Summary

The in vivo CAR-T investment boom is fundamentally a bet on a manufacturing paradigm shift. The episode says Lilly agreed on April 20, 2026, to acquire “Colonya” (the show’s transliteration) for up to $7B, including $3.25B upfront, even though its CLN1010 remains in Phase 1; the deal was triggered by early signals from just 4 or 5 patients, with roughly 4 achieving near-complete responses that lasted about 6 months, according to 刘诚. His view is that big pharma has to secure a position as soon as the path appears viable, because “once it works, its future commercial value and potential become infinite.”
The clearest commercial value of in vivo CAR-T is not higher efficacy but converting a roughly 28-day, $150K bespoke manufacturing process into a one-time dose infused in about 1 hour at a cost of roughly $10K-$20K. Conventional autologous CAR-T requires collecting a patient’s cells, editing and expanding them ex vivo, performing batch-by-batch QA/QC, and shipping them through a cold chain; in vivo approaches put a mass-producible gene-delivery tool in a vial and let the patient’s body do the manufacturing. 刘诚’s rough estimate is that preparation costs could fall 10x, while stressing that “in vivo CAR-T does not solve an efficacy problem; it solves a cost problem.”
Blood cancers have already shown that CAR-T can expand the benefit of cancer drugs from a minority of patients to a majority, while the market ceiling rises as treatment moves into earlier lines. 刘诚 contrasts the roughly 10%-20% efficacy of ordinary late-stage cancer drugs with CAR-T response rates of about 90% in some last-line blood cancers, and says that after more than 10 years of follow-up on first-generation CD19 CAR-T, nearly 40% of patients remain alive, with some surviving more than 7 years. Nanjing Legend and J&J’s myeloma product generated about $1.9B in sales in 2025, and moving from fourth-line to second-line treatment would expand the addressable population further; he nevertheless stresses that efficacy is not the same as a cure and that manufacturing cost remains the bottleneck.
Moving CAR-T inside the body does not solve its biggest biological ceiling: solid tumors. A 10-centimeter tumor contains roughly 10B-30B cancer cells, versus about 200M CAR-T cells in a typical infusion—an initial force ratio of roughly 1:100—after which the cells that reach the tumor still face infiltration barriers and TME-mediated immune suppression. 刘诚’s blunt conclusion is that if a disease cannot be treated by ex vivo CAR-T, “in vivo CAR-T is even less worth discussing”; more than 90% of cancer patients have solid tumors.
The two foremost risks of in vivo CAR-T are editing the wrong cells and losing control of the actual dose, and neither has been dispelled by response data from 4 or 5 patients. The human body contains roughly 40T cells across hundreds of cell types, and the delivery system must modify only target T cells—not endothelial, heart, or vascular cells; meanwhile, the injection contains manufacturing inputs rather than a defined number of finished cells, leaving the question of “how many shoots will grow from this handful of seeds” unresolved. The host’s central challenge is that this is close to performing gene modification inside the body while not knowing either “where the edits land” or “how many edits are made.”
China’s CAR-T edge comes from starting at roughly the same time, engineering down costs, high patient density, and rapid IIT feedback—but the same mechanism creates quality and regulatory risks. 刘诚 believes China and the US now form the leading CAR-T bloc: China’s concentrated network of top-tier hospitals, large patient pool, and investigator-initiated trials produce feedback several times faster than in the US. Local supply-chain substitution has also reduced marketed-product manufacturing costs from about $150K in the US to roughly RMB400K in China, with further room to fall. But inconsistent hospital standards and abuses such as stem-cell and anti-aging programs packaged as IITs leave “a hundred flowers blooming” alongside the risk that patients become experimental subjects.
刘诚’s 20-year forecast is not that every cancer cell will be eliminated, but that most people will no longer die from cancer. He does not expect in vivo CAR-T to replace ex vivo CAR-T: the former would function as a mass-market industrial product focused on cost and access, while the latter would handle the hardest and latest-stage cases through tighter dose and engineering control. His definition of “cure” is closer to turning cancer into a chronic disease—after diagnosis, the reaction would be “what a hassle, I need to see a doctor,” rather than a death sentence. “Within 20 years” is his personal optimistic judgment, not an extrapolation from current clinical evidence.

