Cancer cell therapy has already shown that a patient’s own immune cells can sometimes do something ordinary drugs cannot: hunt down cancer cells, multiply inside the body and keep attacking after treatment. That idea changed blood cancer care and helped turn CAR-T therapy from an experiment into an important part of modern oncology.
But the first generation of cell therapy also exposed a long list of problems. Treatments can take weeks to make. Manufacturing is expensive. Patients may need chemotherapy before receiving the cells. Some treatments work very well against blood cancers but struggle against solid tumors. Others lose strength, fail to reach the right tissue or cause dangerous immune reactions.
Boston’s biotech industry is now attacking those problems from several directions at once.
Some companies are trying to engineer immune cells without removing them from the patient. Others are creating cell therapies with built-in biological controls. Some are combining several immune cell types in one treatment. Another group is using RNA, gene delivery and computational design to turn the body itself into the manufacturing site.
Artificial intelligence is starting to sit inside this new system.
In some cases, AI is being used directly to design RNA that behaves differently in different cells. In other cases, computational tools can help select targets, predict which genetic designs are more likely to work, identify patients who may respond and analyze the huge amount of biological data created during development.
This does not mean every Boston cell-therapy startup is an “AI company.” In fact, that would be a misleading way to describe the market.
The more important story is that cell therapy is becoming an engineering problem with an enormous search space. Once developers can test thousands of receptors, RNA sequences, delivery particles, genetic switches, cell states and patient profiles, computational systems become increasingly useful for deciding what should be built next.
That is where Boston becomes particularly interesting.
The region has cancer hospitals, immunology researchers, computational biology teams, genetic medicine startups, clinical-trial infrastructure and investors packed into a remarkably small area. The result is an emerging ecosystem in which AI does not have to replace cell biology. It can help cell biology learn faster.
Our analysis of nine Greater Boston companies working on cancer cell therapy or closely related in-vivo immune-cell programming suggests an even bigger shift is taking place.
Seven of the nine are developing approaches that can program cells directly inside the body or are explicitly building toward that model. Four already have oncology cell-therapy programs in human clinical development. Five companies in our dataset announced at least $652 million of financing or private-placement capital during 2026 alone.
That combination matters.
Boston’s next generation of cancer cell therapy may not look like today’s CAR-T industry with slightly better software.
It may look like programmable medicine.
The Short Version: AI Is Becoming Part of the Cell-Therapy Design Loop
Traditional cancer drug development usually begins with a molecule. Researchers find a biological target, design a drug against it, test the drug and slowly learn whether the original theory was right.
Cell therapy creates a much larger design problem.
Researchers must decide which immune cell to use, what that cell should recognize, how strongly it should react, how long it should survive, what happens when it enters a tumor, how it should respond to danger signals and how the treatment should be manufactured.
The number of possible combinations becomes enormous.
That is exactly the type of problem where computation becomes useful.
AI can potentially help researchers narrow the search before expensive laboratory work begins. It can find patterns across genomic, protein, imaging and patient-response data. It can help rank possible cancer targets. It can improve RNA or protein designs. It can also help researchers understand why one engineered cell succeeds while another fails.
The Boston opportunity is especially large because the city is not building only one type of cell therapy.
Obsidian Therapeutics is developing engineered tumor-infiltrating lymphocytes with a drug-controlled IL-15 system. Its lead therapy, amsoki-cel, entered a registration-enabling melanoma cohort in 2026 after earlier Phase 1/2 data showed a 67% objective response rate at the recommended Phase 2 dose in 15 evaluated patients.
Kelonia Therapeutics is attempting to create CAR-T cells directly inside patients using targeted lentiviral particles. Lilly announced an agreement in April 2026 to acquire the Boston company, whose KLN-1010 program had entered Phase 1 testing in relapsed or refractory multiple myeloma.
CREATE Medicines is taking the in-vivo idea further. Its MT-304 program is designed to program multiple immune-cell types against HER2-positive solid tumors, and the company dosed the first patient in a Phase 1/2 study in April 2026.
Kernal Biologics is explicitly combining targeted mRNA delivery with AI models designed to control where therapeutic RNA is translated. The company describes its approach as a path toward scalable in-vivo CAR-T treatment for cancer and autoimmune disease.
BrainChild Bio, meanwhile, is taking CAR-T therapy directly into one of oncology’s hardest areas: brain tumors. In September 2026, the Seattle-and-Cambridge company launched a pivotal Phase 2 study of BCB-276 in diffuse intrinsic pontine glioma and announced a $116 million Series A financing.
These are not five versions of the same idea.
They show how widely the definition of cell therapy is beginning to expand.
Why Today’s Cell Therapy Still Needs a Major Redesign
The success of CAR-T therapy can make the field look more mature than it really is.
The FDA already lists multiple commercial cellular and gene therapies, including CAR-T products such as Abecma, Breyanzi, Carvykti, Tecartus and Yescarta, as well as the TIL therapy Amtagvi and TCR-engineered Tecelra.
That is real progress.
Yet most of the biggest opportunities remain unsolved.
