Profluent: The Company Teaching AI to Write Biology
Nature has spent billions of years designing the proteins that make life possible. A California startup believes artificial intelligence can design entirely new ones, and pharmaceutical companies are beginning to pay attention.
In April 2026, pharmaceutical giant Eli Lilly announced a research partnership with a relatively young biotechnology company in California. The agreement carried a remarkable figure: up to $2.25 billion in potential milestone payments, with additional royalties if resulting medicines eventually reached the market. Yet the smaller company was not bringing Lilly an approved medicine, a late-stage clinical candidate, or even a conventional drug discovery pipeline. What Profluent offered was something more fundamental: an artificial intelligence platform capable of designing biological machinery that could make entirely new kinds of medicines possible. The partnership focused on developing custom enzymes that might one day enable scientists to insert large sections of DNA into the human genome, addressing genetic diseases that remain difficult to treat using existing technologies. Profluent was not simply trying to discover the next medicine. It was trying to design the tools with which future medicines could be created.
That distinction reveals why Profluent is such an interesting company. Founded in 2022 and headquartered in Emeryville, California, it belongs to an emerging generation of biotechnology startups attempting to transform biology from a discipline largely built around discovering and modifying natural molecules into one increasingly capable of designing them for specific purposes. Its scientists use large language models, conceptually similar to the systems underlying generative AI applications, except that instead of generating sentences, the models generate proteins. These proteins might become gene-editing enzymes, components of new therapies, antibodies or tools for improving agricultural crops. By September 2026, Profluent had secured collaborations with companies including Eli Lilly, BioMarin, Corteva, Revvity and Integrated DNA Technologies. It had also raised $150 million from investors including Bezos Expeditions, Altimeter Capital, Spark Capital and Insight Partners. The company’s ambition is expansive, but its underlying commercial logic is remarkably precise: if biology can become more programmable, the ability to design reliable biological components could become an exceptionally valuable position in the biotechnology industry.
From Discovering Nature to Designing What Nature Never Made
For most of biotechnology’s history, scientists have worked with the biological materials that evolution has already produced. Proteins, the molecular machines responsible for an extraordinary range of functions inside living organisms, provide a good example. Some accelerate chemical reactions, others transport molecules or recognise threats, while others help organisms repair damage, communicate or defend themselves. Their behaviour depends on their structure, which in turn is influenced by the sequence of amino acids from which they are assembled. There are twenty standard amino acids used to build proteins, and changing their order can fundamentally alter how a protein folds and functions. This creates a staggering range of possible combinations, the overwhelming majority of which have never been observed in nature.
Traditional protein engineering has generally relied on identifying a useful natural protein and modifying it to improve particular characteristics. A naturally occurring enzyme might be sufficiently active for one application but too unstable for another. Scientists can introduce mutations, test the results, select improved variants and repeat the process. This approach has produced enormous scientific advances, but it also involves considerable experimentation. The number of possible protein sequences is so vast that systematically searching them is impossible, while improving one property can sometimes damage another. An enzyme may become more powerful but less stable; a gene editor may recognise more DNA sequences but become less precise. The challenge is not merely finding proteins that work. It is finding proteins with the right combination of properties for a particular commercial or medical application.
Profluent’s central idea is to reverse the starting point. Instead of searching nature for an existing protein and attempting to adapt it, scientists can begin with the desired functionality and use artificial intelligence to generate molecular candidates that might possess those characteristics. The distinction resembles the difference between discovering a useful material and designing one according to engineering specifications. In principle, a pharmaceutical company could specify that it needs a gene-editing enzyme capable of recognising a particular DNA sequence, operating efficiently inside human cells and being small enough to fit within a delivery vehicle. Profluent’s technology would then explore possible protein sequences in search of candidates satisfying those requirements. This does not eliminate experimentation, uncertainty or biological constraints. Rather, it changes how scientists navigate an otherwise impossibly large design space.
The implications extend well beyond medicine. Agriculture depends on enzymes and genetic tools that can help develop crops with improved resilience, disease resistance and productivity. Industrial biotechnology requires proteins capable of functioning under demanding manufacturing conditions. Diagnostics rely on molecular systems that recognise biological targets with exceptional specificity. These markets differ significantly in their regulatory requirements, economics and routes to commercialisation, yet the molecular design challenges often share common features. Profluent’s strategy rests on this technological commonality. Rather than developing a separate technology for every biological industry, it is trying to build a general-purpose protein design capability that can be commercialised across several of them.
The Origins: What If Biology Had a Language?