Deep dive

1. CAR-T’s First Rewrite Is the Philosophy of Cancer Treatment

刘诚 summarizes conventional chemotherapy, antibodies, and ADCs as “using external force to kill tumor cells,” whereas CAR-T modifies a patient’s own immune cells and then uses them to treat that patient’s cancer: the person who tied the knot must untie it. In his view, this is not another drug molecule but a revolution in the philosophy of treatment.
Unlike a chemical compound packaged in a vial, CAR-T uses “one living immune cell to kill one living cancer cell.” Cancer cells originate in the body; the immune system may fail to see them or may be suppressed after recognizing them. Genetic engineering therefore has to give T cells both target recognition and the ability to resist suppression and mount an attack.
刘诚 explains the chimeric antigen receptor with a warship analogy: CAR is the navigation system, while the antibody it carries identifies protein targets on the surface of cancer cells—the antigens. Once it finds the target, the engineered immune cell launches its attack.
His career also explains the basis for his judgment: Eureka Therapeutics has worked on CAR-T since 2014, focusing on late-stage solid tumors, and its liver-cancer program is in Phase 2 in the US with a planned move into Phase 3 soon. Earlier in his career, he worked on ex vivo CAR-T for multiple myeloma; that product now has 2 BMS programs in Phase 3.

2. “One Patient, One Drug” Turned a Revolutionary Therapy into a Manufacturing Problem

Conventional autologous CAR-T begins with leukapheresis, keeping the patient’s blood circulating outside the body for about 2 hours while immune cells are separated. The cells are then genetically edited in a laboratory, expanded by roughly 10x-100x, and finally sent back to the hospital for intravenous infusion.
The vein-to-vein time—from collection to infusion—is usually about 28 days: manufacturing takes 9-10 days, followed by roughly 2 weeks of quality testing, plus shipping and hospital scheduling. QA/QC must be completed separately for every patient and every batch.
The early debate was whether CAR-T should be classified as a technology or a drug. The US ultimately regulated it as a drug, creating a commercial contradiction: the process looks more like a personalized surgery but must meet drug-manufacturing standards, preventing it from being mass-produced and distributed through pharmacies like an ordinary drug.
A one-course CD19 CAR-T treatment for lymphoma in China is priced at more than RMB1M, versus roughly $500K in the US. But 刘诚 stresses that a high price does not mean high margins: bespoke manufacturing, quality testing, and cold-chain logistics consume a large share of the revenue.

3. Roughly 90% Efficacy Produced Long-Term Survival in Last-Line Blood Cancers

刘诚’s industry baseline is brutal: in a Phase 3 trial for late-stage cancer, a drug that benefits roughly 10% of patients may be enough to win FDA approval. Typical response rates are 10%-20%; the rest pay money, endure side effects, and may lose time as the disease progresses.
CAR-T initially targeted last-line patients with lymphoma, acute lymphoblastic leukemia, and multiple myeloma—people for whom all existing options had failed—yet achieved efficacy of roughly 90%. Among those who responded, 刘诚 says half could nearly be considered cured, explaining why the medical community’s view of the technology changed so sharply.
First-generation CD19 CAR-T has now been followed for more than 10 years. Citing data from 刘诚, nearly 40% of patients remain alive, with survival beyond 7 years—an outcome that can be regarded as a cure in the context of cancer treatment. But he also stresses that efficacy is not the same as a cure: some patients relapse or die within months.

4. Fast CAR Cannot Eliminate Quality Testing, While Allogeneic CAR-T Lost to Immune Rejection

Fast CAR tried to compress 9-10 days of manufacturing into 2-3 days, but it could not remove the subsequent quality testing, shipping, or hospital scheduling. It therefore shortened vein-to-vein time only marginally and did not change the basic one-patient-at-a-time manufacturing model.
Allogeneic CAR-T uses cells from healthy donors to build inventory in advance, theoretically shortening the wait and lowering costs through centralized manufacturing. For 7-8 years it was as hot as in vivo CAR-T is today, but its clinical efficacy ultimately fell well behind autologous products.
The problem cuts both ways: donor cells can attack the patient and cause GVHD, while the patient’s immune system can rapidly recognize and clear the donor cells as foreign before they finish clearing the tumor. Even knocking out 1 or 2 relevant genes may not prevent rejection, and extensive editing can cost T cells some of their function.
刘诚 has not declared the approach dead—companies continue to pursue it—but acknowledges that the clinical results have been disappointing and industry enthusiasm has cooled. That failure is also the direct backdrop to big pharma’s shift toward modifying patients’ own cells inside the body.

