Science is being rebuilt—startup by startup.
In the latest wave of funding rounds, pharma partnerships, and lab-scale breakthroughs, a new class of companies is turning research into something that views less like a linear pipeline and more like an always-on production system: AI proposes an idea, robots run the experiments, new data improves the model, and the cycle repeats—rapider each time. [1]
This is not just “tech in science.” It’s startups redesigning how science itself obtains done: autonomous labs that run 24/7, AI models that design antibodies like software, quantum processors with record fidelities, and carbon-removal technologies being pulled into the market via large advance purchase agreements. [2]
Below is a news-driven view at the startups that are altering science right now—why they matter, what’s new, and what to watch next.
The new “science stack”: AI models + automated labs + proprietary data loops
A core pattern displays up again and again in today’s most ambitious science startups:
- Models generate hypotheses (molecules, materials, mechanisms, tarobtains).
- Automation tests them at scale (robotic synthesis, high-throughput asdeclares, self-driving labs).
- Fresh experimental data becomes a competitive moat—and the training set for the next iteration.
Reuters captured the logic behind this shift in coverage of autonomous lab company Lila Sciences: rather than relying only on internet-scale text data, AI-for-science leaders increasingly argue that owning the experimental data engine is the long-term advantage. [3]
That idea—science as a closed-loop data factory—supports explain why investors, industrial giants, and governments are clustering around a handful of high-impact domains.
1) Autonomous labs: turning experiments into a 24/7 engine
Lila Sciences: “AI Science Factories” for biology, chemisattempt, and materials
Lila Sciences has become one of the most watched “AI for science” companies after raising a $115 million extension that brought its Series A total to $350 million, its overall capital to $550 million, and its valuation to more than $1.3 billion, according to Reuters. [4]
Lila’s pitch is straightforward—and radical: pair specialized AI models with robotic lab systems that can run experiments continuously. Reuters reported the company signed a 235,500-square-foot lease in Cambridge, Massachutilizetts, and plans to open its platform to commercial customers via enterprise software. [5]
Why it matters: autonomous labs aren’t just about speed. They’re about reproducibility, standardized data, and the ability to explore experimental spaces too large for traditional R&D teams.
ChemLex: a self-driving chemisattempt lab in Singapore
In December 2025, Singapore-based ChemLex raised $45 million to build out an autonomous chemisattempt lab, with Granite Asia leading the round, according to company announcements and Semafor’s reporting. [6]
ChemLex describes its core system as a 24/7 autonomous chemisattempt workflow where AI and robotics design experiments, run them, and capture data with limited human intervention. Semafor notes the vision: software and hardware doing the heavy lifting, overseen by a tiny number of people—if it scales, it could reshape how drug discovery and other R&D-heavy industries allocate time and talent. [7]
Important nuance: several key performance claims (like customer counts and market-size forecasts) are coming from the company’s own release, which is common at this stage—something readers should keep in mind when comparing platforms. [8]
Galatek: automation infrastructure for “smart labs” and semiconductor packaging
Also in December 2025, Singapore-based Galatek announced roughly US$30 million in Series A funding to expand its automation and AI products across life sciences and advanced semiconductor manufacturing. [9]
Galatek positions itself as infrastructure: not a single drug program, but the software, robotics integration, and data systems (ELN/LIMS/SDMS-style workflows) that build modern labs run rapider and with cleaner traceability. [10]
Why it matters: the “lab automation layer” is increasingly the connective tissue between AI models and real-world experimental validation.
2) Programmable biology: startups turning molecules into software
Chai Discovery: $130M for AI-designed molecules and antibodies
In one of the hugegest biotech rounds of December, Chai Discovery announced a $130 million Series B at a $1.3 billion valuation, co-led by Oak HC/FT and General Catalyst, with participation including OpenAI, Thrive Capital, Menlo Ventures, and others. [11]
The company frames its mission as predicting and “reprogramming” interactions between biochemical molecules—and building a “computer-aided design suite” for molecules. In its release, Chai also described “double digit experimental success rates” in de novo antibody design and characterized improvements versus older computational approaches—claims that are exciting, but still early and highly depfinishent on benchmark definitions and real-world generalization. [12]
The broader story: major funding is increasingly going to teams that can link AI performance to experimental validation—and then translate it into partnerships and pipelines.
