Biotech: what are startups building now?

Why are investors funding drugs instead of vague biotech platforms?

Biotech investors currently want proof much earlier, and that is changing what startups get created. The platform-company model has not disappeared, but a technology without a convincing near-term medicine has become dramatically harder to finance than a drug already approaching or inside the clinic.

The contrast in funding stages is the clearest evidence. About two-thirds of the venture rounds tracked by BioPharma Dive during the first half of 2026 involved companies with clinical-stage assets. S&P Global separately found that the number of U.S. biotech venture rounds had fallen 25.2% year over year to 237, its lowest level since at least 2021, while the capital raised declined much less sharply. Fewer companies are being financed, while relatively large amounts continue flowing to selected winners.

That selection pressure changes company design. Beeline Medicines did not emerge asking investors to wait years for its discovery platform to produce something interesting. Its Series A eventually reached $426.3 million while the company prepared afimetoran for pivotal development in lupus and additional candidates for clinical studies. Alveus raised $197 million around a portfolio that already included a Phase 2 obesity program. Bionyra launched with $165 million and three licensed immunology assets.

BCG has quantified how brutal this repricing has become at the opposite end. Its biopharma review estimated that the average value of preclinical companies had fallen from roughly $500 million during the 2021 market to below $50 million. An undifferentiated platform can no longer rely on distant optionality to justify its existence.

The strongest companies still build reusable technology, but increasingly use it to produce measurable clinical bets. Chai can raise $400 million around molecular-design technology partly because major pharmaceutical companies are already using it. Generate can make expansive claims about generative biology because it has pushed programs into human testing. Iambic combines its computational engine with proprietary clinical-stage molecules. Xaira, despite being one of the largest pure platform bets of the current cycle, is explicitly building an internal drug pipeline alongside its AI models.

What investors appear to reject is reusable technology with no testable path to value. They are much more willing to fund a platform when the platform is already producing a medicine, a partner program or a clinical readout.

If you want more recent data on this point, please see our latest biotechnology market report.

Is AI biotech finally building drugs rather than demos?

Yes. AI biotech is currently crossing from computational tooling into actual drug creation, although the decisive proof — repeated clinical success — is still missing. The leading startups now design molecules, run wet labs, advance proprietary pipelines and increasingly partner with pharmaceutical companies on live discovery programs.

Chai Discovery is an unusually clear example. The company raised $400 million at a $3.8 billion valuation after developing models designed to generate molecular structures and therapeutic candidates. It has subsequently announced antibody-discovery collaborations with argenx and Bristol Myers Squibb, on top of relationships with pharmaceutical companies including Lilly and Pfizer. What customers are buying goes well beyond an AlphaFold-like answer to “what shape does this protein have?” They want the model to propose molecules that did not exist before.

Generate Biomedicines pushes the model further by integrating computational design, physical protein production, testing and model retraining inside one system. It says it has generated, built and tested more than 42,000 proteins. More importantly, the company is no longer preclinical: GB-0895 is in global Phase 3 asthma studies, while other programs are in Phase 1.

Iambic offers another useful test because its lead internally developed molecule has entered patients. IAM1363, an AI-designed HER2 inhibitor intended to penetrate the brain, has shown early antitumor activity in Phase 1/1b testing. Bayer and Takeda have separately signed discovery collaborations with the company.

Financial Times reporting recently noted that no AI-developed drug has yet produced the body of regulatory approvals needed to establish that AI systematically improves drug-development success rates. UBS nevertheless estimates that roughly 850 AI-generated candidates are somewhere in development globally.

AI biotech has moved beyond the demo stage. The remaining test is clinical: whether those computationally generated drugs actually succeed more often than conventionally discovered ones.

Are startups trying to design biology instead of discovering it?

Increasingly, yes. One of the strongest directions in biotech today is the move from searching nature for useful molecules toward specifying a desired biological function and generating a molecule engineered to perform it.