Profluent’s scientific foundations emerged from research led by Ali Madani, its co-founder and CEO, who previously worked on machine learning at Salesforce Research. Madani had been involved in developing ProGen, a language model trained to generate artificial protein sequences. Profluent was co-founded with Alexander Meeske, a microbiologist associated with the University of Washington whose expertise includes CRISPR biology. The combination of machine learning and experimental biological knowledge would become central to the company’s identity. While many AI startups begin with access to a model and search for industries where it might be commercially useful, Profluent emerged from research investigating whether language modelling itself could provide a fundamentally different approach to understanding and engineering proteins.
In January 2023, research involving Madani and collaborators was published in Nature Biotechnology under the title Large language models generate functional protein sequences across diverse families. The study described a model trained on approximately 280 million protein sequences spanning more than 19,000 protein families. Its central proposition was that proteins contain statistical patterns that an AI system can learn, much as a language model learns relationships between words, phrases and grammatical structures. Rather than requiring scientists to manually specify every relevant rule of protein folding and function, the model could learn relationships embedded within large quantities of biological sequence data. It could then generate novel sequences that were different from natural examples while retaining useful biological properties.
The researchers tested artificial proteins across several families, including lysozymes, enzymes that can break down components of bacterial cell walls. Some generated proteins displayed catalytic efficiencies comparable with naturally occurring equivalents despite sharing as little as 31.4% of their amino-acid sequences with natural proteins. The significance was not that AI had immediately produced a market-ready medicine, but that a model trained on biological sequences could generate genuinely functional molecules rather than merely plausible-looking strings of amino acids. It offered experimental evidence that generative models could learn relationships extending beyond superficial sequence similarity. For a company founded around programmable biology, this was an important starting point: a scientific demonstration that the underlying technological premise could work in the physical world.
Profluent announced its initial $9 million seed financing in January 2023, led by Insight Partners with participation from other investors. The capital was intended partly to establish its own experimental laboratory and expand a team combining machine learning with biological research. This decision is strategically important. Computational models can generate enormous numbers of hypothetical proteins, but a model’s prediction is not equivalent to proof of biological function. Proteins must still be produced, examined and tested under realistic conditions. By building internal experimental capabilities alongside computational research, Profluent sought to control more of the process through which AI-generated sequences become validated biological tools. The laboratory was not an accessory to the AI business. It was essential to proving that the business could create something valuable.
The Protein Atlas: Building an Engine for Biological Invention
Profluent’s technology has evolved substantially since its original research. By November 2025, the company reported that its proprietary Protein Atlas contained more than 115 billion unique protein sequences, making it, according to Profluent, the world’s largest resource of its kind. This enormous collection of biological information provides the foundation for developing increasingly sophisticated models capable of recognising patterns across different proteins. In April 2025, the company introduced ProGen3, a family of protein language models supported by research examining whether the scaling relationships familiar from text-based AI also apply to biological design. The associated work explored models ranging from millions to tens of billions of parameters, with experiments involving training on billions of protein sequences and testing whether larger models could generate better functional candidates.
The distinction between data scale and scientific capability is worth examining. A larger collection of protein sequences can expose an AI system to more of the diversity generated by evolution, while larger models may become better at capturing complex relationships within that information. Yet biological applications introduce a difficulty that does not exist in quite the same form for ordinary text generation. A convincing sentence can be evaluated by a human reader almost immediately. A newly designed protein must function in the physical world, under conditions that may be difficult to predict computationally. Its stability, activity, specificity, interaction with other molecules and behaviour inside cells all matter. Consequently, the relevant measurement of progress is not simply how sophisticated a model appears, but how reliably its outputs perform when experimentally tested.
This explains Profluent’s emphasis on what is often described as a closed-loop discovery process. The model generates candidate sequences, scientists test selected designs in the laboratory, and the resulting measurements can help researchers improve future models and design decisions. Over time, this process has the potential to produce a valuable feedback advantage. Publicly available biological databases provide enormous quantities of information, but carefully controlled laboratory experiments can generate more specific data about why particular protein designs succeed or fail. Such measurements are especially useful when a model must optimise several properties simultaneously. If Profluent can systematically connect computational prediction with high-quality experimental evidence, the resulting capability may become more difficult for competitors to reproduce than a language model trained exclusively on public data.
ProGen3 also demonstrated how Profluent was expanding beyond gene editing. The company reported generating antibody candidates against twenty therapeutic targets and developing an ultra-compact gene editor. Its announcement associated the antibody targets with treatments that had collectively served approximately seven million patients and generated more than $660 billion in historical pharmaceutical sales. Those numbers describe the markets represented by the reference medicines, not revenue earned by Profluent or commercial performance achieved by its AI-generated antibodies. The scientifically interesting claim concerned the ability to produce antibody designs that performed competitively in selected experimental tests. However, experimental activity is only an early step toward a successful therapeutic product. An antibody must also demonstrate appropriate safety, stability, manufacturability and ultimately clinical benefit. Profluent is attempting to accelerate the invention of biological candidates, but the journey from a promising molecule to a commercially successful medicine remains long.