Profluent: $106M to scale frontier models for protein design
In November 2025, Profluent announced $106 million in financing co-led by Altimeter Capital and Bezos Expeditions, bringing total funding to $150 million. [13]
Profluent positions itself as “programmable biology,” applying large-scale AI models to design proteins—genome editors, antibodies, and enzymes. The company also cites prior research milestones (including AI-generated protein functionality demonstrations and AI-designed CRISPR claims) and declares its Protein Atlas contains 115B+ unique proteins. Those are company-stated metrics and should be read as part of how frontier-model startups signal scale and momentum. [14]
Nabla Bio: Takeda expands AI protein-design collaboration
In October 2025, Reuters reported that Nabla Bio signed a new multi-year research partnership with Takeda, with double-digit millions in upfront and research payments and potential success-based payments over $1 billion. [15]
Nabla’s platform, Joint Atomic Model (JAM), is designed to generate protein therapeutics—“like ChatGPT, but for molecular design,” as Reuters paraphrased CEO Surge Biswas. The company claims a three-to-four-week loop from design to lab testing and expects first-in-human data within one to two years for its AI-designed molecules. [16]
Why it matters: partnerships like this display how pharma is increasingly “renting” cutting-edge AI biology capability—rather than building everything in-houtilize.
Algen Biotechnologies: AstraZeneca’s up-to-$555M AI + gene-editing bet
In October 2025, Reuters reported that Algen Biotechnologies granted AstraZeneca a license to develop therapies discovered applying Algen’s AI-driven gene-editing platform in a deal worth up to $555 million (upfront plus milestones). [17]
Algen’s platform, AlgenBrain, maps genes to disease outcomes to prioritize tarobtains, according to Reuters. The report also notes Algen was spun out of UC Berkeley work connected to CRISPR pioneer Jennifer Doudna, and that AstraZeneca would have exclusive rights to develop and sell approved therapies (if any) in the relevant areas. [18]
What’s notable: deals structured as milestone-heavy “biobucks” reflect both enthusiasm and uncertainty—huge upside if programs work, but risk-sharing if they don’t.
Iambic Therapeutics: $100M+ funding as AI drug programs enter the clinic
AI-driven biotech Iambic Therapeutics raised more than $100 million in an oversubscribed round, reported in November 2025. Fierce Biotech highlighted that Iambic is advancing AI-designed therapeutics and pairing funding momentum with clinical progress and collaborations. [19]
Why it matters: the credibility bar rises sharply once programs hit human trials. Companies that combine AI claims with clinical execution are likely to define what “AI drug discovery” really means over the next few years.
3) Clinical trials startups: applying AI to find patients rapider and more fairly
Paradigm Health: $78M Series B + Flatiron clinical research acquisition
If drug discovery is the front finish, clinical trials are the bottleneck—and Paradigm Health is attempting to treat that bottleneck as a data and workflow problem.
In December 2025, Paradigm announced an oversubscribed $78 million Series B and the acquisition of Flatiron Health’s Clinical Research Business, according to MobiHealthNews. The companies also entered a multi-year partnership to integrate platforms and expand access to trials. [20]
MobiHealthNews reports Paradigm’s platform is available in 45 U.S. states across 166 healthcare provider organizations (including community oncology practices, health systems, and academic medical centers), aiming to connect patients to clinical trial services directly through provider integration. [21]
Why this matters: rapider trial enrollment can compress timelines, reduce costs, and—if done right—improve diversity and representativeness in research populations.
4) Quantum computing startups: precision breakthroughs that could reshape research
Quantum computing has long promised scientific breakthroughs (chemisattempt simulation, materials design, optimization). The challenge is error: more qubits often means more noise. Two recent headlines suggest the precision gap may be narrowing.
Silicon Quantum Computing: an 11-qubit silicon atom processor with high fidelity
A December 2025 Nature-linked report highlighted by Phys.org describes an 11-qubit atom processor in silicon built by Australian startup Silicon Quantum Computing (SQC), applying precision-placed phosphorus atoms in isotopically purified silicon—what the team calls the “14|15 platform.” [22]
Phys.org reports the system maintained strong benchmarks while scaling connectivity and cites two-qubit gate fidelities reaching 99.9% in silicon qubits—an important signal for fault-tolerant roadmaps. [23]
Why it matters for science: chemisattempt and materials simulation often require very high fidelity and scale. Hardware that improves error characteristics could accelerate the practical window for scientific applications.