Protein design is where this is most visible. Generate describes a closed design-build-measure-learn loop in which models propose protein sequences, automated systems physically create them, experiments measure their behavior and the resulting proprietary data improve subsequent models. The company says its system can work across antibodies, enzymes, peptides and other protein formats rather than optimizing a single drug class.

Chai is taking a related approach through general-purpose molecular design models. Xaira is attacking a complementary problem: instead of only predicting which molecule fits a target, it is building models of how cells respond to perturbations so that researchers can choose better biological targets in the first place. Its X-Cell system was trained using a genome-wide perturbation dataset containing billions of genomic measurements.

These companies are pushing beyond the computational chemistry model that dominated the first AI-drug-discovery wave. The ambition now is to make three previously empirical decisions computational: what biological process should we alter, what molecule should perform that alteration, and what molecular properties should be designed in from the beginning.

A useful indication of where this could go came from the $95 million AI BioDesign research initiative involving the Allen Institute, University of Washington and Fred Hutch. Its intended outputs include completely new proteins, biological switches and molecules capable of selectively degrading or stabilizing particular proteins.

If you want more recent data on this point, please see our latest biotechnology market report.

Is obesity still swallowing biotech?

Obesity remains one of biotech’s largest magnets for startup formation, but what companies are building has already moved beyond straightforward GLP-1 imitation. The current startup race is about longer dosing intervals, oral drugs, preservation of muscle, better weight maintenance and mechanisms that can outperform or complement the first blockbuster incretin medicines.

Alveus illustrates how much capital the opportunity can absorb. It closed a $197 million Series A to advance several metabolic programs simultaneously. Its lead drug, ALV-100, combines GIP-receptor antagonism with GLP-1-receptor agonism, while another program targets amylin biology and additional work is focused on oral small molecules.

Superluminal Medicines is attacking a much narrower entry point. Its $60 million Series B is financing a small-molecule program for rare genetic and hypothalamic obesity while using AI, structural biology and protein dynamics to pursue difficult GPCR targets. The company expects to move the lead candidate into Phase 1 and is already backed by investors that include Lilly and NVIDIA.

The strongest evidence that the market has moved past “build another weekly GLP-1” comes from dealmaking. AstraZeneca's multi-program partnership with CSPC covers eight metabolic programs and carries up to $18.5 billion in potential value, including a once-monthly GLP-1/GIP candidate. Whether all of those milestones are ever paid is another matter, but the architecture of the transaction shows what large drugmakers want: entire next-generation metabolic pipelines rather than one incremental molecule.

Has autoimmune disease become biotech’s new gold rush?

Autoimmune disease is currently one of biotech’s clearest concentrations of startup capital, and the attraction is shifting from incremental immune suppression toward something much more ambitious: resetting or selectively deleting the immune cells responsible for disease.

BioPharma Dive calculated that cancer and immune-focused drugmakers captured more than 40% of the venture funding it tracked during the first half of 2026. Autoimmune and immune-disease companies alone received about $2.1 billion. That is not one unusually large financing distorting the picture; the money is spread across antibodies, T-cell engagers, conventional cell therapy and in vivo immune engineering.

Beeline's $426.3 million Series A is one end of the spectrum. Re-Aim Therapeutics is taking a more targeted approach, raising £7 million to develop antibodies that selectively deplete disease-causing T-cell populations. Prolium launched with $50 million and began developing a CD20xCD3 T-cell engager intended to produce deep B-cell depletion without manufacturing a personalized cell therapy.

Then there are the CAR-T companies. Early clinical evidence has suggested that profound B-cell depletion can produce long remissions in diseases such as lupus, creating a completely different therapeutic objective from chronic immunosuppression. Nature Biotechnology has described the emerging competition explicitly as one between approaches including ex vivo CAR-T, in vivo CAR-T and T-cell engagers.