Manufacturing Is Still Part of the Medicine
With many autologous cell therapies, treatment begins by collecting cells from the patient.
Those cells must then be processed, modified, expanded, tested, transported and eventually returned. Every extra step creates another place where time, cost or operational complexity can enter the system.
That is very different from manufacturing a bottle of tablets months before a patient needs it.
The product is partly created for one individual.
This is why Boston’s movement toward in-vivo programming could matter so much. If an injectable medicine can safely turn the correct immune cells into therapeutic cells inside the patient, much of the traditional manufacturing chain could eventually change.
Instead of:
collect → ship → engineer → expand → test → ship back → infuse
the future could move toward:
design → manufacture a standardized delivery product → inject → program cells in the patient
That is not a small improvement.
It changes the economics of the category.
Blood Cancer Has Been Easier Than Solid Tumors
CAR-T’s greatest commercial successes came first in blood cancers.
There is a biological reason.
Blood cancer cells can expose targets such as CD19 or BCMA that make them easier for engineered immune cells to recognize. The therapeutic cells also do not have to penetrate a dense physical tumor in the same way they must enter a lung, brain or breast tumor.
Solid tumors create several additional problems.
A tumor can contain different groups of cells with different markers. It can suppress immune activity around itself. Engineered cells may struggle to reach the tumor. The target may also appear on healthy cells, increasing the risk of damaging normal tissue.
The challenge therefore changes from simply giving a T cell a new target.
Researchers need to design an entire operating system for the cell.
The Cell Must Make Decisions
Consider what an ideal engineered immune cell would need to do.
It would need to recognize cancer accurately enough to avoid attacking healthy tissue. It would need to enter the tumor. It would need to keep functioning in a hostile environment. It might need to react differently depending on the biological signals around it.
It may also need a safety switch.
Once those requirements are considered together, cell therapy starts to look less like traditional drug chemistry and more like biological programming.
That is where computational design becomes increasingly valuable.
Where AI Actually Fits Into Cell Therapy
AI’s role is easy to exaggerate, so it is useful to be precise.
An algorithm cannot make a bad biological idea good simply by processing more data. A model also cannot replace the laboratory experiments needed to prove that an engineered immune cell behaves safely.
What AI can do is help researchers search very large biological design spaces faster.
Layer 1: Finding Better Cancer Targets
The first question is deceptively simple:
What should the engineered cell attack?
An ideal target would appear strongly on cancer cells and rarely on healthy cells. It would be present across enough tumor cells to prevent easy escape. It would also remain stable as the cancer evolves.
Those requirements dramatically narrow the field.
AI can help combine gene-expression data, protein measurements, tissue atlases and patient outcomes to rank possible targets.
The important change is not merely speed.
Better models could eventually search for target combinations.
Instead of programming a cell to attack when it sees marker A, a future treatment might react only when it detects a particular pattern involving A, B and C.
That potentially creates more precise biological logic.
Layer 2: Designing the Receptor
Once the target is chosen, scientists still have to build a receptor that can recognize it.
There can be many possible protein sequences capable of interacting with the same target. Small changes can influence binding strength, stability, specificity and the way the engineered T cell behaves.
Protein-design models can help narrow that search.
Instead of manually producing a few candidates, teams can computationally explore much larger libraries and prioritize the most promising designs for laboratory testing.
That does not eliminate experiments.
It changes which experiments are worth running.
Layer 3: Programming the Genetic Payload
The genetic instructions delivered to a cell can contain much more than a receptor.
Researchers can add molecules intended to improve cell survival, alter immune signaling or help the cell withstand the tumor environment. They can also build systems designed to turn activity up or down.
Each additional component creates another design variable.
This is especially important for companies using RNA.
Kernal Biologics provides one of the clearest Boston examples. The company says its therapeutics combine targeted delivery with AI models intended to make RNA translate selectively inside the desired cells.
That creates a different computational challenge.
The question is no longer only:
Can we design useful RNA?
It becomes:
Can we design RNA that produces the desired protein in one cellular context while remaining inactive in another?
That kind of selectivity could become extremely important as in-vivo cell programming expands.
Layer 4: Choosing the Delivery System
A brilliant genetic program is useless if it enters the wrong cell.
In-vivo cell therapy therefore turns delivery into one of the most important design problems in medicine.
Developers are experimenting with viral particles, lipid nanoparticles, extracellular vesicles and other systems.
Each delivery system contains design choices around size, surface chemistry, targeting molecules, payload capacity and immune response.
Computational methods can help model some of those interactions and rank candidate designs before testing.
The market may eventually discover that the delivery layer is just as valuable as the cell program itself.
Layer 5: Understanding the Patient
Even a well-designed cell therapy may work very differently from one patient to another.
Cancer varies between people. The tumor microenvironment varies. Previous treatments matter. Immune fitness matters.
The industry therefore needs tools that can help answer another question:
Who should receive this therapy?
This is where companies such as BostonGene become relevant to the wider ecosystem even though BostonGene is not itself developing a CAR-T product.
BostonGene describes its platform as using biologically grounded AI and multimodal data for drug development, treatment strategy and patient selection.
