Periodic Labs funding: Every round, investor, and valuation (2026)

Periodic Labs is building an AI scientist for materials science. Its models suggest new materials, its robotic lab in Menlo Park, California, makes and tests them, and the results go back into training. The company's stated goal is a superconductor that works at higher temperatures than any known today.

Periodic Labs has raised $300 million in confirmed funding, all of it in one seed round announced in September 2025. A second round, led by Anjney Midha's AMP at a reported $7.5 billion valuation, has been reported since May 2026 but never announced by the company.

The founders explain most of the money. Liam Fedus helped create ChatGPT at OpenAI, and Ekin Dogus Cubuk led materials and chemistry research at Google DeepMind. Let's dive into Periodic Labs' funding history.

Periodic Labs went from incorporation to a billion-dollar valuation in a single round. Its second raise has played out through press reports rather than announcements, so only the seed is confirmed.

Fedus and Cubuk started Periodic Labs in May 2025, after both left the big AI labs. Fedus had been vice president of post-training research at OpenAI. Cubuk's team at Google built GNoME, a model that predicted the structure of millions of possible crystals.

The seed came together fast. TechCrunch reported that Peter Deng of Felicis committed over coffee before the company had a name, a bank account, or incorporation papers. Andreessen Horowitz then led the round, according to the deal note from the law firm Wilson Sonsini, which advised the company.

The company announced the round on September 30, 2025. DST Global, Accel, and Nvidia's venture arm NVentures joined, along with angels Jeff Bezos, Elad Gil, former Google CEO Eric Schmidt, and Google chief scientist Jeff Dean. OpenAI, Fedus's former employer, is not an investor, according to TechCrunch.

The $300 million paid for three things at once: a research team, a physical laboratory, and the computing power to train large models. A software AI startup needs only the first and the last. Periodic Labs could not, because its pitch depends on producing its own experimental data.

The launch post made that case directly. Its argument was that frontier models have already absorbed most of the text on the internet, so the next gains in science must come from new data generated in a lab. Superconductors were the headline target, with magnets and chip materials listed next to them.

Six months after the seed, Bloomberg reported that Periodic Labs was in talks to raise at a valuation of about $7 billion. Reports at the time added that the company had begun earning revenue from semiconductor customers, though it has not disclosed any figures.

In May 2026, Forbes reported that Periodic Labs was in advanced talks to raise at least $500 million at a $7.5 billion valuation. The lead was AMP, a public benefit corporation started by Anjney Midha, a former general partner at Andreessen Horowitz. Forbes also reported that the round was heavily oversubscribed and that talks had begun on a follow-on round at a higher price.

The round's status is still unclear. Forge Global lists a June 2026 Series A at a $7.5 billion valuation, and Contrary Research said in mid-2026 that the close had not been confirmed. Periodic Labs' own news page carries no funding announcement after September 2025, and neither the company nor AMP commented to Forbes.

Periodic Labs' reported valuation rose almost sixfold in about eight months. None of that rise came from a published discovery or a disclosed revenue figure.

The seed valued the company at roughly four times the money it raised, a ratio that implies a pre-money valuation of about $1 billion for a team with no product. Bloomberg and Forbes reported the figure. The company has never confirmed it.

The jump to $7.5 billion puts Periodic Labs among the research labs priced on talent and plans alone, a pattern visible in AI's share of new unicorns. A close comparison is Physical Intelligence's robot-model funding, another lab founded by former Google researchers that reached a multibillion-dollar price before selling a product at scale.

The two seeds show how far prices moved in about 18 months. Physical Intelligence raised $70 million in its 2024 seed at a reported valuation of about $400 million. Periodic Labs' seed was more than four times larger and priced at more than three times the valuation.

The price also sets a high bar for the next round. At $7.5 billion, Periodic Labs would be valued at 25 times its confirmed funding, against a business that still describes its core goal as a scientific breakthrough.

Periodic Labs' investors fall into three groups: the venture firms that wrote the seed, the AI insiders who invested personally, and AMP, which leads the reported second round.

Andreessen Horowitz led the seed. The pitch it backed is that the next gains in AI for science will come from new experimental data rather than more internet text. It is also the firm Midha left to start AMP, so the leads of both rounds share a history.

Felicis wrote the first commitment through partner Peter Deng, a former OpenAI product leader. TechCrunch reported that his quick yes started a scramble among other firms to get into the round.

DST Global and Accel joined the seed as growth and early-stage investors. NVentures, Nvidia's venture arm, adds a supplier to the cap table. Periodic Labs trains its models on Nvidia GPUs, so each dollar it spends on compute flows back to Nvidia's data center chip business.

No stakes, board seats, or strategic terms have been disclosed for any of these firms.

The angel list is short and senior. Jeff Bezos also backed Physical Intelligence, and his personal checks in AI labs sit apart from Bezos's stake in Amazon, which runs a competing cloud and AI business.

Eric Schmidt and Jeff Dean add weight from Google, where Cubuk did his materials work. Dean still runs research at the company whose DeepMind unit is now a direct competitor in AI for materials. Elad Gil, a prolific angel investor in AI startups, completes the group.