OpenCRISPR-1: The Experiment That Changed the Company’s Position
Perhaps Profluent’s most important scientific achievement arrived in April 2024, when it introduced OpenCRISPR-1, an AI-designed gene-editing system. To appreciate its significance, it is useful to understand why CRISPR technology had already transformed biotechnology. CRISPR-Cas systems originated as part of the defence mechanisms used by certain microorganisms. Researchers discovered that components of these systems could be repurposed to recognise and modify DNA at selected locations. CRISPR-Cas9 subsequently became one of the most influential tools in modern biology, making it possible to investigate genes, engineer organisms and develop new approaches to treating genetic disease. However, the natural origins of these systems also introduce limitations. An enzyme evolved for bacterial defence was not necessarily optimised for safely editing human cells, operating inside a delivery vehicle, or recognising every clinically useful DNA sequence.
Profluent sought to demonstrate that AI could design a functional alternative. Rather than relying solely on a naturally occurring Cas9 protein, the company trained models using extensive information about the diversity of CRISPR systems. In research eventually published in Nature in July 2025, its scientists described assembling a dataset containing more than one million CRISPR operons, derived from approximately 26 terabases of genomic and metagenomic information. They used this information to generate diverse artificial CRISPR-associated proteins, including some that differed substantially from naturally occurring examples. OpenCRISPR-1, the best-known outcome, was a Cas9-like protein differing by more than 400 amino-acid substitutions from the commonly used SpCas9 enzyme. It was not simply a minor adjustment to an existing protein, but a demonstration that AI could generate a considerably different molecular sequence while preserving the complex functions required for gene editing.
Experiments showed that OpenCRISPR-1 could edit DNA in cultured human cells, with results demonstrating comparable or improved performance relative to conventional SpCas9 in the selected assays reported by the researchers. The system also showed compatibility with base editing, an approach that can change individual DNA bases rather than relying on a conventional double-strand break. The study was an important demonstration of what AI-generated biological tools might achieve, and a Nature commentary highlighted its contribution to expanding the available repertoire of gene-editing proteins. Yet an essential distinction remains: successful editing in cultured cells does not establish that a technology is ready to treat patients. Safety, delivery, immune responses, performance across different tissues and unintended genetic changes all require further investigation. The study showed that AI could design functional gene-editing machinery; it did not demonstrate that OpenCRISPR-1 was a clinically validated therapy.
The achievement nevertheless changed Profluent’s strategic position. Until then, the company could be understood primarily as a promising developer of generative biological models. With OpenCRISPR-1, it had produced a recognisable and scientifically published demonstration of those models performing a difficult biological task. That mattered enormously for commercial credibility. Pharmaceutical companies evaluating AI platforms face a familiar problem: many technologies generate impressive computational results, but relatively few establish that their outputs behave as expected in biological experiments. An experimentally validated gene editor provides a far more tangible demonstration than theoretical claims about the possibilities of programmable biology.
OpenCRISPR-1 was therefore more than a scientific publication. It was a demonstration product for Profluent’s underlying technology. It showed prospective partners what the platform could accomplish while inviting the broader research community to experiment with the results.
Why Give Away a Technology You Could Sell?
Profluent’s next strategic decision was particularly unusual. Rather than restricting access to OpenCRISPR-1, the company released its sequences publicly and made the technology available without licence fees for permitted research and commercial applications. The decision appears counterintuitive in an industry where valuable patents and exclusive licensing arrangements frequently underpin business models. Gene-editing technology has been characterised by complicated intellectual property disputes and significant licensing requirements, making access to important molecular tools commercially consequential. A young biotechnology startup might therefore be expected to protect a newly created gene editor as aggressively as possible.
The reality of Profluent’s release is more nuanced than simply making everything unrestricted. The company has filed intellectual property covering OpenCRISPR and requires a licence agreement for commercial therapeutic uses, including conditions governing ethical and safe use. It has nonetheless chosen to make the underlying sequences widely accessible without charging the conventional licensing fees that can make some gene-editing tools difficult to adopt. This creates an important distinction between access to a particular molecule and access to the full proprietary capability required to develop new ones. Researchers may use OpenCRISPR-1, but they do not thereby gain Profluent’s entire model infrastructure, datasets, experimental expertise or ability to generate customised proteins on demand.
