COMPANY PROFILE / MATERIALS INFORMATICS
Lila Sciences
Aiming at "scientific superintelligence" with AI and automated labs for materials, chemistry and life sciences: reading the track record apart from the slogans
Lila Sciences (written "LILA" on its website) is a privately held U.S. company that says it will take on the whole flow of AI forming hypotheses, automated labs running the experiments, and learning from the results. Its website calls its goal "Scientific Superintelligence", but that is a phrase expressing company philosophy, not a state that has been reached. The concrete result in materials that primary sources let us confirm is a search for catalysts for hydrogen production (September 2026). This profile uses only the company's official announcements and a preprint (a paper not yet peer reviewed), and keeps philosophy, claims and confirmed facts apart.
- Lila Sciences in 30 seconds, and where it stands in materials informatics
- The logo and how the name is written
- Where the company sits in the materials informatics field
- Scale
- Resources committed to materials informatics
- What it has done so far
- What comes next
- What the company is aiming for in materials informatics
- Map of materials informatics partnerships
- Technologies, products and services
- The risks this company carries
- Glossary, references and claim-to-source audit
1. Lila Sciences in 30 seconds, and where it stands in materials informatics
Cambridge (U.S.)
LondonThe website's contact page
Foundation models & compute (secondary)Our own classification Our calculation
Public in March 2025The company says it began inside Flagship Pioneering
2,942 evaluated in 3 monthsCompany's claim. A preprint exists (not peer reviewed)
Materials informatics (MI) is a way of searching for new materials by gathering data, using AI to predict and design candidates, and testing them by experiment. Lila Sciences wants one company to own all of it, and it applies the approach to science as a whole, including drugs and chemicals, not only materials Sourced. Its website uses big words: "scientific superintelligence" and "the world's most advanced operating system for science". This profile treats such phrases as the company's claims and philosophy, and reads them apart from the facts with measured values, counts and conditions (chapter 10). On its hydrogen-catalyst result the company itself says plainly that it is "not a finished technology and not a commercial product".
2. The logo and how the name is written
A note on names: the legal entity is Lila Sciences, Inc., and the website headings write it as "LILA" in capitals. The website's logo is white lettering spelling "L I L A" with wide spacing (it does not include the word "Sciences"), and because it is used on dark backgrounds it is also placed here on a dark panel. To avoid confusion with other companies and institutes with similar names, this article writes "Lila Sciences" throughout.
Official site: https://www.lila.ai/3. Where the company sits in the materials informatics field
First, the figure sorts MI companies into four types and shows where Lila Sciences sits. To keep the company's claim (scientific superintelligence) apart from the business that can actually be confirmed (linking AI with labs), the types are our own classification.
Next, the automated experiment loop that Lila Sciences actually ran in its hydrogen-catalyst search, based on the company's account. "PVD (physical vapour deposition)" is a way of making a thin film by evaporating a material onto a substrate.
4. Scale
The company is privately held, so there are no statutory financial filings. Only what can be confirmed is shown Sourced.
| Item | Detail | Source and type |
|---|---|---|
| Founding and launch | Founded in 2023. The company says it prepared inside Flagship Pioneering's labs for about three years and went public in March 2025 | The company's description (About page, Series A posts) |
| Locations | San Francisco, Cambridge (Massachusetts) and London | The website's contact page. The head office is not named |
| Employees | Not disclosed. Leaders' names and titles are on the team page | No total headcount in official material |
| Revenue | Not disclosed | Not stated in any official material. This article does not estimate it |
| Funding raised | US$200 million seed, US$350 million Series A, US$550 million in total (October 2025) | Company announcements. No valuation is given in official material and this article does not discuss one |
| Listing | Privately held | No share information on the website |
The Series A close post carries the date 10 October 2025, but another post on the same website says it was announced on 15 October 2025. Official material does not settle which is right Not yet confirmed. The website's news list also links to outside articles that report a valuation, but this article does not use press-only numbers. The seed post likewise shows 3 October 2025 on the site, while its text says the company went public in March 2025, so the dates do not agree.