5. In Vivo CAR-T Turns the Patient’s Body into a Manufacturing Plant

The in vivo approach preserves the core of autologous CAR-T: a patient’s own immune cells still treat that patient’s cancer. The difference is that the cells are no longer removed; genetic modification and manufacturing happen directly inside the body.
What can truly be scaled is the delivery tool. It can be mass-produced in advance, bottled, shipped, and stored at hospitals, then administered like an ordinary intravenous drug. “In theory, it is an almost perfect solution,” because it bypasses bespoke manufacturing, the cold chain, and the waiting period at the same time.
刘诚 describes in vivo CAR-T as CAR-T’s “second wave,” with the first arriving about 10 years ago. AstraZeneca, AbbVie, Gilead, BMS, and others have been aggressively acquiring companies still in the preclinical, Phase 1, or Phase 2 stages, showing that capital is buying platform potential rather than mature revenue.

6. Four or Five Complete Responses Can Trigger an Acquisition, but Not Validate a Platform

The episode says Lilly has acquired 2 in vivo CAR-T companies in succession. The latest deal closed on April 20, 2026, with total consideration of up to $7B for “Colonya” (the show’s transliteration), including $3.25B upfront.
The target has no marketed product, and its lead asset CLN1010 targets relapsed/refractory multiple myeloma and remains in Phase 1. The data described by 刘诚 cover only 4 or 5 patients, of whom roughly 4 achieved near-complete responses that lasted about 6 months.
刘诚’s interpretation is that the data first show that using lentivirus to modify T cells inside the patient “might work,” not that the drug is finished. Big pharma is willing to bet early because, if the technology holds, the efficacy already demonstrated by blood-cancer CAR-T can be attached to a much lower-cost delivery model.
The more important boundary is that “in vivo CAR-T itself is a change in manufacturing method and treatment pathway.” It does not upgrade CAR-T’s biology; a cancer that cannot be treated by an ex vivo product does not automatically become treatable just because manufacturing moves inside the body.

7. In Vivo Editing Must Answer “Whom Did You Edit, and How Much?”

The first top-tier risk is specificity. The human body contains roughly 40T cells across hundreds of cell types, and the delivery system must edit only the intended T cells. 刘诚 repeatedly asks: “What happens if it edits endothelial cells? What happens if it edits the blood-vessel cells in your heart?”
The second risk is an uncontrollable dose. Ex vivo production at least establishes how many qualified cells were made and how many were infused back into the patient. In vivo treatment supplies only the manufacturing ingredients; the final number of CAR-T cells depends on the patient and is difficult to calibrate in advance.
刘诚’s seed analogy captures the substance of the problem: “I scatter a handful of seeds, but I don’t know how many sprouts will emerge or how many trees will grow.” The dose of the final active therapy is therefore difficult to control.
The host’s concern is not erased by technological optimism: this sounds like genetic modification inside one’s own body, with the possibility of missing the target as well as producing too little or too much. 刘诚 says only that the technology will improve step by step; he does not claim that either risk has been solved.

8. A 10x Cost Reduction and Immediate Dosing Open the Existing Blood-Cancer Market

刘诚 estimates that a US ex vivo CAR-T product costs about $150K to manufacture per patient: roughly $130K for cell production and about $20K for the lentivirus used to deliver the gene. Marketed products in China can already bring the comparable cost down to about RMB400K.
In vivo CAR-T still requires the delivery component, such as lentivirus, but can eliminate most of the cell-manufacturing process. If the per-patient cost falls to $10K-$20K, total preparation costs could decline by roughly 10x, while the timeline shrinks from 28 days to about 1 hour.
China’s cost advantage comes from both labor and the supply chain. 刘诚 also disclosed that he serves as chairman of JW Therapeutics; he said early CD19 CAR-T production relied heavily on imported raw materials, and that costs still have significant room to fall as domestic suppliers and local alternatives mature.
The economics initially apply to the already validated markets of lymphoma, leukemia, and multiple myeloma—not to every cancer at once. Recent in vivo CAR-T deals have concentrated on myeloma precisely because the indication has a clearer efficacy profile and commercial benchmark.

9. Solid Tumors Drag CAR-T into a 1:100 Siege

Blood-cancer cells and infused CAR-T cells share the bloodstream, while lymphatic tissue also has abundant blood supply, making encounters relatively easy. Solid tumors grow inside organs and tissues; even after CAR-T finds the target, it still has to “drill in layer by layer.”
刘诚 calls this obstacle immune-cell infiltration. A 10-centimeter tumor contains roughly 10B-30B cancer cells, compared with about 200M CAR-T cells in a typical infusion—an initial force ratio of approximately 1:100. Only a fraction actually enters the tumor, potentially worsening the local ratio to 100:1 or 1,000:1.
CAR-T’s advantage is that it can activate and expand after killing: “Kill one, and it becomes 2; then it kills 2 and becomes 4.” But in solid tumors, getting onto the battlefield is already difficult. 刘诚 summarizes the combined disadvantage of numbers and terrain as one good fighter being unable to defeat 4 opponents.

