Last year, Bristol Myers Squibb put a spotlight on the kind of cash that comes with pharma’s conviction in a new oncology strategy.
The drugmaker agreed to pay BioNTech $1.5 billion upfront and another $2 billion in noncontingent anniversary payments to jointly develop and commercialize pumitamig, a bispecific antibody that targets PD-L1 and VEGF-A. Another $7.6 billion could follow if development, regulatory and commercial milestones are met, making the partnership a potential $11.1 billion wager on a drug that is still investigational.
But that does not make the wager irrational. Pumitamig is part of a broader wave of medicines designed to combine checkpoint inhibition with a blockade of VEGF signaling, which helps tumors build blood vessels and shape an immunosuppressive microenvironment. The approach is accumulating substantial clinical evidence.
It’s also a useful example of a broader tension that’s always run through drug R&D but may now be accelerating alongside the pace of science: Companies often have to commit enormous amounts of capital before the evidence is mature enough to provide anything resembling certainty.
The same tension is playing out in B7-H3 antibody-drug conjugates and in vivo CAR-T, two other fields drawing major pharma investment at very different points on the proof curve.
“It is important to separate the biological rationale for the target from the clinical evidence for an individual ADC.”
Eric Richards
SVP, head of oncology product leadership, GSK
The crucial question for pharma isn’t whether any of these biological approaches are “real” or “hype.” It is what they’ve actually demonstrated in patients — and how much confidence the available evidence can reasonably support.
Ultimately, companies are “paying promise prices on premise evidence,” Mike Rea, CEO of a new venture focused on drug R&D decision-making called Then, If…, said in an email.
A rational early bet, he said, should be able to identify the premise that has to hold for the price to make sense, specify how and when that premise will be tested and define what result would make the buyer walk away.
“Hype is when the premise is inferred from the fact that other people have bought it,” Rea wrote.
PD-(L)1/VEGF: The evidence is catching up
Few oncology ideas have moved as quickly from intriguing clinical signal to strategic priority as PD-L1/VEGF bispecifics.
The strongest evidence for the strategy so far comes from ivonescimab, developed by Akeso and licensed to Summit Therapeutics in several major markets. In the China-only phase 3 HARMONi-2 trial, ivonescimab beat pembrolizumab, aka Merck’s star immunotherapy Keytruda, on both progression-free and overall survival in previously untreated PD-L1-positive advanced non-small cell lung cancer, cutting the risk of progression or death by 49% and the risk of death by 27%.
But the global picture is more complicated.
Summit’s multiregional HARMONi trial studied ivonescimab in a different setting: EGFR-mutated non-small cell lung cancer after progression on a third-generation EGFR inhibitor, with both arms also receiving chemotherapy. The trial met its progression-free survival endpoint, reducing the risk of progression or death by 48%. At the prespecified primary overall survival analysis, however, the hazard ratio was 0.79 and the result narrowly missed the trial’s threshold for statistical significance.
Longer follow-up has continued to show an overall survival trend, including similar hazard ratios in Western and Asian patients, but those later analyses do not retroactively change the statistical outcome of the prespecified primary analysis.
That distinction matters because saying a PD-L1/VEGF “works” can mean different things: that a particular molecule works in a particular population; that the biological strategy can outperform established therapy; or that the entire class will repeatedly improve outcomes across geographies, tumor types and treatment settings.
Pumitamig adds another piece to the puzzle. Interim global phase 2 data from 40 response-evaluable patients presented this year showed encouraging response rates when the drug was combined with chemotherapy in first-line non-small cell lung cancer, including in patients with low or negative PD-L1 expression. BioNTech and Bristol Myers have already moved the drug into a broad global phase 3 program.
But response rates from an interim phase 2 cohort without a standard-of-care comparator are not the same thing as randomized survival evidence.
The field has therefore moved well beyond pure speculation. The next proof point is whether multiple PD-L1/VEGF drugs can repeatedly beat modern standards of care in global randomized trials — and whether those wins hold across populations rather than clustering around a small number of unusually successful assets.
Summit, in written responses to PharmaVoice, argued that the evidence base is already broader than any single trial. The company said five randomized phase 3 ivonescimab studies have read out across its program and Akeso’s China studies, four with statistically significant improvements in overall survival; HARMONi later showed a nominal survival improvement with longer follow-up.