Trials involving autoimmune CAR-T programs at Novartis and Bristol Myers Squibb have recently been paused after serious adverse events, including deaths. The biological promise is substantial, but immune reset is far from a solved engineering problem.

Are biotech startups trying to make CAR-T without manufacturing CAR-T cells?

Yes — and this may be one of the most consequential engineering shifts in current biotech. A growing group of startups wants to turn CAR-T from a bespoke manufacturing process into an injectable medicine that programs a patient's immune cells inside the body.

Traditional CAR-T requires extracting a patient's T cells, genetically engineering and expanding them in specialized facilities, testing the manufactured product and reinfusing it. That architecture helped produce extraordinary responses in some blood cancers, but it also makes the therapy slow, expensive and operationally difficult to scale.

In vivo CAR-T attempts to remove that factory step. Instead of shipping cells out of the patient, companies deliver genetic instructions directly to selected immune cells. CREATE Medicines raised $122 million to advance an in vivo CAR program for autoimmune disease and oncology, including a CD19 therapy designed for repeat dosing. Sail Biomedicines has gone far enough that Johnson & Johnson entered a collaboration covering its in vivo CAR-T technology for immune-mediated disease and obtained an exclusive option to acquire the company.

This is no longer purely theoretical. Researchers have demonstrated stable, site-specific genetic engineering of T cells directly in living animals, while other groups have used targeted nanoparticles carrying CAR mRNA to generate therapeutic T cells in vivo. A recent Nature Materials study showed a polymer-lipid carrier that could both activate and transfect T cells without an external targeting antibody.

The difficulty shifts from cell manufacturing to delivery: finding the right cells, transferring enough genetic material and controlling expression without triggering unacceptable immune effects.

Model	Where engineering happens	Main attraction	Main unresolved problem
Autologous CAR-T	Factory, using patient's cells	Proven clinical efficacy	Cost and manufacturing
Allogeneic CAR-T	Factory, using donor cells	Off-the-shelf inventory	Rejection and persistence
In vivo CAR-T	Inside patient	Injectable, potentially scalable	Precise delivery and control
T-cell engager	No cell engineering	Conventional drug manufacturing	May not reproduce durable reset

Is gene editing finally moving inside the body?

Yes. Gene editing is increasingly being built as an in vivo medicine rather than a procedure performed on cells outside the body, and startup activity suggests the industry now views direct editing of organs as a practical development path rather than a distant objective.

The earliest commercial gene-editing model depended heavily on ex vivo intervention: remove blood-forming cells, edit them, condition the patient and return the edited cells. It can work, but the complexity resembles transplantation. In vivo editing asks a different question: can an editor and its instructions be injected into a person, reach the correct organ and make the intended DNA change there?

Liver-targeted editing is furthest ahead because lipid nanoparticles naturally accumulate in the liver. Intellia is approaching potential commercialization with in vivo CRISPR programs for hereditary angioedema and transthyretin amyloidosis. Beam is developing liver-targeted base-editing programs, while Editas has been moving an in vivo candidate aimed at dramatically reducing LDL cholesterol toward human testing.

New company formation shows investors still believe the opportunity is open. Serapha Bio launched with $230 million in concurrent financing and merger funding around SERP-01, an in vivo base-editing treatment for alpha-1 antitrypsin deficiency licensed from China's YolTech. Its objective is a one-time edit that corrects a disease-causing mutation rather than repeatedly suppressing the resulting pathology.

The more interesting direction, however, is expansion beyond easy liver targets. Programs are increasingly being built around neurological disease, cardiovascular disease and direct immune-cell engineering. An ARPA-H-backed consortium involving 12 organizations is even developing a gene-editing platform for rare pediatric epilepsies, with the aim of reaching a first-in-human study within three years.

If you want more recent data on this point, please see our latest biotechnology market report.

Have drug delivery and RNA become the same biotech problem?