10. The Tumor Microenvironment and Integration Site Together Determine Durability

Peripheral blood does not impose the same level of immune suppression, so CAR-T can activate, attack, and expand directly after encountering blood-cancer cells. The solid-tumor TME (tumor microenvironment), by contrast, can leave immune cells that enter it “weak in all four limbs.”
How many CAR-T cells can penetrate a tumor and how long they can persist after entering remain a “black box.” Durability therefore depends not only on whether the vector can express the construct over time, but also on whether the cells can arrive, survive, and retain function in a suppressive environment.
Lentivirus can stably integrate the CAR gene into the host genome, but the integration site is random. The roughly 200M cells in a single patient may produce thousands or tens of thousands of distinct integration patterns, with potential risks including activation of oncogenes or insertion into a site with poor expression.
The host mentioned “Azelia” (the show’s transliteration), a company involving Jennifer Doudna that raised $82M and is trying to combine CRISPR with in vivo CAR-T. 刘诚 explained that site-specific insertion can improve quality, stability, and expression efficiency, but “does not solve the CAR-T problem itself.”

11. Immune Suppression Is Not a Bug, and Cancer Cells Are Not Foreign Enemies to the Immune System

刘诚 stresses that the immune system is a double-edged sword: excessive activation attacks healthy tissue, with allergies, lupus, and arthritis among the costs. The body therefore needs braking mechanisms such as immune-suppressive factors, Treg (regulatory T cells), and myeloid-derived suppressor cells.
Cancer cells may lose function, proliferate without limit, and eventually become fatal, but they are still transformed from the body’s own normal cells rather than bacteria or viruses. In terms of cellular behavior, they do not actively invade from outside, so the immune system does not naturally identify them as foreign enemies.
刘诚’s police analogy best explains CAR-T: “The police do not shoot an ordinary civilian.” The CAR gene artificially tells a force naturally built for patrol and combat who the bad guys are, then makes it treat cancer cells as invaders.
That is why he considers asking the innate immune system to attack cancer on its own “unrealistic and counterintuitive.” CAR-T’s ingenuity is to make immune cells perform a task that is not part of their evolutionary mandate but is highly compatible with their combat capabilities.

12. CAR-T’s Market Truly Opens Up When Treatment Moves from Fourth-Line to Second-Line

The host cited the commercial benchmark: since 2017, 7 ex vivo CAR-T products have been approved in the US, generally priced at about $500K per infusion. Nanjing Legend and J&J’s myeloma product generated about $1.9B in 2025 sales, clearing the $1B blockbuster threshold commonly used by big pharma.
At that price, the product treats only several thousand patients a year, while the US sees about 40K new multiple-myeloma diagnoses annually and may have more than 200K prevalent patients. The actual coverage will depend on treatment line, manufacturing capacity, cost, and the load that treatment centers can absorb.
Innovative drugs usually start in the last line. The product was initially approved for fourth-line use, after the first 3 lines had failed; after showing a clear advantage in later lines, it was compared with second-line therapy and ultimately moved into second line, where early-line patients actually had better efficacy.
刘诚’s conclusion is that “efficacy is not the problem, especially in blood cancers; manufacturing cost is the bottleneck.” Continued movement into earlier lines, combined with lower-cost in vivo production, is what gives myeloma in vivo CAR-T deals such large upside.

13. High Response Rates Still Hide a Short PFS Tail and Extramedullary Relapse

The host’s objection is worth preserving: several products have very high response rates, but median progression-free or event-free survival is only several months to the low teens, while a single infusion costs about $500K. A high initial response is not the same as a durable cure.
刘诚 clarifies that median PFS is the midpoint of a distribution: if it is 6 months, roughly half of patients have relapsed by month 6 while the other half have not. CAR-T’s unusual feature is that the latter group may remain in remission for many years, but before treatment there is no reliable way to identify who will fall into that long-term-benefit tail.
The research goal therefore has 2 layers: identify the characteristics of patients who do not relapse over the long term, and move more patients into that group. The initial explanation was antigen escape, in which cancer cells lose the target; years of data now show that this cannot explain roughly half of eventual relapses.
Another important hypothesis is extramedullary disease: late-stage blood-cancer cells move into tissue and form outposts resembling solid tumors, where CAR-T struggles to infiltrate and clear them. The host summarizes the boundary sharply: “Once blood cancer crosses into the territory of solid tumors, CAR-T becomes somewhat powerless.” 刘诚 agrees that this is the current technical limit.

