View source for Periodic Labs

{{Short description|AI company with autonomous laboratories for materials science}}
{{Infobox company
| name = Periodic Labs
| industry = Artificial intelligence, materials science
| founded = 2025
| hq_location = Menlo Park, California, United States
| products = Periodic Neon
| website = [https://periodic.com periodic.com]
}}
'''Periodic Labs''' is a company with laboratories in Menlo Park, California, that is building an "AI scientist" and the automated laboratories for it to work in. Its home page calls it "an AI research and deployment company creating models and autonomous labs to accelerate science", and its tagline is "From bits to atoms."<ref name="home" /> The company made its first public statement on September 30, 2025, under the title "Introducing Periodic Labs", and gave its goal as "to create an AI scientist."<ref name="intro" /> One of its stated goals is to find "superconductors that work at higher temperatures than today's materials", and a model it trained on data from its own laboratories is called Periodic Neon.<ref name="home" /><ref name="labs" />

[[Molecular Reality Corporation]], a nanopore company, names Periodic Labs third in its [[brain lab]] column, with the caption "AI-native biotech platform".<ref name="records" />

== Approach ==
The company began with the physical sciences, the introduction says, because experiments there "have high signal-to-noise and are (relatively) fast", "physical simulations effectively model many systems", and "physics is a verifiable environment". AI, it adds, has progressed fastest "in domains with data and verifiable results", such as math and code, and in physics "nature is the RL environment." In reinforcement learning the environment is what an agent acts on and is scored by. For a chess program it is the board. For Periodic it is a laboratory.<ref name="intro" />

"Science works by conjecturing how the world might be, running experiments, and learning from the results", the introduction says, and "New knowledge is created when ideas are found to be consistent with reality." Autonomous labs are "central to our strategy" because they "give our AI scientists the tools to act". They also "provide huge amounts of high-quality data" that "exists nowhere else", and "generate valuable negative results which are seldom published."<ref name="intro" />

== Laboratories ==
In a post of September 15, 2026, Periodic wrote that "we have built high-throughput labs in Menlo Park", and described "scaled-up high-throughput labs that run 24/7 with vastly enhanced throughput". The materials they are meant to discover are "superconductors and magnets". Each cycle starts with a hypothesis about what to make: "We first hypothesize what target to make, driven by predictions of stability and properties." AI is meant to sit inside the equipment too: it "embeds in our equipment, identifies issues in data, and learns the scientific process."<ref name="labs" />

Physical work scales more slowly than software. "The number of agents is not elastically scalable", the post says, because "increasing the number of concurrent experiments requires power, equipment, and engineering", "each experiment can take days", and "the results are often ambiguous". Where outside models "fall short or become too costly, we train our own." The post closes with the line "Our labs are learning how."<ref name="labs" />

== Periodic Neon ==
Periodic Neon is a model that interprets X-ray diffraction (XRD) data from the company's laboratories. In XRD a powdered sample is exposed to X-rays, and the angles and strengths of the scattered beams form a pattern that identifies the crystalline phases in the powder. A sample from a synthesis run often contains several phases at once. The post reports Neon's results on 134 samples from the company's labs that "human experts take hours to resolve".<ref name="nature" />

The company post-trained Neon from Kimi K2.6, an open-weight model, on "a multimodal blend of academic literature, code, and experimental data", and describes the result as a trillion-parameter model.<ref name="nature" /><ref name="labs" /> It set out to "outperform GPT-6 Astra and Claude Fable 5.1 at lower cost". On FrontierXRD, "our most difficult internal evaluation set", Neon "achieves a 55.3% success rate", a "20x improvement over the initial 2.7% success rate of Kimi K2.6". Its final training run "used 1,300 H200 GPUs, far less than Astra's reported 100,000+ Blackwell GPUs".<ref name="nature" /> The post also states a limit of its method: "Science is falsifiable, but not easily verifiable", and in the company's setup the success of an XRD analysis "cannot be cheaply verified" by pattern-fitting alone.<ref name="nature" />

== In Molecular Reality's usage ==
{{Main|Brain lab}}
Molecular Reality's materials cite Periodic Labs in support of their argument: "As Periodic Labs itself argues, AI now needs physical, experimental verification."<ref name="records" /> Periodic's introduction says "Here, nature is the RL environment." Molecular Reality's own proposal, [[RLMF]], takes the reward for an AI system from a nanopore measurement.<ref name="intro" /><ref name="records" /> Periodic Labs is the only one of the five companies in the brain column that does not appear on the list of eventual customers in those materials.<ref name="records" />

The chart's caption for Periodic Labs is "AI-native biotech platform". Periodic's own pages name superconductors and magnets.<ref name="records" /><ref name="labs" />

== See also ==
* [[Anthropic]]
* [[Brain lab]]
* [[Lila Sciences]]
* [[OpenAI]]
* [[Self-driving laboratory]]

== References ==
{{Reflist|refs=
<ref name="records">{{cite web |title=Company records |publisher=Molecular Reality Corporation |date=October 2026}}</ref>
<ref name="home">{{cite web |url=https://periodic.com/ |title=Periodic Labs |publisher=Periodic Labs |access-date=2026-10-25}}</ref>
<ref name="intro">{{cite web |url=https://periodic.com/news/introducing-periodic-labs |title=Introducing Periodic Labs |publisher=Periodic Labs |date=2025-09-30}}</ref>
<ref name="labs">{{cite web |url=https://periodic.com/news/building-labs-that-learn |title=Building Labs that Learn |publisher=Periodic Labs |date=2026-09-15}}</ref>
<ref name="nature">{{cite web |url=https://periodic.com/news/nature-is-our-learning-environment |title=Nature Is Our Learning Environment |publisher=Periodic Labs |date=2026-09-15}}</ref>
}}

{{Brain labs}}

[[Category:Artificial intelligence laboratories]]
[[Category:Materials science organizations]]
[[Category:Companies established in 2025]]