For many ambitious biotech startups, almost. We increasingly find that the valuable invention is not another RNA sequence or another editing enzyme, but a delivery system capable of placing those instructions into the right cells.

The liver explains why. Lipid nanoparticles and conjugate technologies have made hepatocytes unusually accessible, producing an expanding pipeline of RNA and gene-editing drugs. Moving into the lung, muscle, brain, immune system and other organs is much harder. Nature Reviews Drug Discovery identifies tissue-specific localization, stability, immunogenicity and manufacturing as central remaining constraints on DNA- and RNA-based gene therapies.

Startups are therefore being formed specifically around extrahepatic delivery. Vivatides raised $54 million to develop RNA therapeutics and delivery systems intended to reach tissues beyond the liver. Ractigen raised more than $31 million for small activating RNA programs that include a CNS candidate and proprietary extrahepatic delivery technology.

In vivo CAR-T makes the same point from another direction. The underlying CAR sequence is not necessarily the scarce technology. The challenge is delivering RNA or DNA preferentially into T cells while achieving enough expression to matter clinically.

RNA itself is simultaneously becoming more versatile. It can instruct cells to produce proteins, suppress gene expression, activate genes or temporarily express engineered receptors without permanently changing DNA. CREATE's in vivo CAR strategy includes RNA-based engineering, while experimental nanoparticle systems can transiently turn ordinary T cells into CAR-T cells inside the body. Cardiovascular startups are also pursuing RNAi candidates such as PCSK9 and angiotensinogen programs designed around very infrequent dosing.

The practical distinction between an RNA company, a gene-regulation company and an in vivo cell-engineering company is already starting to blur.

Why are startups still obsessed with “undruggable” proteins?

Because the set of proteins that medicines can manipulate is expanding again. Biotech startups are building molecular glues, degraders and AI-designed binders to attack proteins that traditional small molecules cannot easily inhibit.

Conventional drug discovery strongly favors proteins with suitable pockets that a molecule can bind. Many disease-driving proteins do not provide one. Targeted protein degradation changes the objective: instead of disabling a protein's active site, a drug recruits the cell's own disposal machinery and causes the entire protein to be destroyed.

Degron Therapeutics, for example, has raised $95 million in total and moved DEG6498 into human trials. The candidate is a molecular-glue degrader targeting HuR, an RNA-binding protein implicated in cancer and other diseases that conventional inhibitor design has struggled to address. The company has already expanded its degradation pipeline into autoimmune and metabolic disease.

AI protein design broadens the attack surface further. Xaira and Chai are attempting to generate binding molecules against targets traditional discovery tools struggle to reach. Iambic explicitly describes part of its strategy as converting difficult or apparently undruggable targets into drug programs through computational design and rapid experimentation.

Why is radiopharma suddenly producing so many biotech startups?

Radiopharmaceuticals have become one of biotech's strongest startup categories because they combine an old therapeutic mechanism — radiation — with much better molecular targeting, and recent commercial validation has made the economics easier for investors to believe.

The idea is straightforward. A targeting molecule carries a radioactive isotope to cancer cells, concentrating radiation at the tumor instead of exposing large areas of healthy tissue. What has changed is the availability of better targeting molecules, isotopes and manufacturing capabilities.

AdvanCell is a useful measure of investor conviction. The company raised a $315 million Series D to accelerate its targeted alpha-radiotherapy pipeline, including plans to move its prostate-cancer program toward Phase 3 development. That single round is larger than the entire lifetime funding of many traditional early-stage biotechs.

Ratio Therapeutics raised $70 million to advance an actinium-based program for advanced sarcomas and expand manufacturing. Aktis Oncology had already demonstrated that radiopharma startups could reach public markets, while Novartis's Pluvicto provided commercial evidence that targeted radioligand therapy can become a significant oncology product.