14. CRS Has Shifted from a Fatal Risk to a Manageable One, but Still Limits Adoption

In early CD19 CAR-T, cytokine release syndrome was extremely severe. 刘诚 says roughly 15%-20% of patients needed the emergency room; without rescue treatment, some could die, and a small number of deaths did occur in early clinical trials.
The first US products therefore carried black-box warnings and required infusion sites to be within about 15 minutes of an emergency room. That constraint kept treatment concentrated around large medical centers and continues to limit adoption, even though clinicians are now better at managing high fever, chills, and confusion.
Prevention and clinical management have improved substantially over the past 7-8 years, sharply reducing the risk of death. But 刘诚 believes current measures treat the symptoms rather than the cause. Traditional CAR structures fuse recognition and activation, effectively “welding the accelerator to the steering wheel”: once the target is seen, the accelerator stays down with no brake.
Continuous activation causes T cells to release large amounts of cytokines. 刘诚 advocates separating the recognition and activation systems at the molecular level and using the T cell’s own regulatory capacity to prevent overactivation; he describes this as the problem Eureka’s technology platform is trying to solve at the source.

15. China’s R&D Flywheel Is Fast, but the Endgame Still Depends on Solid Tumors and Control

The host cited industry data showing more than 250 CAR-T companies globally and more than 500 development programs, with several hundred clinical projects in China alone. 刘诚 believes China started small-molecule drug development roughly 100 years behind the US and biologics about 20 years behind, but began CAR-T at nearly the same time; as a result, China and the US lead the field by a wide margin, with the third-place player far behind.
China’s speed of feedback comes from the combination of policy, patients, and hospital structure. IITs can enter the clinic relatively quickly, the large patient population creates a “patient dividend,” and cancer patients are heavily concentrated in top-tier hospitals in Beijing, Shanghai, Guangzhou, Shenzhen, and provincial capitals. A program can quickly determine whether to stop because it is poor or advance because it is good, producing feedback several times faster than in the US.
刘诚 views in vivo CAR-T as an engineering application: once the direction and metrics are clear, China is good at turning frontier technology into something “as cheap as cabbage.” He compares the process with Tesla opening the EV market and China later accelerating past it. But engineering intensity also creates a “problem of the century”: when technology is managed as technology, standards vary across hospitals and quality becomes difficult to reproduce and regulate.
The host noted that stem-cell and anti-aging programs could also be packaged as IITs, creating a risk of gray-market abuse of innovation channels. After the 魏则西 incident, regulators tightened the balance, which has since shifted back toward support. 刘诚 said “Order 818” (as described on the show) adopts a two-track compromise: a single hospital may regulate a project as technology, but national commercialization requires drug-style regulation. He believes the US single-track approach is too restrictive, driving costs ever higher.
Across the broader cancer landscape, 刘诚 likens the human body to a society of roughly 40T cells—equivalent to “5,000 Earth populations.” Cancer cells “do not work,” consume resources, proliferate without limit, and metastasize, eventually squeezing out organ function. Chemotherapy distinguishes friend from foe roughly by division speed; targeted drugs seek specific molecules; ADCs use antibodies to carry toxins; surgery and radiation remove concentrated disease sites.
He still calls CAR-T “the best weapon,” because immune cells are already the body’s police and lack only the ability to identify the bad guys. But more than 90% of patients have solid tumors, so the next phase must solve infiltration and TME suppression. In vivo and ex vivo approaches will coexist: the former will provide scale and lower costs, while the latter will handle the hardest cases through tighter dose and engineering control.
“Within 20 years” is 刘诚’s personal forecast for cancer becoming curable, but his definition does not require eliminating every cancer cell. It means that “most people should not die from cancer,” with the end state being a diagnosis that prompts “what a hassle, I need to see a doctor,” while the patient continues living normally and coexisting with the disease rather than treating cancer as a death sentence.

Cliff Weitzman: What I Learned from 100 of the World’s Top CEOs & Why Tokens Will Outspend Salaries Elon’s Anthropic Deal, The Next AI Monopoly?, “FDA for AI” Panic, Trading the AI Boom