Platform draws interest across scientific domains.

Key difference: Lila is broader ("scientific superintelligence" across all science, rooted in Flagship Pioneering's biotech DNA). Periodic is narrower (condensed matter physics, materials). Periodic has stronger ML pedigree (ChatGPT co-creator, attention mechanism inventor). Lila has more total capital and physical infrastructure. Both are racing to close the autonomous loop — neither has publicly demonstrated fully autonomous discovery-to-synthesis cycles at scale.

CuspAI

$100M SERIES A · $520M VAL

$100M Series A (Sep 2025) led by NEA and Temasek, at $520M valuation. Founded by Dr. Chad Edwards and Prof. Max Welling (Qualcomm VP, attention mechanism co-contributor). Positioning as a "search engine for the material world" — synthesis-aware foundation models for materials design. Claims: customer request → novel material meeting specs in 6 months (vs ~decade traditionally), targeting 1–2 months within 2 years. Partners: Hyundai Motor Group, Kemira (PFAS filtration). NVentures and Samsung Ventures also invested.

Key difference: CuspAI focuses on computational design + customer specs without building its own physical labs. Periodic builds the full stack (models + labs). CuspAI's speed-to-customer-specs claim is a strong commercial signal but depends on partner labs for synthesis. If Periodic's autonomous labs work, they have a tighter feedback loop.

Orbital Materials

$21M · PIVOTING DOWNSTREAM

No longer a pure discovery-platform comp. Orbital's current site emphasizes AI-designed data-center hardware: modular data centers, direct-to-chip liquid cooling, and carbon capture. Multi-year AWS partnership for data center decarbonization. Open-source 'Orb' AI model available on AWS. 10x improvement in material performance since early 2024.

Key difference: Orbital has pivoted from "discovery platform" to "AI-designed products" — competing downstream on manufactured goods rather than upstream on discovery tools. More comparable to an advanced hardware company than to Periodic's research-platform model. The AWS partnership validates the materials-to-product pipeline but is a different business.

Citrine Informatics

$74M · INCUMBENT

The OG — materials informatics platform since 2013. Series B ($20M) from Prelude Ventures, Innovation Endeavors, Next47. Established customer base in chemicals and manufacturing. Data/analytics SaaS, not autonomous lab.

Big Lab Research

INCUMBENTS

Google DeepMind GNoME — 2.2M new crystal structures (Nature, 2023). Cubuk co-authored this work before leaving.
Microsoft Research MatterGen/MatterSim — generative models for inorganic materials. Periodic hired Matt Horton, a creator of these tools.
Berkeley A-Lab — 41/58 autonomous synthesis (71% success), though results disputed by Leeman et al. reanalysis.

Talent consolidation: Periodic is systematically hiring FROM these incumbents. GNoME co-author, MatterGen creator, national lab researchers. They're concentrating the talent that built the field's foundational tools into one company. This is the strongest bull signal in the competitive map — but it also means DeepMind/Microsoft lose key contributors and may deprioritize materials work, reducing the external research ecosystem that Periodic also benefits from.

Credibility & Data Hygiene Warning

Periodic maintains strong signals of active frontier research rather than operating as a closed commercial black box. But public data has known staleness issues.

Research Credibility

POSITIVE

Scientific Advisory Board: 5 professors publicly listed on periodic.com — including ZX Shen (Stanford, ARPES/superconductivity), Mercouri Kanatzidis (Northwestern, materials chemistry), Carolyn Bertozzi (Nobel laureate), Steven Kivelson, Chris Wolverton. Advisor: Tal Broda (ex-VP Compute, OpenAI).

Academic Grant Program: Funding complementary university research — a signal of ecosystem building, not isolation.

arXiv publications: Recent papers show Periodic Labs affiliations for Cubuk, Bahdanau, and Agarwal — the company is still publishing, not going dark. This supports the view that they're embedded in active frontier research.

Data Hygiene Warning

CAUTION

Funding figures vary across sources: some report "$200M at $1B" (earlier reporting), others "$300M at $1.3B" (final close). Always prefer the most recent, most specific source — in this case, the Bloomberg March 2026 reporting and a16z's own announcement. Public team pages and LinkedIn profiles can also lag — always cross-check headcount data against multiple sources.

For this memo: Our headcount of 39 and departure tracking (Zhang) comes from direct LinkedIn profile scraping on March 25, 2026 — not from Periodic's own site. Treat our scrape as more current than their public page, but note that LinkedIn itself can lag by weeks.

Market Size & Opportunity

The materials informatics market is relatively small today but growing fast. The real opportunity is the industries that better materials unlock — semiconductors, energy, aerospace, defense — representing ~$15 trillion in global GDP according to a16z.

Materials Informatics (Narrow TAM)

SOFTWARE

$170M (2025) → $410M (2030) at 19.2% CAGR per MarketsandMarkets. Precedence Research estimates reaching $1.14B by 2034 at 20.8% CAGR. This is the software/platform slice — tools like Citrine, not the end-market value.

AI in Chemical & Materials (Broad TAM)

PLATFORM

$17B (2025) → $195B (2032) at 41.6% CAGR per 360iResearch. This broader definition includes AI-augmented R&D across chemicals, pharma, and advanced manufacturing.

End-Market Industries (Total Value)

$15T+

Advanced manufacturing, semiconductors, energy, aerospace, defense. If Periodic's AI can compress materials R&D from decades to months, even capturing a fraction of this value represents a massive outcome. McKinsey estimates generative AI could unlock $100B+ in value for materials-intensive industries.

The tension: Citrine's founder noted: "If you've raised a quarter of a billion dollars, you need a $100 billion outcome." At $7B valuation, the bar is even higher — Periodic needs to capture material-discovery economics, not just software margins, to justify the price.

Public Presence & Media

Official Channels

ONLINE

Website: periodic.com
Twitter/X: @periodiclabs
LinkedIn: Periodic Labs
Careers: Ashby Job Board

Founder Key Quotes

INTERVIEWS

"Intelligence is necessary, but not sufficient. New knowledge is created when ideas are found to be consistent with reality." — Fedus, launch

"Nature is our RL environment." — Fedus, Cognitive Revolution podcast

"Experiments are extremely important for solid-state physics; theory and simulations are not enough." — Cubuk, Physics Today

"LLM researchers and physicists don't usually work together, and that was crucial for our aims." — Cubuk on why he left Google

Major Press Coverage

MEDIA

Bloomberg (Mar 2026) — "AI Science Startup Periodic Labs Is in Deal Talks at About $7 Billion"
TechCrunch — "$300M seed to automate science"
TechCrunch — "$300M VC frenzy"
a16z — Investor thesis
MIT Tech Review — "AI materials discovery now needs to move into the real world"
Latitude Media — "Inside a $300M bet on AI for physical R&D"
Physics Today — Cubuk profile: "accelerate physics R&D using AI"
Bloomberg (Jun 2025) — Pre-launch: "OpenAI staffers' new startups"

Podcasts

DEEP DIVES

Cognitive Revolution — "Training an AI Scientist with Feedback from Reality" (both founders)
The a16z Show — "Building an AI Physicist"
Latitude Media "Catalyst" — Cubuk on business model and capital allocation (Nov 2025)

Key Tweets

X / TWITTER

@LiamFedus — Launch: "Our goal is to create an AI scientist..."
@a16z — "We're leading a $300M founding round..."
@agarwl_ — Company name origin: "Copper Age → Bronze Age → Iron Age → Silicon Age"
@nmasc_ — Pre-launch scoop: "$1B+ valuation"
@AndrewCurran_ — March 2026 update