As cell therapies become more complex, patient-selection systems like these could become increasingly important.
The best cell therapy may still fail commercially if it is given to the wrong biological population.
BostonBizScene Original Research: Mapping Greater Boston’s Emerging Cancer Cell-Therapy Stack
To understand what is actually changing in Boston, we created a small original dataset of Greater Boston companies developing cancer-focused cell therapy or enabling technologies that can directly create therapeutic immune cells.
This is not a ranking of company quality, investment attractiveness or probability of clinical success.
It is a structural analysis.
We wanted to answer four questions:
- How many Boston companies are moving toward programming cells inside the body?
- How much of the ecosystem has already reached human oncology studies?
- Where is AI explicitly visible today?
- What does recent capital formation tell us about investor interest?
Methodology
We reviewed public company websites, pipeline pages, corporate releases, clinical updates and industry landscape information available through October 6, 2026.
For geographical inclusion, a company needed to be headquartered in Greater Boston, maintain a meaningful Cambridge/Boston operation tied to the program or publicly identify Boston/Cambridge as a principal company location.
For scientific inclusion, the company needed an active cancer cell-therapy program or a platform explicitly designed to engineer immune cells with a stated cancer application.
We excluded companies developing ordinary cancer drugs, antibodies or vaccines without a direct cell-programming strategy.
We also avoided treating every computational biotech company as an “AI cell-therapy startup.” AI status was recorded only when the company publicly described AI or similar computational design as part of its platform.
The final working sample contains nine companies.
Boston Cancer Cell-Therapy Dataset
| Company | Boston-area presence | Core approach | Oncology status in our review | In-vivo/direct programming? | Publicly explicit AI layer? |
|---|---|---|---|---|---|
| Obsidian Therapeutics | Cambridge | Engineered TIL therapy with drug-regulated IL-15 | Clinical | No | Not central in public platform description |
| Kelonia Therapeutics | Boston | Targeted lentiviral in-vivo CAR-T | Clinical | Yes | Not central in public description |
| CREATE Medicines | Cambridge | RNA-based multi-immune in-vivo CAR programs | Clinical | Yes | Data-driven platform, but AI not required for inclusion |
| BrainChild Bio | Cambridge/Seattle | Locoregional CAR-T for brain tumors | Phase 2 | No | Not central in public description |
| Ensoma | Boston | In-vivo hematopoietic and immune-cell engineering | Oncology preclinical work | Yes | Not central in public description |
| Kernal Biologics | Boston | T-cell-targeted mRNA and selective translation | Preclinical/development | Yes | Yes |
| STRM.BIO | Cambridge | Megakaryocyte-derived vesicle delivery for in-vivo engineering | Preclinical pipeline | Yes | Not central in public description |
| Typewriter Therapeutics | Cambridge/Tokyo | RNA-delivered non-viral gene insertion | Preclinical/development | Yes | Not publicly central |
| Factor Bioscience | Cambridge | RNA reprogramming, allogeneic and in-vivo CAR-T | Preclinical/development | Yes | Not publicly central |
Sources include company pipeline and corporate materials from Obsidian, Kelonia/Lilly, CREATE, BrainChild, Ensoma, Kernal, STRM.BIO, Typewriter and Factor.
The most interesting finding is not the AI column.
It is the in-vivo column.
Original Finding #1: Roughly 78% of the Sample Is Moving Toward In-Vivo Cell Programming
Seven of the nine companies in our dataset have platforms designed to create or modify therapeutic cells inside the patient, or are actively developing that capability.
That equals roughly 78% of the sample.
BostonBizScene Analysis
Greater Boston cancer cell-therapy sample
n = 9 companies
In-vivo/direct cell programming 7 ████████████████████ 78%
Primarily ex-vivo in this dataset 2 ██████ 22%The two primary ex-vivo approaches in our selected group are still highly sophisticated.
Obsidian engineers tumor-infiltrating lymphocytes outside the body and then returns them to the patient. BrainChild’s BCB-276 is an autologous CAR-T product delivered into the central nervous system.
But the wider direction of travel is difficult to ignore.
Boston companies are increasingly asking whether the cell-manufacturing factory can be moved into the patient.
Why That Matters More Than It First Appears
Traditional autologous cell therapy requires two products to work together.
There is the biological product itself.
Then there is the entire manufacturing and logistics system needed to create it.
In-vivo programming has the potential to collapse those two layers into something closer to a normal pharmaceutical supply chain.
Kelonia’s approach shows what this might look like. Its iGPS platform uses engineered lentiviral particles designed to enter particular cells inside the patient, allowing CAR-T cells to be generated without the conventional remove-engineer-return process.
CREATE is pursuing RNA-based immune programming and has extended the concept beyond T cells. Its HER2-directed MT-304 program is designed around coordinated programming involving NK and myeloid immune biology as well.
STRM.BIO uses megakaryocyte-derived vesicles that naturally home toward bone marrow. Its pipeline specifically includes in-vivo CAR-T, with the company highlighting potential advantages around payload capacity and repeat dosing.