From a business perspective, this resembles an open-core strategy, although biological molecules are not software and the licensing arrangements are not identical. By releasing a working gene editor, Profluent creates opportunities for researchers to experiment with its technology, observe its limitations and explore potential applications. More importantly, it demonstrates what its underlying system can produce. An organisation that discovers OpenCRISPR-1 is useful but needs a smaller enzyme, different targeting characteristics or more specialised functionality may become a natural customer for Profluent’s proprietary design services. The freely accessible molecule becomes evidence of the value of a much larger capability that remains commercial.
There is also a reputational advantage. In emerging technological industries, scientific credibility is often established through external validation and adoption. By allowing academic and commercial researchers to work with OpenCRISPR-1, Profluent increases the visibility of its approach and potentially encourages the development of complementary tools. In November 2025, the company claimed that OpenCRISPR-1 was being used by thousands of commercial operations and academic researchers, including major pharmaceutical companies. This adoption figure is company-reported and does not establish how many users were paying Profluent. Nevertheless, it illustrates the possible role of open scientific assets as distribution and credibility mechanisms. Instead of measuring their value solely through direct licensing revenue, the company may benefit from visibility, relationships and demand for further customisation.
The strategy is not without risk. Widely distributing a functional tool can encourage experimentation by competitors, while technological improvements elsewhere may reduce its distinctiveness over time. There is also no guarantee that users of a free gene editor will eventually purchase related services. But Profluent’s apparent assumption is that the most valuable asset is not any single protein. It is the ability to continually create new ones according to changing scientific and commercial requirements. If that assumption proves correct, protecting every individual molecular design may be less important than establishing leadership in the underlying design capability.
A Different Way to Commercialise Biotechnology
The most revealing aspect of Profluent’s strategy is how it approaches the transition from scientific discovery to commercial value. Traditional biotechnology companies frequently develop their own drug candidates, raise capital to conduct preclinical research and clinical trials, and eventually seek regulatory approval or partner with larger pharmaceutical companies. This can create enormous value when a treatment succeeds, but it also exposes companies to extensive development timelines, regulatory risks and substantial financial requirements. A new therapy may take many years and considerable investment before generating its first commercial sale.
Profluent has selected a different primary route. Rather than positioning itself principally as the developer of a portfolio of medicines, it increasingly presents itself as a supplier of specialised biological design capabilities. Its website identifies two principal partnership models: discovery collaborations, in which customers work with Profluent to develop proteins tailored to their needs, and asset licensing, through which organisations can gain access to existing protein designs that may be further developed or optimised. This enables the company to serve pharmaceutical businesses, research organisations and industrial partners without necessarily assuming responsibility for every stage of product development.
The model resembles a technology platform business more than a conventional single-product biotechnology venture. A pharmaceutical company may understand the disease it wants to treat and possess extensive clinical development expertise, yet lack an enzyme with the molecular characteristics required to make the treatment work. Profluent can concentrate on that difficult design problem, while the partner contributes the disease expertise, delivery technologies, manufacturing infrastructure, clinical trials and regulatory capabilities needed to progress a medicine. Economic value can then be divided through upfront payments, research funding, milestone arrangements, licensing fees or future royalties, depending on the agreement.
This division of responsibilities helps explain the company’s appeal to large pharmaceutical groups. A single drug developer must choose which diseases to pursue and commit resources to each programme. A protein design platform can potentially contribute to several unrelated programmes and earn income from more than one customer. The technology developed for one project may also improve capabilities that are useful elsewhere, although contractual restrictions and intellectual property ownership will influence how much knowledge can be reused. Profluent is attempting to monetise the underlying capacity for biological invention, rather than tying its entire future to the commercial success of one medicine.
However, this structure involves its own trade-offs. Licensing designs to larger partners may reduce Profluent’s direct exposure to clinical development costs, but it can also limit its share of the eventual economics. Pharmaceutical partners control many of the activities determining whether a programme advances, and failure at a later stage may prevent substantial milestone or royalty payments from materialising. Moreover, Profluent must continually demonstrate that its specialised technology provides enough value to justify collaboration rather than internal development. Its business is therefore neither a conventional software subscription company nor a traditional drug manufacturer. It occupies an intermediate position in which scientific capabilities are converted into intellectual property, research services and development-stage commercial relationships.
The Lilly Deal: A $2.25 Billion Vote of Confidence, With Conditions
The collaboration announced with Eli Lilly on 28 April 2026 provides the clearest example of this strategy in practice. The agreement focuses on designing custom site-specific recombinases, enzymes capable of rearranging DNA at particular locations. One of the major limitations of many existing gene-editing technologies is the difficulty of making large, precise changes to a genome. Some genetic diseases might be addressed by correcting a single mutation, while others require inserting larger functional sequences or entire genes. Achieving these more substantial modifications safely and predictably remains a difficult scientific challenge.