5. Resources committed to materials informatics
Lila Sciences spans materials, chemistry and life sciences, and the resources spent on materials alone are not disclosed. What follows is what could be confirmed for the company as a whole, under four headings: people, money, outside investments, acquisitions and joint ventures, and ties to universities and public bodies.
| Item | What could be confirmed | Status |
|---|---|---|
| People | Leadership includes CEO Geoffrey von Maltzahn, CTO Andrew Beam and Rafael Gómez-Bombarelli, CSO for physical sciences (also introduced in company posts as an MIT associate professor). John Gregoire is described in the September 2026 post as chief autonomous science officer. The total headcount and the size of the materials group are not disclosed | Sourced (headcount: not disclosed) |
| Money (raised) | US$200 million seed (Flagship Pioneering and others); US$350 million Series A (co-led by Braidwell and Collective Global, with NVentures, Analog Devices and others). US$550 million in total | Sourced (company announcements) |
| Money (facilities and computing) | The company says it will use funds to build more AI Science Factories. Capital spending and the scale of computing are not disclosed | Could not be confirmed |
| Outside investments, acquisitions, joint ventures, divestments | None stated in official material. (The Creation offering is said to include launching new companies, but no actual case could be confirmed) | Could not be confirmed |
| Universities, research institutes and public funding | Three projects were selected for Phase I of the U.S. Department of Energy (DOE) Genesis Mission (July 2026), with Caltech, Lawrence Berkeley National Laboratory (LBNL), Northwestern University, Argonne National Laboratory and the National Laboratory of the Rockies (NLR) | Sourced (Phase II is Not yet confirmed) |
Status of the main models, products and facilities
The table lists name, type, specification, form of release, stage and date, as far as official material allows. Numbers measured under different conditions are not ranked against each other.
| Name | Type | Main specifications (including company claims) | Release | Stage and date |
|---|---|---|---|---|
| Lila Iris | Scientific reasoning model | In the company's account it combines Bayesian models and language models to balance uncertainty against information gain (hydrogen-catalyst post). Performance metrics and training data: could not be confirmed. The site says it "consistently outperforms other models", with no data or conditions given | Not public. Said to be available through Catalyst | Offered (company description). October 2026 |
| AI Science Factory (AISF) | Automated lab integrating AI and instruments | For the hydrogen-catalyst search: a closed loop of synthesis, pre-test characterisation, testing and post-test characterisation. 96 samples made in parallel by PVD. Up to 100 samples a day, about 240 a week in practice. Automation above 90% (preprint) | Operated in house. Access offered through Catalyst and others | In operation. Full automation is under development |
| Catalyst | Form of offering | Direct access to Lila Iris and the AISF, plus expert support. Described as "lab as a service", used on demand. No price given | Commercial (by contract) | Offered (company description) |
| Creation | Form of offering | Lila runs the whole discovery campaign to a client's targets and delivers validated structures, protocols and data. Said to include launching new companies | Commercial (by contract) | Offered (company description) |
| OER catalyst search result | Experimental result (materials) | 2,942 catalysts evaluated across 53 systems and 26 elements. Six palladium-based families. The lead candidate has run over 1,000 hours in acid (details in chapter 10) | Company post and preprint | Announced. 24-25 September 2026 |
In an electrolyser that makes hydrogen from water, the anode runs the OER (oxygen evolution reaction). It proceeds in a strongly acidic, high-potential environment that dissolves most metals. Today's commercial machines therefore use iridium oxide, which is both active and durable, and iridium is a rare element produced in very small amounts Sourced. The difficulty of the search is a trade-off: the more active a material, the more easily it tends to break down, so both must be measured together before a material can be judged good or bad. In its preprint Lila Sciences takes, as its measure of durability, how long the overpotential (the extra voltage needed beyond the ideal) stays below 0.5 V when 10 mA/cm² flows in 1 M sulfuric acid. But this condition is a test on a small thin film, and real electrolysers are generally run at much higher current densities (our own note). The company itself says it will evaluate the material in a form closer to real use.