Typewriter Therapeutics emerged from stealth in September 2026 with $56 million and an RNA-delivered, site-specific gene-insertion technology intended to support durable and potentially repeatable in-vivo medicines. Cancer is one of the company’s stated disease areas.
This is beginning to look like a local technology cluster rather than a collection of unrelated experiments.
Original Finding #2: Boston Is Compressing the Distance Between Platform Science and Human Testing
Four of the nine companies in our sample have cancer cell-therapy programs that have reached human clinical development.
That is roughly 44%.
Oncology clinical translation in selected sample
Human clinical oncology program 4 ███████████ 44%
Earlier oncology development 5 ██████████████ 56%The four are Obsidian, Kelonia, CREATE and BrainChild.
More important than the raw percentage is the variety of cancers being tested.
Obsidian is working in advanced melanoma and moving into other solid tumors such as non-small-cell lung cancer. BrainChild is targeting devastating pediatric brain tumors. CREATE is testing an in-vivo immune-programming approach in HER2-positive solid tumors. Kelonia’s KLN-1010 is being studied in multiple myeloma.
This gives Boston something extremely valuable:
human feedback.
Computational biology improves when researchers have high-quality data about what actually happens in people.
Animal models are useful, but they cannot reproduce every feature of human cancer.
Once therapies reach patients, developers can begin asking much more meaningful questions.
Which cell features correlate with response?
Which patients relapse?
What happens to the engineered cells over time?
Which tumors exclude them?
Which genetic programs persist?
Which biological signals appear before toxicity?
Those questions produce data that can eventually improve the next design cycle.
That is how AI becomes strategically interesting.
The value is not merely having an advanced algorithm.
The value is connecting computation to a rapid experimental and clinical feedback loop.
Obsidian Therapeutics: Making Cell Therapy Controllable
One of the hardest problems in cell therapy is control.
Once genetically modified cells enter the patient, they can expand and remain biologically active. That persistence is part of what makes cell therapy powerful, but it can also make the treatment harder to manage.
Obsidian Therapeutics has built its approach around this issue.
The Cambridge company develops engineered tumor-infiltrating lymphocytes, commonly called TILs.
TIL therapy begins with immune cells already found inside a patient’s tumor. Those cells have one extremely attractive property: they have already found the cancer.
The problem is that they may not be strong enough or persistent enough to destroy it.
Obsidian’s lead therapy, amsoki-cel, is engineered to express membrane-bound IL-15, which is designed to support immune-cell activity. Importantly, Obsidian’s cytoDRiVE system allows that IL-15 expression to be regulated using an approved small-molecule drug.
This introduces a different model for cell therapy.
Instead of building the most powerful cell possible and hoping it behaves correctly, developers can build a cell whose activity has a control layer.
That could become increasingly important as therapies become more complex.
Clinical Progress Is Turning the Concept Into Evidence
In August 2026, Obsidian said it had enrolled the first patient in a registration-enabling cohort of its Agni-01 study.
Earlier data at the recommended Phase 2 dose showed responses in 10 of 15 evaluated patients, including two complete responses, according to the company’s release.
Those numbers come from a small early dataset and should not be interpreted as proof of eventual approval.
But they move the discussion beyond platform theory.
The program is generating clinical information.
Obsidian also completed a transaction in August 2026 that left it with approximately $350 million in post-transaction cash and equivalents, according to the company, with funding expected into the second half of 2028.
Boston’s cell-therapy ecosystem therefore includes something many emerging biotech clusters lack:
a locally developed engineered cell product moving toward a possible registration pathway.
Kelonia Therapeutics: What If the Patient Becomes the CAR-T Factory?
Kelonia represents the opposite manufacturing philosophy from traditional autologous CAR-T.
Instead of removing the T cell, modifying it in a factory and returning it, Kelonia wants to deliver the genetic instructions directly to the right T cells.
The company calls its system iGPS.
Its engineered lentiviral particles are designed to enter particular cells inside the body and insert the therapeutic genetic program there.
The strategic implication is enormous.
If successful, cell therapy begins moving away from a personalized manufacturing service and toward an injectable programmable medicine.
The First U.S. Clinical Step Has Already Happened
In May 2026, Emory’s Winship Cancer Institute announced that a patient had received the first U.S. dose of investigational KLN-1010 in the inMMyCAR Phase 1 study for relapsed or refractory multiple myeloma.
That matters because in-vivo CAR-T is no longer only a presentation-slide idea.
Patients are receiving these experimental medicines.
Lilly clearly noticed the opportunity.
In April 2026, the company announced an agreement to acquire Kelonia. Lilly specifically highlighted the potential of Kelonia’s gene-delivery technology beyond a single program.
For Boston founders, this is an important signal.
A platform capable of reliably targeting the correct immune cell inside the body may be valuable even before the full clinical potential of every program is known.
The delivery system itself can become a strategic asset.
CREATE Medicines: From One Engineered Cell to an Engineered Immune System
Most discussions about CAR therapy focus on T cells.
CREATE Medicines is exploring a broader idea.
Its platform is built around programming multiple immune-cell populations inside the body.