Recombinases offer one possible approach, but naturally occurring enzymes may not recognise the exact DNA sequences needed for a particular therapeutic application. Profluent’s proposed advantage is the ability to design recombinases tailored to chosen genomic targets. Instead of depending on whether nature has already evolved a suitable enzyme, its AI models attempt to generate proteins with the required characteristics. Lilly, in turn, has the capabilities to evaluate promising designs, investigate their therapeutic applications and carry selected programmes through preclinical and clinical development. Under the agreement, Lilly receives exclusive rights to advance selected recombinases through these stages and potentially commercialise resulting products.
The financial arrangement illustrates why early-stage biotechnology partnerships can attract so much attention. Profluent is entitled to an undisclosed upfront payment and committed research and development funding, alongside eligibility for as much as $2.25 billion in development and commercial milestone payments, plus tiered royalties on net sales. The distinction between potential and realised revenue is crucial. The headline figure does not mean Lilly has paid Profluent $2.25 billion, nor does it guarantee that the company will receive that amount. Milestone payments depend on specified achievements, which may include development progress, regulatory events or commercial outcomes. The value of the agreement is therefore conditional on the progress of programmes that remain scientifically and clinically uncertain.
Nevertheless, the agreement is a significant commercial signal. A major pharmaceutical company is prepared to commit resources to a technology that could address limitations of current gene-editing methods. From Profluent’s perspective, the partnership also validates an important aspect of its business model: it can potentially command substantial contractual value before any resulting medicine has reached the market. The revenue available through successful partnerships may help finance the development of new models and laboratory capabilities, potentially making the platform more sophisticated and commercially attractive over time.
There is a strategic reason that Profluent’s approach could appeal to companies such as Lilly. Pharmaceutical organisations are increasingly interested in genetic medicines, but the biological tools required to make certain therapies possible may be highly specialised. Developing every such capability internally would require expertise across numerous fields of protein engineering, computational biology and experimental validation. Working with a company focused specifically on designing these molecules allows the pharmaceutical partner to access external innovation while concentrating its own resources on translating promising tools into viable treatments. Profluent is effectively attempting to become a specialist supplier of the molecular capabilities that make other companies’ therapeutic ambitions possible.
BioMarin, Corteva and the Strategy Beyond Gene Editing
The Lilly deal is the most financially striking announcement, but it does not tell the whole story. On 30 September 2026, Profluent announced a collaboration with BioMarin Pharmaceutical to design and optimise enzymes for potential new medicines. BioMarin has extensive experience developing treatments for genetically defined conditions, including enzyme replacement therapies. These treatments can supply patients with functional versions of enzymes that their bodies lack or cannot produce effectively, addressing certain metabolic disorders. Developing suitable therapeutic enzymes is difficult because they must possess the required biological activity while maintaining characteristics such as stability and compatibility with the intended treatment approach.
Under the collaboration, Profluent will use its AI models to generate and optimise potential enzyme candidates, while BioMarin will evaluate them and consider advancing promising molecules toward preclinical development. The commercial logic is subtly different from the Lilly agreement. Rather than concentrating on gene-editing machinery capable of changing DNA, the BioMarin partnership focuses on designing the therapeutic proteins themselves. It demonstrates how the same underlying computational infrastructure can potentially serve different pharmaceutical needs. The announced financial terms were not disclosed, and the partnership does not establish that any resulting therapy will succeed. Yet it broadens the evidence that established companies see value in Profluent’s protein design capabilities beyond its original CRISPR applications.
Another important collaboration began in October 2025 with Corteva, one of the world’s major agricultural technology companies. The partnership focuses on developing AI-designed gene-editing systems for applications in crop improvement. Corteva contributes agricultural research expertise, plant biology capabilities and infrastructure for testing innovations in crops, while Profluent provides its protein design technology. The potential applications include developing more resilient plants, improving agricultural characteristics and enabling more precise crop engineering. Importantly, the commercial pathway is different from pharmaceutical development. Agricultural technologies must demonstrate value through crop performance, production economics and regulatory compliance, while distribution depends on agricultural supply chains rather than pharmaceutical sales channels.
This makes Corteva particularly interesting as a strategic partner. A company that can design useful biological molecules for both human medicine and plant science may possess a genuinely versatile underlying technology. However, the relationship also tests whether that versatility can be commercialised effectively. Each industry has its own technical requirements, customers and economics. Producing a protein with desirable characteristics in a laboratory is only one component of successfully introducing a crop technology. Corteva’s role in validating and developing agricultural applications therefore addresses capabilities that Profluent would otherwise need to build independently. The company is not attempting to master every market that biology touches. It is working with organisations that already understand how to bring innovations into those markets.