6. What it has done so far
| When | What happened | Meaning |
|---|---|---|
| 2023 | Founded (About page) | The company says it prepared inside Flagship Pioneering |
| From late 2024 | Began building the AISF and the automated electrocatalysis workflow | From the hydrogen-catalyst post |
| March 2025 | Went public (emerged from stealth). Announced a US$200 million seed round | The post is dated 3 October 2025 on the site, which disagrees |
| 15 September 2025 | Announced a US$235 million Series A | Co-led by Braidwell and Collective Global |
| October 2025 | Series A closed at US$350 million in all; total raised US$550 million | The date is given as the 10th and the 15th within the same website |
| 22 July 2026 | Three projects selected for Phase I of the DOE Genesis Mission | Catalysis, polyethylene production and semiconductor charge transport. Phase II is a future application |
| 24 September 2026 | Posted the OER catalyst preprint to arXiv | Not peer reviewed. The breakdown of measured values can be checked |
| 25 September 2026 | Published the hydrogen-catalyst search result as a company post | 2,942 catalysts evaluated in three months; six families identified |
7. What comes next
Plastics precursorsThe company says the method can serve electrocatalysis and electrosynthesis
None of the following is stated in official material Not yet confirmed.
The number of AISFs and their throughput by site; the headcount of the materials group; revenue and number of customers; the valuation; details of how Lila Iris was evaluated; results at real-world scale for the OER catalyst and the time to a product; the outlook for a Phase II award; how the intellectual property in the catalyst is handled. The company says it will build more AISFs, but gives neither a number nor a date.
8. What the company is aiming for in materials informatics
This chapter includes things the company has not said directly. They are inferences from primary sources, set out with the evidence.
| What can be read from it | Primary evidence | Type |
|---|---|---|
| It positions its labs as the "scorer" (verifier) from which the AI learns | The Tech page says the AISFs, by running experiments at scale, become "the world's largest verifier for science" and supply the reward signal reinforcement learning needs | Sourced |
| It aims at one general-purpose science AI, not tools for individual fields | The About page says it is building a single general platform rather than many narrow, domain-specific tools | Sourced |
| Choosing catalysts (OER) as its first materials demonstration looks like an effort to build a track record to win clients and public projects | The hydrogen-catalyst post calls OER the "first physical sciences project". Two of the three Genesis Mission projects are also about catalysis and electrochemistry. Its offerings are aimed at clients (Catalyst and Creation) | Inference |
| It appears to assume that productisation and volume manufacturing are left to clients and partners | Creation says it delivers validated structures, protocols and data that clients can drop into their pipelines. No in-house manufacturing is mentioned | Inference |
| Its materials results should be judged separately from its life-science results | Figures on the website (for example "900x" and "100%") are metrics for DNA design and are easily confused with materials metrics. No conditions are given that we could find | Inference |
The reading that "catalysts were chosen first to build a track record" rests on how the company itself frames the problem. The hydrogen-catalyst post calls OER a "very ambitious" choice, yet explains it as a problem well suited to AI search: the combinations of composition, synthesis route and concentration are vast, and the criterion for winning is clear Sourced. Moreover, two of the three projects selected for the Genesis Mission concern electrocatalysis or electrochemistry. From this, this article reads the company as following an order in which it builds a success in catalysis, where "fix the conditions and you can measure it", and then widens to other reactions and clients Not yet confirmed.
9. Map of materials informatics partnerships
10. Technologies, products and services
Prices and contract terms are not disclosed Not yet confirmed. The website shows two forms of offering: Catalyst (use of Lila Iris and the AISF) and Creation (discovery work for clients). The concrete materials result that can be confirmed is the hydrogen-catalyst search below. For background, see our explainers on self-driving laboratories, closed-loop optimisation, applications to catalysts, Bayesian optimisation and active learning, and platinum-group metals and catalysis.