Xanadu: up to $23M in Canadian federal support
Canada is also leaning into quantum commercialization. The University of Toronto reported that photonic quantum startup Xanadu was selected to receive federal support of up to $23 million per company under the Canadian Quantum Champions Program, alongside Anyon Systems, Photonic, and Nord Quantique. [24]
Why it matters: government programs like this can support quantum startups bridge the gap between lab prototypes and industrial systems—especially when the cost of hardware and talent is high.
5) AI materials discovery startups: speeding up “hard science” R&D
Materials science underpins batteries, motors, sensors, semiconductors, and critical infrastructure—but discovering and manufacturing new materials has traditionally taken years. Several startups are now compressing that cycle by combining AI with lab automation.
Altrove: $10M seed to industrialize AI-designed materials
Paris-based deeptech startup Altrove announced a $10 million seed round led by Alven (with Bpifrance participation, among others), bringing total funding to $14 million, according to the investor’s release. [25]
Altrove declares it combines proprietary AI with automated synthesis and self-learning characterization, aiming to cut discovery from years to weeks. The release also states the company has achieved early technical milestones (including rare-earth-free, cobalt-free magnetic materials and lead-free compounds for sensors/actuators) and has secured 12+ indusattempt partnerships. [26]
As always with early materials startups, the key question is scale: can a promising lab recipe become a manufacturable, cost-competitive product?
Novyte Materials: pre-seed funding for AI-driven material discovery
In India, Novyte Materials raised pre-seed funding led by Theia Ventures (amount undisclosed), according to The Economic Times. Novyte describes its system as “predict and verify” (weeks, not decades), applying physics-aware machine learning and generative methods to propose feasible materials formulations. [27]
The Economic Times report attributes to the company claims such as cutting R&D time by up to 10x and reducing early-stage physical testing costs by up to 90%—ambitious tarobtains that will ultimately depfinish on pilot outcomes and industrial validation. [28]
6) Climate science startups: carbon removal and carbon utilization obtain real purchase orders
NULIFE GreenTech: Frontier’s $44.2M offtake for biowaste-based carbon removal
Carbon removal is shifting from “concept” to “contracts.” In December 2025, Reuters reported that Frontier—a coalition backed by companies including Google and Stripe—agreed to pay $44.2 million for carbon credits from Canadian biowaste carbon removal firm NULIFE GreenTech. [29]
The deal covers 122,000 metric tons of CO₂ to be stored between 2026 and 2030, with Reuters reporting an average weighted price of $362 per ton. NULIFE’s process converts agricultural and industrial waste (including grease from food processing) into bio-oil that is injected into deep underground salt caverns for permanent storage. [30]
Why it matters: advance purchase agreements “derisk” early projects and are increasingly how climate science startups prove bankability.
Colipi: turning CO₂ exhaust into “Climate Oil” via microbes
In industrial carbon capture and utilization (CCU), German biotech startup Colipi is partnering with Continental’s ContiTech unit to utilize CO₂-rich exhaust gases at a Hamburg site to cultivate microorganisms, starting in summer 2026, according to Continental’s announcement. [31]
The partnership aims to build one of the world’s largest bioreactors to convert CO₂ (plus hydrogen and oxygen) into biomolecules such as “Climate Oil,” positioned as an alternative to plant-based oils like palm oil. [32]
7) Fusion energy startups: science, scale, and the politics of “powering the future”
Fusion is one of the highest-risk, highest-reward frontiers in science-based startups. It also sits at the intersection of physics, advanced materials, and grid-scale engineering—building it a bellwether for “deep science” commercialization.
Commonwealth Fusion Systems: a billion-dollar power purchase signal
In September 2025, Reuters reported that Italian energy company Eni struck a power purchase agreement worth more than $1 billion tied to Commonwealth Fusion Systems (CFS), purchaseing power from CFS’s planned 400-megawatt ARC fusion plant project in Virginia, expected (if it comes online) in the early 2030s. [33]
CFS—an MIT spinout founded in 2018—has raised nearly $3 billion, Reuters reported, and the PPA followed a similar agreement with Google for power from the same project. [34]
Why it matters: PPAs don’t solve physics, but they do signal market demand—especially as data centers push power requireds higher.