What makes radiopharma structurally different from many biotech modalities is that chemistry alone is insufficient. Companies need isotope access, radiochemistry expertise, manufacturing capacity and distribution capable of handling materials that decay over time. Consequently, startups are building supply chains at the same time as medicines.

Are startups still building cell and gene therapies after the sector's setbacks?

Yes, but they are increasingly trying to fix the economics and logistics rather than simply producing another conventional cell or gene therapy. The category has not disappeared; its engineering priorities have changed.

The first commercial generation showed that extraordinary biology does not guarantee an attractive product. Personalized CAR-T requires complex manufacturing. One-time gene therapies can carry very high prices but address small patient populations. Manufacturing failures, slow uptake and reimbursement difficulties have repeatedly damaged companies with clinically interesting products.

Current startup formation reflects those lessons. In vivo CAR-T removes individualized cell manufacturing. Gene-editing programs increasingly use systemic delivery instead of stem-cell transplantation. Encoded Therapeutics recently raised $275 million while advancing a gene-regulation therapy for Dravet syndrome and expanding its own manufacturing capability.

Recent cell-and-gene-therapy work has increasingly emphasized scalability, manufacturing, access and commercialization alongside efficacy. That is a notable change from the earlier phase of the field, when demonstrating that the underlying biology worked was often the central milestone.

If you want more recent data on this point, please see our latest biotechnology market report.

Are biotech startups building factories as well as medicines?

Increasingly, yes. Automation and manufacturing are becoming part of the biotech product because AI-designed biology creates little value if experiments and production remain slow, manual and difficult to reproduce.

AI exposes the bottleneck clearly. A model can propose thousands or millions of molecules rapidly. A laboratory cannot manually synthesize and test them at comparable speed. That creates a growing gap between computational throughput and physical biological throughput.

Automata raised $45 million specifically to build integrated, AI-ready laboratory automation. Its proposition is effectively an operating layer connecting instruments, robotics and experimental workflows so that life-science labs can run more continuously and generate machine-readable experimental data.

The major AI-biotech companies are building similar infrastructure internally. Generate integrates computational protein generation with physical building and testing of those proteins, then feeds the experimental results back into its models. Iambic describes an AI-driven high-throughput experimental platform producing new biological data every week. Xaira pairs computational models with large-scale perturbational datasets rather than relying entirely on information already published by others.

A company that continuously runs experiments chosen by its model and feeds proprietary measurements back into the next model can build a dataset competitors do not possess.

Are Western biotech startups increasingly being built around Chinese drugs?

Yes. The China licensing boom is changing both where Western biotech startups source their drugs and what it means to create a biotech company. A new U.S. or European startup may begin by financing and developing an asset already discovered — and sometimes already tested clinically — in China rather than originating the science itself.

The scale of the underlying shift is enormous. IQVIA estimates that Chinese-originated assets represented 40% of all drugs in-licensed by big pharmaceutical companies in 2025. During the first half of 2026, disclosed potential value from therapeutic licensing transactions involving Chinese assets had already reached roughly $92 billion — equivalent to 88% of the entire 2025 total.

Startup formation is following that pipeline. R1 Therapeutics launched with a $77.5 million Series A and rights outside Greater China to AP306, a chronic-kidney-disease candidate from Alebund Pharmaceuticals. Serapha's core gene-editing asset came from YolTech. Other newly financed companies are similarly being assembled around already-developed external assets rather than years of proprietary discovery work.

The model is attractive because Chinese biotechs have become exceptionally fast at pushing programs through discovery and early clinical development. Western venture investors can finance a new company around a partially de-risked molecule, then use experienced U.S. or European teams for global trials, regulation and commercialization.

It is no longer necessarily cheap. Evaluate data cited by Fierce Biotech show average upfront payments in Western licensing deals with Chinese companies rising from about $52 million in 2022 to $172 million in early 2026, a 230% increase.