Its collaboration with Integrated DNA Technologies, a Danaher company, offers another variation on the same strategy. Announced in October 2025, the partnership combines Profluent’s computational design capabilities with IDT’s experience in enzyme development, genomics and manufacturing. Here, the intended commercial outputs may include improved molecular tools used by scientific and biotechnology customers. Such products can have shorter routes to market than entirely new therapeutic drugs, although they still require reliable manufacturing and performance validation. Meanwhile, a September 2025 partnership with Revvity integrated Profluent’s AI-engineered deaminases with Revvity’s Pin-point base editing platform, allowing customers to access enhanced gene-editing components through bundled licensing arrangements.
Taken together, these partnerships suggest that Profluent is developing several complementary routes to commercialisation. Pharmaceutical collaborations offer potentially substantial long-term milestone and royalty opportunities, while partnerships involving research tools and enabling technologies may create more immediate opportunities to license functional molecules. Agricultural collaborations broaden the application of the underlying platform into another major sector. The exact revenues and margins associated with these arrangements are not publicly established, and it would be premature to describe any one of them as commercially proven at scale. Still, their diversity is strategically informative: Profluent is testing how the same underlying invention engine can create different forms of value for customers with very different business models.
The Rare Disease Opportunity: When Conventional Economics Fail
One of Profluent’s more consequential areas of work concerns rare genetic diseases, where the scientific promise of precision medicine often collides with difficult commercial economics. Many genetic disorders arise from mutations affecting relatively small patient populations. Developing a different medicine for every disease, or even every relevant mutation, can be prohibitively expensive when the potential number of patients is limited. Yet these patients may face severe conditions for which few effective treatments exist. The challenge is therefore not exclusively scientific. It is also about finding a sufficiently scalable development process that can support treatments for diseases whose individual markets may be too small to attract traditional investment.
In November 2025, Profluent announced a collaboration with the Rett Syndrome Research Trust, a nonprofit organisation supporting research into Rett syndrome, a serious neurodevelopmental disorder caused by mutations affecting the MECP2 gene. The partnership aims to develop customised base editors targeting particular disease-causing mutations. One important technical objective is making these editing systems sufficiently compact to fit within an adeno-associated virus, or AAV, a commonly investigated delivery vehicle for genetic therapies. The size of a gene-editing system can be a decisive practical constraint because the tools must reach the relevant cells before they can produce any therapeutic benefit. Designing a highly effective enzyme is of limited use if it cannot be delivered to the tissue where it is needed.
This example demonstrates why AI-based protein design may be commercially and medically significant beyond simply improving existing technologies. If a model can generate molecular tools with carefully specified properties, it may become easier to develop related therapies for different mutations without rebuilding every component from the beginning. The advantage would not necessarily be the invention of one universally effective medicine, but a more modular process for creating many different therapies. This could be especially valuable in rare diseases, where conventional development approaches struggle to distribute research costs across sufficiently large patient populations.
A related development arrived in July 2026, when the US Advanced Research Projects Agency for Health, ARPA-H, announced its THRIVE programme, committing up to $160 million over five years across multiple research teams to advance personalised genetic medicines for rare diseases. One participating team, led by GEMMABio with support from Profluent, will develop an AI-based platform for modular gene-editing technologies, including approaches to inherited disorders associated with extreme cholesterol levels. The $160 million figure represents the total potential programme commitment across participating teams, not an award made solely to Profluent. The project’s significance lies in its objective: reducing the need to develop every genetic medicine as a completely separate scientific and regulatory undertaking.
For Profluent, rare-disease research creates an opportunity to demonstrate the advantages of a reusable molecular design platform. It also introduces important responsibilities. Producing many customised gene editors does not remove the need to evaluate safety, unintended effects, tissue delivery and long-term outcomes. Nor does cheaper molecular design automatically guarantee affordable access to resulting therapies. Clinical trials, manufacturing and medical delivery remain expensive. The broader promise is that better design technologies could remove one part of the economic burden, potentially allowing more conditions to become viable targets for therapeutic development.
What Investors Are Really Funding
In November 2025, Profluent raised $106 million in financing co-led by Altimeter Capital and Bezos Expeditions, bringing its disclosed funding to $150 million. Earlier financing included the initial $9 million seed round and a $35 million raise announced in March 2024. The participation of established technology and venture investors reflects growing interest in artificial intelligence as an enabling technology for the physical sciences. Unlike software applications that can often be distributed rapidly once developed, however, biological design businesses require substantial expenditure on computational infrastructure, scientific personnel, laboratory equipment and experimental validation. Developing a competitive protein model is expensive; demonstrating that it generates useful molecules repeatedly is an additional investment.