The hydrogen-catalyst search: the company's claims and the numbers in the preprint
| Item | What is stated | Type |
|---|---|---|
| Aim | To find catalysts free of iridium and ruthenium for the oxygen evolution reaction (OER) in acid. The company calls it a problem that has stumped scientists for 50 years | Company's claim |
| Scale | 2,942 catalysts evaluated across 53 material systems and 26 elements in three months. The first campaign took four months (while the infrastructure was still being built) and the second four weeks | Company post and preprint |
| Speed | Up to 100 samples a day, about 240 a week in practice. 17x faster than a standard lab, with more than 90% less hands-on human time per sample | Company's claim (the post gives no conditions for the comparison) |
| Result (post) | Six palladium-based material families lie on or near the Pareto front of activity versus stability. The lead candidate has been tested in acid for over 1,000 hours. The company describes it as matching ruthenium | Company's claim |
| Result (preprint) | In 1 M sulfuric acid at 10 mA/cm², PdOx exceeded 0.5 V overpotential at about 200 hours and NiTaPdOx at about 470 hours. InMnPdOx kept the overpotential below 0.5 V over 1,000 hours. The paper reports these as long-term tests under limited conditions | Preprint (not peer reviewed) |
| Search method | The preprint reports that its sequential-learning method advanced the frontier faster than fixed-policy Bayesian optimisation or in-context selection by a language model in retrospective benchmarking | Preprint's claim |
| Limits | Automation above 90% with human supervision. In the first attempt people carried samples between instruments, and a person checked safety. The company states it is "not a finished technology or a commercial product" | Sourced (company statement) |
This result rests on the company's post and a preprint that has not been peer reviewed. The preprint states measurement conditions and times, so it is easier to check than the company's post. But the measurements are tests on small thin films, not on conditions in a real electrolyser, and the company itself says it is "not a commercial product". This article reads it as a report that a composition the AI proposed, and that had not been thought promising, lasted a long time under limited conditions. Until independent repeat tests or results at real-world scale appear, its practical value cannot be assessed.
Materials engineers can take three points from this search Sourced.
(1) It measures activity and durability together. If only one is measured, good-looking materials are easy to find. The company includes in the loop a step that examines what degraded after the test, which is a useful design for measurement.
(2) The AI proposed a composition that went against conventional wisdom, and an experiment tested it. Experts had thought palladium unsuited to OER in acid. The company writes that without a lab to confirm the finding, it would have been written off as a hallucination (an AI error). The value of linking prediction and experiment in one place shows in cases like this.
(3) Moving from thin film to powder or electrode is a separate problem. Thin films made by PVD let composition be varied finely, but real electrolysers use powders or electrodes. Whether a result on a thin film is reproduced by a powder of the same composition has to be checked separately (our own note).
11. The risks this company carries
| Risk | Detail | Type |
|---|---|---|
| Little outside verification of results | Its materials result rests on the company's post and a preprint not yet peer reviewed. No independent replication could be found | Sourced |
| Distance from thin film to real device | The result comes from tests on small thin films; evaluation at real-device scale is still to come. The company itself says it is not a commercial product | Sourced |
| Automation is not complete | In the first attempt people carried samples between instruments. Robotics is under development | Sourced |
| Gap between rhetoric and record | "Scientific superintelligence" and "the most advanced operating system for science" are company philosophy. The website's "consistently outperforms others" also lacks data or conditions that we could find | Sourced |
| Breadth of business | It handles materials, chemistry, life sciences, energy and more at once, and the resources given to materials are not disclosed | Inference |
| Scale not visible from outside | Headcount, instrument counts, revenue and valuation are not disclosed, leaving few ways to check progress from outside | Not yet confirmed |
| Public projects are at an early stage | The Genesis Mission awards are Phase I. Phase II is an opportunity to apply, and no award is assured | Sourced |
| Date mismatches within the website | The dates of the seed and Series A posts disagree within the website (chapter 4) | Sourced |
12. Glossary
- Materials informatics (MI)
- Searching for new materials efficiently with data and AI, in the flow of data, then prediction and design, then testing by experiment.
- AI Science Factory (AISF)
- Lila Sciences' name for an automated lab in which AI and instruments run as one system.
- Scientific superintelligence
- The phrase the company uses for its goal: AI that reasons scientifically beyond human level. It is a philosophy and does not show that this has been achieved.
- OER (oxygen evolution reaction)
- The reaction at the anode that makes oxygen when water is split by electricity. In an acidic electrolyser, high potential and strong acid dissolve catalysts easily.
- Overpotential
- The extra voltage needed beyond the ideal to drive a reaction. Smaller is more efficient, and growth over time signals degradation.
- PVD (physical vapour deposition)
- A method of evaporating a material onto a substrate as a thin film. It is well suited to making many samples with slightly different compositions in parallel.