TAE Technologies: a headline-grabbing path to public markets
In one of the most unusual science-startup business stories of the year, Reuters reported that Trump Media & Technology Group announced a $6 billion all-stock merger with TAE Technologies, a Google-backed fusion company. The deal is expected to close in mid-2026 (pfinishing approvals), with shareholders of each company owning about 50% of the combined entity. [35]
Reuters also reported Trump Media agreed to provide up to $300 million in funding to TAE, and that the companies stated they plan to launch building what they described as the world’s first utility-scale fusion power plant in 2026—while noting that fusion has not yet produced a commercially viable reactor. [36]
What it signals: fusion startups are increasingly being pulled into the mainstream capital markets conversation—sometimes in unpredictable ways.
Agriculture is obtainting the “pharma discovery” treatment
Hexagon Bio + Corteva: nature-inspired crop protection via AI and microbial genetics
On December 16, 2025, Corteva and Hexagon Bio announced a multi-million-dollar joint venture to accelerate nature-inspired crop protection development. Corteva’s statement describes Hexagon’s platform as combining microbial genetics, AI, chemisattempt, and synthetic biology to identify and characterize natural products. [37]
AgTechNavigator added context: Corteva Catalyst has partnered with 11 companies since launching in 2024, and this deal reflects the push to industrialize discovery approaches familiar from pharma—now applied to agriculture. [38]
Why it matters: food systems are chemisattempt-and-biology systems. Startups that translate discovery engines into safer, more sustainable inputs can have outsized impact.
What ties these science startups toobtainher
Across biotech, climate, quantum, and fusion, today’s science startups tfinish to share three characteristics:
- They build platforms, not single experiments.
The winners aim to be repeatable discovery engines—so each success compounds. - They treat data as a product.
Not just publications: structured experimental datasets that train better models and attract partners. - They relocate “science risk” into contracts and ecosystems.
Milestone-heavy pharma deals, advance carbon credit purchases, PPAs for fusion—these are ways to fund uncertain science while forcing real-world accountability. [39]
What to watch next in 2026
If you’re tracking which startups truly alter science—not just raise capital—these are the signals that will matter most:
- Indepfinishent replication and benchmarks (especially for AI biology claims).
- Cycle time improvements that translate into real outputs: candidates entering the clinic, manufacturable materials, verified carbon storage, and (eventually) grid-relevant fusion milestones.
- Integration into incumbent workflows: pharma R&D, hospital systems, semiconductor fabs, and industrial plants don’t adopt new science stacks lightly.
- Regulation and governance: gene editing, clinical trial infrastructure, and environmental claims will face increasing scrutiny as scale grows. [40]
The bottom line: the most consequential science startups aren’t just applying AI—they’re building machines for discovery. And the last few months of news suggest those machines are starting to relocate from promise to proof.
References
1. www.reuters.com, 2. www.reuters.com, 3. www.reuters.com, 4. www.reuters.com, 5. www.reuters.com, 6. www.prnewswire.com, 7. www.semafor.com, 8. www.prnewswire.com, 9. www.prnewswire.com, 10. www.prnewswire.com, 11. www.businesswire.com, 12. www.businesswire.com, 13. www.businesswire.com, 14. www.businesswire.com, 15. www.reuters.com, 16. www.reuters.com, 17. www.reuters.com, 18. www.reuters.com, 19. www.fiercebiotech.com, 20. www.mobihealthnews.com, 21. www.mobihealthnews.com, 22. phys.org, 23. phys.org, 24. www.utoronto.ca, 25. alven.co, 26. alven.co, 27. m.economictimes.com, 28. m.economictimes.com, 29. www.reuters.com, 30. www.reuters.com, 31. www.continental-indusattempt.com, 32. www.continental-indusattempt.com, 33. www.reuters.com, 34. www.reuters.com, 35. www.reuters.com, 36. www.reuters.com, 37. www.corteva.com, 38. www.agtechnavigator.com, 39. www.reuters.com, 40. www.mobihealthnews.com
Leave a Reply