The strategic question is what these investments will ultimately produce. If Profluent were merely a research organisation generating isolated scientific breakthroughs, its value would depend heavily on individual discoveries. Its more ambitious proposition is that improvements in data, models and experimental infrastructure can create a reusable capability with applications across many commercial programmes. This resembles a platform strategy in which upfront investment produces technology that can be deployed repeatedly. The attraction for investors is that a successful underlying platform could support numerous partnerships without requiring Profluent to independently create a complete pharmaceutical company around every candidate molecule.
Yet a biological platform does not automatically possess the economics of a traditional software business. Each new application may require different laboratory assays, specialist personnel and extensive partner collaboration. Some discoveries may be straightforward to validate, while others could demand years of experimental optimisation. There are also substantial differences between generating a protein sequence, demonstrating its activity and creating a product that meets manufacturing or regulatory standards. The company’s ability to scale will therefore depend partly on how much of its discovery and validation process can be standardised across customers. If every new partnership requires entirely bespoke scientific work, growth may remain closely tied to research capacity. If the models increasingly generate useful designs with relatively little additional experimentation, the economics could become more attractive.
Profluent has not publicly disclosed sufficient information about recurring revenue, research service margins, licensing income or profitability to assess how far this transition has progressed. Nor should its financing be confused with evidence of a sustainable business model. Raising $150 million provides the company with resources to expand its platform, but future value will depend on the quality and commercial performance of the molecules it produces. The investment thesis is not simply that AI makes protein design possible. It is that AI may eventually make valuable biological design sufficiently repeatable to support a scalable commercial enterprise.
Can Profluent Defend Its Advantage?
A major challenge for Profluent is that it is entering an increasingly crowded field. Artificial intelligence is transforming protein engineering and drug discovery, with companies adopting different approaches to the relationship between computational design and biological experimentation. EvolutionaryScale, for example, developed the ESM3 model and demonstrated its ability to generate a functional fluorescent protein substantially different from known natural proteins. Cradle, another specialist in AI-based protein engineering, provides software tools that researchers can integrate into existing laboratory workflows. In January 2026, Bayer announced a three-year collaboration with Cradle to support antibody discovery and optimisation. Other businesses, including Generate:Biomedicines and Chai Discovery, are developing capabilities across therapeutic protein design, molecular prediction and biological engineering.
These companies are not interchangeable. Some emphasise broadly applicable foundation models, others provide tools for pharmaceutical scientists, and others integrate computational design into the development of their own therapeutic candidates. Profluent’s distinguishing emphasis has been the design of functional biological machinery, particularly gene editors, and its willingness to combine published scientific demonstrations with proprietary custom-design partnerships. Nevertheless, improvements in AI research can spread quickly. Model architectures, experimental techniques and scientific findings are frequently published, while better computing infrastructure can become accessible to multiple competitors. The fact that a company develops an innovative model first does not guarantee that it will retain a technological lead.
Profluent’s possible competitive advantages lie in several interconnected capabilities. Its Protein Atlas provides access to an enormous collection of biological sequences, although the usefulness of that information depends on quality and relevance rather than size alone. Its internal laboratory allows computational predictions to be tested experimentally, generating knowledge that may be more specific than what can be learned from public sequence databases. Its published work provides scientific credibility, while commercial partnerships help identify which molecular characteristics matter to real customers. These relationships may also produce detailed knowledge about delivery constraints, manufacturability and the practical demands of different applications.
There is a possible reinforcing relationship between these elements. Better models may generate better candidate proteins, which can attract partners with difficult scientific problems. Those collaborations can reveal new design requirements and, where agreements permit, generate useful experimental learning. That knowledge may improve subsequent designs and make the platform more attractive to additional partners. However, this is a potential learning advantage rather than an established, exclusive data network effect. Pharmaceutical contracts often restrict ownership and reuse of partner-specific information, and competitors may accumulate similar capabilities. The defensibility of Profluent’s position will ultimately depend on whether it can produce meaningfully better results, at competitive cost, across applications that customers value.
Intellectual property introduces another complication. Gene-editing technology has a history of complex patent disputes, and producing an AI-designed protein that differs substantially from natural proteins does not automatically guarantee freedom to operate. A therapeutic product may depend on patented methods, delivery systems, editing architectures or other technologies regardless of the novelty of the particular enzyme. Profluent may be able to establish valuable intellectual property around its own molecular designs, but its partners will still need to consider the broader patent landscape. The promise of designing proteins beyond conventional biological scaffolds is commercially significant, yet it should not be mistaken for a universal solution to biotechnology’s licensing challenges.