In April 2026, the Cambridge company announced that the first patient had been dosed with MT-304, a HER2-targeted in-vivo multi-immune CAR therapy. The company describes the program as using NK and myeloid co-programming in addition to its broader immune-engineering approach.
That changes the unit of design.
Instead of asking how to create one perfect killer T cell, researchers can begin asking whether several parts of the immune system can be coordinated.
Solid Tumors May Require That Complexity
Solid tumors are not merely groups of cancer cells.
They are biological environments.
They contain immune cells, blood vessels, connective tissue and signaling systems that can protect the cancer.
One engineered cell may therefore have too many jobs.
A future treatment could theoretically use one immune population to kill tumor cells, another to reshape the environment and another to maintain longer-term immune pressure.
The exact architecture still needs to be proven clinically.
But the direction is important.
Once therapies start coordinating several cell populations, the number of possible treatment designs rises quickly.
That creates another strong reason to use computational systems.
Researchers will need ways to rank combinations instead of testing every idea manually.
Kernal Biologics: One of Boston’s Clearest AI-to-Cell-Therapy Connections
Kernal Biologics is particularly important to the AI part of this story because the company makes the computational layer explicit.
The Boston company is developing mRNA medicines designed to work selectively in particular cells.
It combines targeted delivery with AI models intended to control selective translation, and it identifies in-vivo CAR-T for cancer and autoimmune disease as an application of the platform.
This solves a subtle problem.
Delivering RNA to a cell is not necessarily enough.
The therapeutic program must also behave correctly after arrival.
Delivery Plus Translation Creates Two Layers of Selectivity
Imagine an mRNA medicine designed for T cells.
The first layer of precision is physical targeting: getting the particle into the T cell.
The second layer is informational targeting: making sure the RNA produces its therapeutic protein where intended.
Kernal is effectively trying to combine both.
That matters because biological delivery is rarely perfect.
If even a small amount of RNA enters an unintended cell type, selective translation could provide an additional safety layer.
AI becomes relevant because cells process RNA differently.
Sequence patterns can affect stability, translation and cell-type behavior. A computational system can search those relationships at a scale that is difficult to handle manually.
The result could be a form of biological software that is not merely delivered to a particular place.
It is designed to execute only under the right cellular conditions.
BrainChild Bio: Cell Therapy Is Moving Into the Brain
Few cancers show the limitations of current treatment more clearly than diffuse intrinsic pontine glioma, or DIPG.
This devastating pediatric brain tumor grows in an extremely difficult location and has historically had very few effective treatment options.
BrainChild Bio is developing BCB-276, a B7-H3-targeted CAR-T therapy for the disease.
In September 2026, the company initiated the ILLUMINATE Phase 2 pivotal study. The trial is designed to enroll children and young adults with newly diagnosed DIPG after initial radiation therapy at six pediatric oncology centers.
The company also announced a $116 million Series A financing to support the study and additional programs.
Local Delivery Changes the Engineering Problem
BrainChild’s work is notable because its CAR-T approach is delivered into the central nervous system rather than relying only on systemic circulation.
This illustrates an important rule for the future of cell therapy:
where the medicine goes may be just as important as what the medicine is.
AI-driven target discovery gets attention because it is easy to understand.
But cancer therapy may increasingly require optimization across target, cell, payload, route and dosing schedule at the same time.
The best target is not automatically useful if the engineered cell cannot reach it.
Ensoma: Turning Blood Stem Cells Into a Long-Term Source of Therapeutic Cells
Ensoma is pursuing another variation on in-vivo engineering.
Rather than focusing only on mature T cells, the Boston company is developing technology capable of targeting hematopoietic stem cells, which can produce many types of blood and immune cells.
In May 2026, Ensoma reported initial clinical safety information from the first participant treated with EN-374 for X-linked chronic granulomatous disease. The program is not an oncology treatment, but it provided an important clinical test of the company’s in-vivo HSC-engineering concept.
At the same meeting, Ensoma described preclinical oncology work showing tumor control in HER2-positive models through multiplexed CAR macrophage, NK and T-cell generation in vivo.
That creates an ambitious possibility.
Instead of programming one mature immune-cell population, developers could potentially modify a source cell that continually creates therapeutic descendants.
The engineering challenge is obviously enormous.
But if it works, durability takes on a new meaning.
STRM.BIO, Typewriter and Factor Bioscience Show How Quickly the Delivery Layer Is Expanding
The Boston market is not converging around one in-vivo technology.
That may be a good thing.
STRM.BIO is developing megakaryocyte-derived vesicles that naturally target bone marrow and can deliver genetic payloads. The Cambridge company raised $8 million in January 2026 and lists in-vivo CAR-T among its pipeline areas.
Typewriter Therapeutics emerged publicly in September with a different solution: RNA-based, site-specific, non-viral insertion of whole genes. The company raised $56 million in Series A financing and states that cancer is one of its intended therapeutic areas.
Factor Bioscience is also developing RNA-based reprogramming technologies. In 2026, the Cambridge company described both allogeneic and in-vivo CAR-T work as well as an IL-12-expressing macrophage program for solid tumors.
These companies are effectively competing over a foundational question:
What is the best way to write new instructions into immune cells?