The pipeline itself is also large enough to change startup formation. BCG estimates Chinese companies now account for roughly 30% of the global biotech pipeline and about half of new antibody-drug conjugates. Reuters reported that the total headline value of Greater China biopharma licensing deals reached approximately $137.7 billion in 2025, almost ten times its 2021 level.

Larger deals are moving beyond single molecules. IQVIA highlighted a $15.2 billion potential Bristol Myers Squibb-Hengrui agreement spanning 13 programs and a $10.5 billion Pfizer-Innovent transaction spanning 12 oncology programs. Headline milestone values should not be mistaken for cash paid upfront, but their scale shows that Chinese pipelines are increasingly being sourced as portfolios rather than isolated bargains.

The tension for Western biotech is obvious: buying a clinical-ready external drug can be far more capital-efficient than financing five years of domestic discovery, but doing so repeatedly could shift more of the industry's original drug creation toward China.

What will separate the biotech startups that matter from the ones that disappear?

The startups most likely to matter are currently the ones converting a technological advantage into a measurable advantage in patients or development economics. A better model, editor, delivery particle or manufacturing process is valuable when it changes what medicine can be built, how quickly it reaches humans, how safely it can be administered or how economically it can scale.

We can already see the hierarchy emerging. Human data carry more weight than animal data. A clinical asset carries more weight than a theoretical pipeline. A delivery system that reaches a previously inaccessible tissue matters more than another small improvement in editing efficiency. A generative model that repeatedly creates differentiated molecules matters more than a benchmark score.

The financing market is enforcing that hierarchy. Of the 68 companies in BioPharma Dive's first-half funding dataset, roughly 62% already had a drug in human testing. J.P. Morgan's first-quarter data showed later-stage rounds absorbing approximately twice the capital committed to seed and Series A rounds.

Some of the potentially transformative projects — programmable proteins, virtual cells, in vivo immune engineering and extrahepatic gene delivery — remain technically risky. But each targets a clear constraint in the existing drug-development system.

So what are biotech startups really building now?

Biotech startups are currently building a more programmable version of medicine. The strongest current wave is defined by technologies that increasingly let researchers specify what they want biology to do and then engineer the molecule, cell or genetic instruction needed to do it.

AI companies such as Chai, Generate, Xaira and Iambic are trying to turn molecular creation into an iterative engineering process. In vivo CAR companies are trying to program immune cells without manufacturing them outside the patient. Gene-editing companies are moving editing machinery directly into organs. RNA companies are building delivery systems that can reach tissues beyond the liver. Molecular-glue developers are attacking proteins previously considered inaccessible. Radiopharma companies are integrating molecular targeting with physical isotope infrastructure.

At the same time, the industry has become much less tolerant of scientific abstraction. Roughly two-thirds of the significant venture financings tracked by BioPharma Dive in the first half of 2026 involved companies with drugs already in human testing. Early-stage financing remains constrained even while very large rounds go to selected clinical or highly differentiated companies.

That combination produces the apparent contradiction defining biotech today. The technology is becoming more ambitious while company formation is becoming more pragmatic.

The geographical definition of biotech is changing too. Chinese laboratories are supplying an expanding portion of the global experimental-drug pipeline, while Western investors increasingly form new companies around those assets. In parallel, large pharmaceutical companies are licensing entire groups of Chinese programs rather than treating China mainly as a market or manufacturing base.

Our final judgment is straightforward: biotech startups are increasingly building programmable medicines and the machinery required to create, deliver and manufacture them. Biotech is slowly becoming less like exploration and more like engineering.

If you want more recent data on this point, please see our latest biotechnology market report.

Who is the author of this content?

NEW MARKET PITCH TEAM

We track new markets so founders and investors can move faster

We build living "market pitch" documents for emerging markets: AI, synthetic biology, new proteins, and more. Instead of outdated PDFs or hallucinated LLM answers, our clients get a clean, visual, always-updated view of what's really happening: key players, deals, regulations, and signals that matter. Learn more about us.