The Biggest Obstacle Is Still Biology
The enthusiasm surrounding AI-generated proteins can sometimes obscure how much scientific uncertainty remains. A language model generating a convincing sequence is only the beginning of the process. A therapeutic protein must perform the intended biological function without producing unacceptable effects elsewhere. Gene editors must reach the correct cells, recognise intended DNA targets and minimise unwanted modifications. Proteins must remain stable under appropriate conditions, be manufacturable at suitable quality and scale, and avoid immune reactions that could make them unsafe or ineffective. A molecule that performs beautifully in a laboratory assay may encounter entirely different conditions inside a living organism.
These limitations are particularly important in genetic medicine. By December 2023, CRISPR technology had already reached a historic milestone when the US Food and Drug Administration approved Casgevy, a CRISPR-based treatment for sickle cell disease. The approval demonstrated that gene editing could successfully progress from experimental research to a regulated medical product. It also illustrated the demanding development infrastructure required to do so. Designing a better editing enzyme may expand what is scientifically possible, but translating that improvement into a treatment requires extensive additional work. Profluent’s successful laboratory demonstrations should therefore be understood as achievements in biological tool development, not evidence that the company has already created approved therapies.
The question of reliability becomes even more significant as the company expands into different types of proteins. Gene editors, antibodies and industrial enzymes have different performance requirements, and a model that succeeds in one area may not transfer equally well to another. The company’s commercial ambitions depend on its ability to distinguish genuine generalisation from isolated successes. Its published research provides evidence of substantial technical progress, but rigorous testing across diverse applications remains essential. As the biological design market becomes more competitive, customers are likely to demand increasingly specific evidence of performance, cost savings and developmental advantages rather than accepting the novelty of AI-generated molecules as sufficient justification.
A final risk concerns the distribution of value between the company inventing a molecular component and the organisation turning it into a product. Profluent can contribute an important enabling technology, but pharmaceutical partners may control the therapeutic programme, clinical development, regulatory submission, manufacturing and commercial relationships. Those capabilities involve enormous investment and are essential to realising the value of a successful medicine. In some cases, the most commercially powerful participant may be the organisation that controls the final product rather than the company that designed one of its components. The long-term challenge for Profluent is to ensure that its technology remains scarce and important enough to command attractive economics even as biological AI becomes more widely available.
A Future in Which Biology Becomes Programmable
The most consequential idea behind Profluent is not that artificial intelligence can generate proteins. It is that biological design could gradually move from an unpredictable process of discovery toward a more intentional form of engineering. For much of modern biotechnology, commercial development has depended on identifying useful molecules and working around their limitations. If AI can increasingly produce proteins that meet detailed functional requirements, some of those limitations may become design parameters rather than fixed constraints. The possibilities extend from molecular tools for studying diseases to improved therapeutic enzymes, customised gene editors and agricultural technologies.
This would represent more than an improvement in laboratory productivity. It could change which scientific opportunities become commercially feasible. A biotechnology company might pursue a genetic medicine previously constrained by inadequate editing tools. An agricultural organisation might develop a crop modification that existing enzymes could not easily support. An industrial manufacturer might gain access to a more stable or efficient catalyst. The common feature is that the underlying scientific opportunity becomes more accessible when an appropriate molecular tool can be designed rather than discovered by chance.
Profluent’s commercial positioning is built around this possibility. Through its proprietary models, Protein Atlas, experimental laboratory and growing network of partners, the company is attempting to establish itself near the beginning of a new biotechnology value chain. It does not necessarily need to control every therapeutic product or industrial application created using its technology. Instead, it could benefit by supplying the tools, designs and expertise required to make those products possible. OpenCRISPR-1 helped establish scientific credibility and broad accessibility; the subsequent partnerships with Lilly, BioMarin and others suggest how that credibility might translate into commercial opportunities. But the ultimate economics remain unproven, and the development of actual medicines will take considerably more than generating promising molecules.
The company’s trajectory so far illustrates something important about the relationship between scientific breakthroughs and entrepreneurship. An invention becomes commercially meaningful not simply when it is impressive, but when an organisation can identify who needs it, why existing solutions are inadequate, and how the value it creates can be captured. Profluent’s strategy is to concentrate on the biological design problems that are difficult enough for established companies to seek external expertise, while relying on partners to provide capabilities that would be expensive or impractical for the startup to replicate independently. It is a model built around specialisation, collaboration and the possibility that AI can make molecular invention more systematic.
For centuries, scientists have studied the proteins that nature created. Profluent is betting that the next chapter of biotechnology will be defined by the proteins we learn to create ourselves. Whether that ambition becomes an enduring business will depend not on how many sequences its models can generate, but on how many of those designs eventually become indispensable to the industries that use them.
And that may be the most interesting part of Profluent’s story. The company is not merely trying to accelerate scientific discovery. It is attempting to make invention itself into a repeatable commercial capability.