The answer may not be the same for every disease.
Original Finding #3: At Least $652 Million of 2026 Capital Flowed Into Five Companies in the Sample
We also examined publicly disclosed 2026 financing events involving companies in our dataset.
To avoid creating a misleading number, we applied several rules.
We counted announced equity financings and Obsidian’s disclosed $350 million private placement tied to its 2026 transaction. We did not assign a value to Lilly’s Kelonia acquisition because the announced transaction value was not disclosed in the source we used.
We also excluded government awards from the total because grant funding is not economically equivalent to venture or private-placement capital.
Using that conservative method, five companies account for at least $652 million.
| Company | 2026 disclosed capital event used in analysis | Amount |
|---|---|---|
| Obsidian Therapeutics | Private placement associated with 2026 transaction | $350M |
| CREATE Medicines | Series B | $122M |
| BrainChild Bio | Series A | $116M |
| Typewriter Therapeutics | Series A | $56M |
| STRM.BIO | Series Seed 2 | $8M |
| Total | $652M |
Sources: Obsidian, CREATE, BrainChild Bio, Typewriter Therapeutics and STRM.BIO corporate releases.
Capital Distribution
Selected disclosed 2026 capital
Obsidian $350M ██████████████████████████████
CREATE $122M ██████████
BrainChild $116M ██████████
Typewriter $56M █████
STRM.BIO $8M █
Total $652MObsidian represents roughly 54% of this selected total.
That concentration is important.
It tells us the market is not simply funding early platform ideas.
A large share of the capital is flowing toward programs with clinical data and nearer-term development milestones.
At the same time, meaningful venture dollars continue moving into new in-vivo technologies.
This creates a healthy barbell.
One end contains clinical proof.
The other contains platform experimentation.
Original Finding #4: Boston’s Bigger Trend Is In-Vivo Programming, Not AI Branding
One result of our dataset surprised us.
Only a minority of the companies publicly position AI as a central part of their cell-therapy identity.
Kernal is the clearest explicit example in our selected group.
That does not weaken the argument that AI will matter.
It strengthens it.
The market appears to be moving toward a model in which computation becomes embedded inside biological development rather than being sold as a separate product category.
This is similar to what happened with cloud computing.
Companies eventually stopped describing every online business as a “cloud startup” because cloud infrastructure became normal.
AI may follow the same path in biotech.
When every cell-therapy program generates genomic data, single-cell measurements, protein data, imaging results, manufacturing data and patient outcomes, computational analysis becomes part of the operating system.
The winning company may not be the one that says “AI” most often.
It may be the one that closes the experimental loop fastest.
The Most Valuable Dataset May Be Failed Cell Designs
There is another implication that deserves more attention.
AI systems learn not only from success.
They learn from failure.
Suppose a cell-therapy company tests 500 receptor designs.
Ten work exceptionally well.
Most people immediately focus on those ten.
But the other 490 may contain equally important information.
Why did some lose activity?
Why did others attack healthy cells?
Why did some fail after repeated exposure to the tumor?
Which ones expanded too aggressively?
Which ones exhausted themselves?
When those outcomes are connected to sequence and cell-state data, the company can begin learning design rules.
That creates a proprietary dataset competitors cannot easily reproduce.
The Learning Loop Becomes the Moat
The strongest long-term system may look like this:
Patient biology
↓
Target selection
↓
Computational design
↓
Cell or RNA construct
↓
Laboratory testing
↓
Clinical testing
↓
Response + safety data
↓
Model learns from outcome
↓
Next-generation design
↺Every completed cycle potentially improves the next one.
That is why Boston’s density matters.
A company does not need to own every part of the stack if the surrounding ecosystem can help it move through the loop quickly.
Original Finding #5: Boston’s Real Advantage May Be Feedback Density
People often describe Boston biotech’s advantage as talent.
That is true, but it is incomplete.
Many regions have strong scientists.
Boston’s deeper advantage may be feedback density.
Within a relatively small geographic area, a company can interact with cancer researchers, physicians, geneticists, computational biologists, venture investors, manufacturing specialists and platform companies.
That increases the number of opportunities to test an idea against reality.
For AI-based biotechnology, this is crucial.
Models improve when high-quality biological feedback arrives quickly.
Why Hospitals Matter to AI More Than People Think
A computational platform trained only on public datasets will eventually face a ceiling.
Public data can tell researchers a great deal about genes, proteins and tumors.
It cannot fully reproduce the data created during a carefully designed clinical program.
Clinical samples can show what changes before and after treatment.
Researchers can compare responders with non-responders.
They can examine tumors that escaped therapy.
They can study immune-cell persistence over time.
This makes institutions capable of collecting detailed patient samples extremely valuable.
In other words, Boston’s hospitals are not simply places where biotech companies run trials.
They can become part of the learning infrastructure.
What AI Could Change About Solid-Tumor Cell Therapy
The biggest prize in cancer cell therapy is probably not another small improvement in blood cancer.
It is finding reliable ways to treat common solid tumors.
That requires solving several problems at once.
Better Multi-Target Logic
A cancer cell can escape a single-target therapy by reducing or losing the target.