- Pareto front
- The boundary of the best trade-offs between two properties (here activity and stability), where improving one means worsening the other. Materials on the line cannot be ranked against each other.
- Preprint
- A manuscript made public before peer review. Its content may change later.
- Genesis Mission
- A U.S. Department of Energy effort to link AI, supercomputers and national laboratories to speed up research (as the company describes it).
- Platinum-group metals
- A collective name for rare metals such as platinum, palladium, ruthenium and iridium, widely used in catalysts.
13. References
- Lila Sciences Official website (home page): the company's description, forms of offering and news list. Checked 22 October 2026. https://www.lila.ai/
- Lila Sciences Tech page: the scientific reasoning model, reinforcement learning and the AI Science Factory. https://www.lila.ai/tech
- Lila Sciences About page: founding year, philosophy and operating principles. https://www.lila.ai/about
- Lila Sciences Team page: names and titles of the leadership. https://www.lila.ai/team
- Lila Sciences Contact page: locations (San Francisco, Cambridge, London). https://www.lila.ai/contact
- Lila Sciences Catalyst page: access to Lila Iris and the AI Science Factory. https://www.lila.ai/lila-catalyst
- Lila Sciences Creation page: discovery campaigns carried out for clients. https://www.lila.ai/lila-creation
- Lila Sciences "Join Our Mission": company introduction, US$200 million seed round and leadership (dated 3 October 2025 on the site). https://www.lila.ai/news/join-our-mission
- Lila Sciences "Welcoming New Partners in Our Mission to Build Scientific Superintelligence" (15 September 2025): the US$235 million Series A. https://www.lila.ai/news/series-a-235-million
- Lila Sciences "Announcing Lila's $350M Series A" (dated 10 October 2025 on the site): Series A close and US$550 million total raised. https://www.lila.ai/news/announcing-the-close-of-our-series-a
- Lila Sciences "Powering American Science: LILA to join DOE's Genesis Mission" (22 July 2026): three Phase I awards. https://www.lila.ai/news/powering-american-science-lila-to-join-does-genesis-mission
- Lila Sciences "How an AI-run lab cracked open green hydrogen's catalyst problem" (25 September 2026): the company's account of the OER catalyst search. https://www.lila.ai/news/how-an-ai-run-lab-cracked-open-green-hydrogens-catalyst-problem
- Paper (arXiv) "AI-guided high-throughput discovery of iridium- and ruthenium-free palladium-oxide catalysts for durable acidic oxygen evolution" (submitted 24 September 2026, arXiv:2609.30133; a preprint, not yet peer reviewed). https://arxiv.org/abs/2609.30133
14. Claim-to-source audit
| Claim in the text | Type | Source |
|---|---|---|
| The name Lila Sciences, Inc.; privately held; locations (San Francisco, Cambridge, London) | Sourced | Reference 1, 5 https://www.lila.ai/ https://www.lila.ai/contact |
| Founded in 2023, prepared inside Flagship Pioneering, went public in March 2025 | Sourced | Reference 3, 10 The company's description https://www.lila.ai/about https://www.lila.ai/news/announcing-the-close-of-our-series-a |
| The leaders' names and titles (von Maltzahn, Beam, Gómez-Bombarelli and others) | Sourced | Reference 4, 8 https://www.lila.ai/team https://www.lila.ai/news/join-our-mission |
| The US$200 million seed and the investors' names | Sourced | Reference 8 https://www.lila.ai/news/join-our-mission |
| The US$235 million Series A (15 September 2025, co-led by Braidwell and Collective Global) | Sourced | Reference 9 https://www.lila.ai/news/series-a-235-million |
| The US$350 million Series A close, US$550 million total, and additional investors | Sourced | Reference 10 https://www.lila.ai/news/announcing-the-close-of-our-series-a |
| The Series A close date differs within the website: 10 and 15 October 2025 | Sourced | Reference 10, 9 https://www.lila.ai/news/announcing-the-close-of-our-series-a https://www.lila.ai/news/series-a-235-million |
| The seed post's date on the site (3 October 2025) disagrees with its text (public in March 2025) | Sourced | Reference 8, 10 https://www.lila.ai/news/join-our-mission https://www.lila.ai/news/announcing-the-close-of-our-series-a |