AMP is the reported lead of the 2026 round. Midha founded it as a public benefit corporation after leaving Andreessen Horowitz, and its AMP Foundry arm co-designs and co-funds new research labs and gives them access to pooled computing power.

Coverage of AMP's launch in May 2026 named Periodic Labs as the flagship Foundry company. That makes AMP more than a financial lead: it is a partner that supplies compute and helps shape the lab. Other investors in the reported round have been named only by sources that cite no evidence.

Some outlets list more names than the company does. The Logic reported that Radical Ventures, a Toronto AI fund, and Khosla Ventures took part in the seed. A June 2026 report from Caproasia listed Coatue and Lightspeed among the backers but gave no source.

None of these four appears on Periodic Labs' own investor list or in the Wilson Sonsini deal note. Until the company announces its next round, they are best treated as reported rather than confirmed.

Periodic Labs spends its funding on three things: physical labs, computing power, and researchers. The labs are what separate it from other AI startups.

Periodic Labs runs a lab of about 30,000 square feet in Menlo Park, according to Contrary Research. It handles powder synthesis, X-ray diffraction, and physics simulations around the clock. The lab is built so that robots mix and heat candidate materials, instruments measure what came out, and the results become training data.

That data is the core of the business case. Published science mostly reports successes, while Periodic Labs keeps its failed experiments as well. Those negative results are rare in public datasets and hard for competitors to copy.

The lab is also the biggest risk to the timeline. Many materials that models predict cannot be made in practice, and building reliable robotic synthesis takes years of engineering that software labs never face.

In September 2026, Periodic Labs released Periodic Neon, a model with about 1 trillion parameters. The company mid-trained it on scientific material and then used reinforcement learning on results from its own lab.

Training Neon used up to about 1,300 Nvidia H200 GPUs at peak, run through CoreWeave's cluster software. The company says it keeps more than 95% of its cluster busy by sharing capacity between training and simulation. The rest of the stack is open source, including Megatron for training, SGLang for inference, and Ray for distributed jobs. It has not disclosed its compute bill, which is far smaller than what the frontier labs spend, as OpenAI's funding history shows.

Talent was the first use of the seed. More than 20 researchers joined from OpenAI, Google DeepMind, and Meta, and the team's past work includes ChatGPT, GNoME, OpenAI's Operator agent, and Microsoft's MatterGen. Contrary Research counted about 64 employees in June 2026.

Hires span both sides of the business. The company recruited machine learning researchers alongside experimental physicists such as Eric Toberer, a superconductivity specialist, because the AI is only as good as the experiments that feed it.

Periodic Labs' commercial model is to sell an intelligence layer to companies in semiconductors, advanced manufacturing, and other physical sciences. It trains custom models and agents on a customer's own experimental data.

Its first public example was an unnamed chipmaker struggling with heat dissipation, which uses Periodic Labs' agents to help its engineers read experimental results. The company has named no customers and disclosed no revenue. That model looks closer to an enterprise services contract than to ChatGPT's subscription business, which Fedus helped launch.

The two halves of the business feed each other. Customer projects bring in industrial problems and data the company could not generate alone, and the lab's results make the models more useful to the next customer. The trade-off is that custom work for each client grows more slowly than selling one product to many.

Periodic Labs' growth focus is the semiconductor industry and advanced manufacturers, where better materials translate into faster chips and lower costs. The reported $7.5 billion round would pay for more labs, larger training runs, and a bigger team, as job listings for X-ray diffraction and equipment roles suggest.

The product roadmap runs from analysis to autonomy. Neon already reads diffraction data in the company's labs, and the next step is a system that designs experiments, picks synthesis recipes, and runs whole research campaigns. The long-term targets are superconductors, magnets, and materials for fusion and space.

The company has no reported IPO plans, and the first job is to confirm the AMP round. The main risk is scientific: Periodic Labs has not yet published a new material, and earlier AI materials claims in the field have faced public challenges. Competition runs from startups such as Lila Sciences, Radical AI, and CuspAI to Google DeepMind and Microsoft, and the Alphabet ownership structure behind DeepMind gives that rival access to search profits no startup can match.

Periodic Labs has raised $300 million in confirmed funding, all from a seed round led by Andreessen Horowitz and announced in September 2025. Forbes reported in May 2026 that the company was raising at least $500 million more in a round led by AMP, but the company has not confirmed that it closed.

Periodic Labs' investors include Andreessen Horowitz, Felicis, DST Global, Accel, and NVentures, Nvidia's venture arm. Angel investors include Jeff Bezos, Elad Gil, Eric Schmidt, and Jeff Dean. AMP, the firm founded by former Andreessen Horowitz partner Anjney Midha, is the reported lead of its 2026 round.

Periodic Labs was valued at $7.5 billion in its 2026 round, according to Forbes. That figure has not been confirmed by the company. Its seed round in September 2025 valued it at about $1.3 billion post-money.

No. Periodic Labs is private and has not filed for an IPO. Its shares are not traded on any stock exchange.

Periodic Labs is owned by its co-founders Liam Fedus and Ekin Dogus Cubuk, its employees, and its investors. Andreessen Horowitz, which led the seed, is likely among the largest outside holders, and AMP would join them if the reported round closed. Exact stakes are not public.