One answer is to engineer cells that recognize more than one marker.
AI can help identify useful combinations by analyzing which proteins appear together across tumor cells and normal tissues.
The goal is not simply higher sensitivity.
It is better logic.
A therapy could eventually act more like:
attack if A AND B are present
rather than:
attack whenever A appears.
That distinction could improve precision.
Better Understanding of the Tumor Environment
Solid tumors actively suppress immune responses.
The ideal cell may therefore need a second genetic program that helps it survive or changes its surrounding environment.
But adding more biological functions can create more risk.
Models that integrate tumor biology, immune-state data and experimental results may help developers identify which combinations are worth testing first.
Better Cell-State Control
T cells can become exhausted.
Other immune populations behave differently depending on their activation state.
Researchers increasingly need to understand not only whether a cell contains the correct receptor, but what state that cell is in.
Single-cell sequencing provides extremely detailed measurements of those states.
Machine-learning systems are well suited to finding patterns in that data.
This could eventually help companies identify what a high-performing therapeutic cell looks like before treatment reaches the patient.
Safety May Become One of AI’s Most Valuable Contributions
Cell therapy has a unusual safety challenge.
A small-molecule drug eventually leaves the body.
A living engineered cell can multiply.
That makes early safety prediction extremely valuable.
FDA history already shows why long-term monitoring matters. The agency has required warnings related to secondary T-cell malignancies following currently approved BCMA- and CD19-directed autologous CAR-T treatments, while stating that the treatments’ benefits continue to outweigh their risks for approved uses.
Interestingly, the regulatory environment is also evolving as experience grows. In June 2025, FDA eliminated the REMS programs for the six then-approved BCMA- and CD19-directed autologous CAR-T therapies, saying the programs were no longer necessary and that removing them would reduce burden on the healthcare system.
This is what a maturing field looks like.
Some risks become better understood.
Other risks require longer observation.
AI will not eliminate that uncertainty.
But better models could eventually help researchers identify warning signals earlier, simulate off-target possibilities more effectively and choose safer starting designs.
What Boston Cell-Therapy Startups Should Build Around AI
The most useful AI strategy is not “add an AI team.”
Companies need to begin with a specific development bottleneck.
Start With the Expensive Decision
Every biotech program contains decisions that cost far more than others.
Running a large animal study is expensive.
Manufacturing a clinical batch is expensive.
Starting a human trial is extraordinarily expensive.
AI creates the most value when it improves decisions made before those costs occur.
A model that helps remove weak candidates before expensive development is more useful than a dashboard that describes what researchers already know.
Capture Negative Data
Biotech companies naturally celebrate successful experiments.
They should be equally disciplined about recording failures.
That includes receptor designs that bound weakly, RNA constructs that expressed in the wrong cells and immune cells that lost function.
Negative experiments define the edges of the design space.
If the dataset contains only winners, the model cannot learn why candidates fail.
Standardize Experiments
AI models struggle when every experiment is performed differently.
Cell-therapy startups should therefore treat experimental consistency as a data strategy.
Standard methods make results more comparable.
Comparable results create better training data.
Better training data improves prediction.
This sounds operational rather than glamorous, but it may become one of the most important advantages in computational biotech.
Connect Clinical and Preclinical Data Early
Teams should not wait until Phase 2 to think about the data system.
The same identifiers and biological concepts used in laboratory development should connect wherever possible to later manufacturing and clinical datasets.
Otherwise, years of information can become fragmented across incompatible systems.
The earlier this is designed, the easier the learning loop becomes.
What Investors Should Ask Before Funding an AI Cell-Therapy Company
Investors should be skeptical whenever the word “AI” appears in a cell-therapy pitch without a clear explanation of where the model changes a real development decision.
The most useful question is:
What does this model allow you to do that a strong conventional biology team could not do efficiently?
If the answer is vague, the AI layer may not be important.
Another question is:
What proprietary data will improve the model over time?
Public datasets are valuable for starting a company.
They are rarely a permanent moat.
The company needs a credible path toward generating private data through laboratory experiments, manufacturing or clinical development.
Investors should also ask what happens when the model is wrong.
In software, a poor recommendation can often be corrected quickly.
In cell therapy, a bad biological decision can consume months of work and significant capital.
Model uncertainty therefore needs to be treated seriously.
What Pharma Companies Should Look for in Boston
Pharma companies searching Boston for cell-therapy partnerships should avoid looking only for a specific CAR target.
The more strategic approach is to evaluate platform bottlenecks.
Does the company own a better delivery system?
Can it program several immune-cell types?
Does it have a control mechanism?
Can the technology work in vivo?
Does it produce proprietary clinical or experimental data that improves with every program?
Those assets can support several products.
Kelonia’s Lilly agreement illustrates why platform value matters.
Lilly did not describe the opportunity merely as one multiple-myeloma asset. It highlighted the broader applicability of Kelonia’s in-vivo gene-delivery and integration technology.
That is an important signal for Boston founders.
Build a product.
But understand what reusable capability sits underneath it.
The Market Is Moving From Cell Products to Cell-Programming Platforms
The first generation of commercial cell therapy was mainly product-driven.