| Three Genesis Mission Phase I awards, the partner institutions and the aim of each project | Sourced | Reference 11 Company announcement; the aims are the company's plans https://www.lila.ai/news/powering-american-science-lila-to-join-does-genesis-mission |
| Phase II is an opportunity to apply, and no award is assured | Not yet confirmed | Reference 11 The company's statement https://www.lila.ai/news/powering-american-science-lila-to-join-does-genesis-mission |
| The AISF as a verifier that supplies the reinforcement-learning reward | Sourced | Reference 2 https://www.lila.ai/tech |
| The philosophy of building a single general-purpose platform | Sourced | Reference 3 https://www.lila.ai/about |
| The forms of offering: Lila Iris, AISF, Catalyst and Creation | Sourced | Reference 6, 7, 1 https://www.lila.ai/lila-catalyst https://www.lila.ai/lila-creation https://www.lila.ai/ |
| OER catalyst search: 2,942 catalysts, 53 systems, 26 elements, 100 samples a day, about 240 a week, 17x, over 90% less human time | Sourced | Reference 12, 13 Company's claim https://www.lila.ai/news/how-an-ai-run-lab-cracked-open-green-hydrogens-catalyst-problem https://arxiv.org/abs/2609.30133 |
| Six palladium-based families, the lead candidate over 1,000 hours in acid, "matches ruthenium" | Sourced | Reference 12 Company's claim https://www.lila.ai/news/how-an-ai-run-lab-cracked-open-green-hydrogens-catalyst-problem |
| PdOx about 200 h, NiTaPdOx about 470 h, InMnPdOx over 1,000 h (1 M H2SO4, 10 mA/cm2, 0.5 V overpotential) | Sourced | Reference 13 Preprint (not peer reviewed) https://arxiv.org/abs/2609.30133 |
| Automation above 90%, human supervision, people carried samples, and the company's statement that it is not a commercial product | Sourced | Reference 12, 13 https://www.lila.ai/news/how-an-ai-run-lab-cracked-open-green-hydrogens-catalyst-problem https://arxiv.org/abs/2609.30133 |
| Phrases such as "scientific superintelligence" and "consistently outperforms other models" | Sourced | Reference 1, 2 Company claim and philosophy; no data or conditions given https://www.lila.ai/ https://www.lila.ai/tech |
| Headcount, instrument counts, revenue, valuation and AISF throughput | Not yet confirmed | Reference 1 Not stated in official material. This article does not estimate them https://www.lila.ai/ |
| Real-scale results for the OER catalyst, the time to a product, and independent replication | Not yet confirmed | Reference 12 Not stated. This article does not assert that none exists https://www.lila.ai/news/how-an-ai-run-lab-cracked-open-green-hydrogens-catalyst-problem |
| Main type autonomous experiments & robotics, secondary type foundation models & compute | Our calculation | Our own classification, not the company's self-description (this article) |
| The experiment-loop illustration and the three-part reading illustration | Our calculation | Our own framing (illustrations in this article) |
| Whether a thin-film result carries over to a real device needs separate checking | Our calculation | Our own note, based on general knowledge (this article) |
| The reading that catalysts were chosen first to build a track record | Inference | Our interpretation of the hydrogen-catalyst post and the Genesis Mission content (this article) |
| The reading that productisation and volume manufacturing appear to be left to clients and partners | Inference | Our interpretation of the Creation description (this article) |
| The reading that materials results should be judged apart from life-science results | Inference | Our interpretation of how the website presents its figures (this article) |
| The risk that business breadth obscures how resources are allocated to materials | Inference | Our interpretation of official material containing no materials-group resources (this article) |
Last updated: 22 October 2026 / Troy Technical
The content of this article is based on Lila Sciences' website and official announcements and on an arXiv preprint. No market-research estimates or press-only figures are used. Phrases such as "scientific superintelligence" are company philosophy, and result figures (17x, over 1,000 hours and so on) are the company's own claims. The preprint has not been peer reviewed and has not been confirmed by independent verification. Passages marked "Inference" are this article's reading of primary sources and are not statements by the company.