A company created a specific cell product for a specific cancer target.
The next generation is becoming platform-driven.
Developers are building systems that can potentially swap targets, payloads, delivery vehicles and genetic programs.
That changes the economics of discovery.
If one platform can create several medicines, information gained from the first program can improve the second.
This is where AI becomes much more powerful.
A company developing only one fixed therapy has a limited internal dataset.
A platform company testing many related designs can create a growing map of cause and effect.
That dataset can eventually become as important as the delivery technology itself.
The Business Model Could Change as Manufacturing Moves Inside the Patient
In-vivo engineering also creates a commercial question.
What happens to cell therapy economics if bespoke manufacturing disappears?
The current cell-therapy business model includes extensive collection, manufacturing, testing and logistics.
A standardized injectable in-vivo product could theoretically be produced at much greater scale.
That does not mean treatment automatically becomes cheap.
Advanced genetic medicines can still be expensive to develop and manufacture.
But the cost structure could change dramatically.
The product becomes less dependent on maintaining a separate manufacturing journey for every patient.
That could make cell therapy accessible to hospitals without the infrastructure traditionally associated with CAR-T.
If that happens, the addressable market expands.
Boston’s Competitive Risk: Great Science Is Not Enough
Boston has one major weakness built into its strength.
Because the region has so many biotechnology companies, competition for capital and experienced development talent is intense.
A technically impressive platform can still fail if it cannot choose the right first indication.
Cell-therapy startups therefore need discipline.
The first program should not simply be the most scientifically exciting use of the technology.
It should create information that validates the platform.
For an in-vivo delivery company, that could mean choosing a disease where target cells are easy to measure after treatment.
For a controllable-cell company, it may mean selecting an indication where changes in cell activity can be connected clearly to dose or response.
The best first program acts as both a medicine and an experiment.
The Most Important KPI May Become Learning Speed
Biotech traditionally measures progress through milestones:
candidate selected.
IND filed.
first patient dosed.
Phase 1 completed.
Those milestones will remain important.
But platform companies may need another internal KPI:
How much did we learn per development cycle?
A company that runs 100 experiments but learns nothing reusable is not building a platform.
A company that discovers why 90 designs failed may be creating tremendous long-term value.
AI makes this distinction even more important.
The model improves only when each experiment generates structured information.
That means the best computational cell-therapy companies will probably look unusually disciplined about data collection.
What Boston Could Look Like by 2030
If the current direction continues, several things could happen.
First, the line between gene therapy and cell therapy will become much less clear.
An injectable genetic medicine that enters a T cell and turns it into a CAR-T cell is technically delivered like gene therapy but behaves like cell therapy.
Second, cell manufacturing could shift from centralized factories toward in-vivo programming.
Third, solid tumors could become a much larger share of cell-therapy development.
Fourth, programmable controls may become standard.
Doctors may eventually expect advanced cell therapies to include ways to adjust activity, limit duration or trigger safety mechanisms.
Finally, AI may become almost invisible.
Every serious platform may use computational design somewhere in its workflow.
At that point, the important question will no longer be:
Does this company use AI?
It will be:
Does this company learn faster because of it?
Final Takeaway: The Smartest Cancer Cell May Be the One Designed by a Learning System
The first era of cancer cell therapy proved that living immune cells could become medicines.
The next era has a harder task.
Those medicines must become easier to manufacture, easier to control, more precise and effective against far more types of cancer.
Boston is already experimenting with nearly every major path toward that future.
Obsidian is building controllable engineered TILs. Kelonia is attempting to generate CAR-T cells directly inside patients. CREATE is moving toward coordinated programming of several immune-cell types. Kernal is explicitly using AI to help control where therapeutic mRNA becomes active. BrainChild is bringing CAR-T therapy into pediatric brain tumors. Ensoma, STRM.BIO, Typewriter and Factor are exploring additional ways to program immune or blood cells without relying on today’s conventional cell-manufacturing model.
Our BostonBizScene analysis finds that seven of nine selected companies are working toward direct in-vivo programming, while four have already pushed cancer cell-therapy programs into patients. Across just five companies in the sample, disclosed 2026 financing and private-placement capital totals at least $652 million.
The deeper lesson, however, is not about the funding number.
It is about the learning loop.
Cancer cell therapy is moving from a world where researchers design one cell product at a time toward a world where they build programmable platforms capable of generating many possible therapies.
That creates enormous amounts of biological data.
AI can help turn those data into better decisions.
Better decisions can produce better experiments.
Better experiments generate more useful data.
And that new data can improve the next generation of cells.
The company that builds the largest model will not necessarily win.
The company with the most complex genetic circuit will not necessarily win either.
The strongest platform may be the one that connects computation, experiments and patients into the fastest reliable learning system.
That is why Boston matters.
Its biggest advantage may not be that it has more AI researchers or more cell-therapy startups than every other city.
It may be that so many pieces of the learning loop already exist within the same ecosystem.
And if cancer cell therapy becomes a programmable, data-driven discipline, that density could be one of the most valuable advantages a biotech city can have.