More than 5,000 patients have been treated across Summit- or Akeso-sponsored studies, the company said. Summit also pointed to updated HARMONi data showing similar survival effects in Western, Asian and global populations and said the program provides “clinical validation” that simultaneous PD-1 and VEGF blockade can improve outcomes over established comparators in certain settings.
But the company also outlined the remaining proof gap. Ivonescimab is not yet approved by the FDA or regulators in other markets where Summit holds rights, and Summit said the next phase of validation will come from global first-line phase 3 studies in lung, colorectal and bladder cancers. Those trials will test how well the strategy reproduces its strongest signals across broader populations, treatment settings and standards of care.
B7-H3 ADCs: Two phase 3 wins change the argument
B7-H3 ADCs may be closer to a different inflection point: the moment when repeated late-stage wins begin to validate the target itself.
Two separate B7-H3-directed ADCs demonstrated phase 3 survival benefits in relapsed small cell lung cancer this year, both in China.
GSK’s risvutatug rezetecan, or Ris-Rez, reduced the risk of death by 54% compared with the chemotherapy topotecan in a pivotal phase 3 trial. Median overall survival was 18.5 months with Ris-Rez versus 10.3 months with topotecan.
Roche partner MediLink also reported a 54% reduction in the risk of death with tambotatug pelitecan, or Tam-Peli, in the phase 3 study. Median overall survival was 13.3 months versus 9.4 months for topotecan, while Tam-Peli reduced the risk of progression by 71%.
One successful drug can validate a molecule. When two independently developed drugs aimed at the same target succeed in randomized phase 3 trials, the target-level argument becomes harder to dismiss.
Pharma has already paid to be part of that argument. GSK licensed Ris-Rez from Hansoh Pharma for $185 million upfront and up to $1.5 billion in potential milestone payments. Roche’s 2026 deal for global rights to Tam-Peli outside mainland China, Hong Kong and Macau included $570 million in upfront and near-term payments, with additional milestones and royalties undisclosed.
“Price stands in for proof, and the more deals there are, the less anyone remembers that no new evidence has been added.”
Mike Rea
Pharma strategist, CEO of Then, If…
Still, B7-H3 is not the only thing those drugs have in common. Both Ris-Rez and Tam-Peli also use topoisomerase inhibitor payloads. That makes “target validation” a more nuanced claim than simply counting wins. The phase 3 results provide powerful evidence that B7-H3 can be exploited therapeutically in small cell lung cancer. They do not establish that every B7-H3 ADC — or every engineering approach built around the target — will work.
Additionally, the pivotal survival data for both Ris-Rez and Tam-Peli were generated in China, leaving global reproducibility as an important next test.
Eric Richards, GSK’s senior vice president and head of oncology product leadership, called the results “strong evidence for Ris-Rez in relapsed SCLC” and said they increase confidence in B7-H3 as a clinically actionable target. But he stressed that an ADC’s performance reflects the target, antibody, linker, payload and drug-antibody ratio together — not the target alone.
GSK is now testing whether the China data reproduce globally in the phase 3 Embold SCLC-301 trial and a broader solid-tumor program. The target-level case is materially stronger than it was a year ago, but global reproducibility and broader class validation remain unresolved.
Rea offered a similar test for when an individual drug begins to validate a broader thesis.
“One drug becomes evidence for a broader thesis when the confound is removed,” he wrote, for example, when a structurally different molecule repeats the effect or a result holds in a different population against a different comparator.
By that standard, the case for B7-H3 has strengthened materially, but shared payload characteristics and China-only pivotal data still leave important confounds unresolved.
In vivo CAR-T: The money is far ahead of the dataset
The proof curve looks very different for in vivo CAR-T.
Traditional autologous CAR-T requires a patient’s immune cells to be removed, engineered outside the body, expanded and reinfused. In vivo approaches aim to turn the patient’s body into the manufacturing site by delivering genetic instructions that reprogram T cells directly.
If that works reliably, it could strip away some of CAR-T’s most stubborn logistical and manufacturing barriers. Pharma is spending as though that possibility matters.
In June, Lilly completed its acquisition of Kelonia Therapeutics in a deal worth up to $7 billion in cash, including $3.25 billion upfront. Kelonia’s lead program, KLN-1010, uses an engineered lentiviral vector to generate anti-BCMA CAR-T cells inside patients with multiple myeloma.
The clinical dataset behind that deal was real — and tiny.
Updated phase 1 results presented at ASCO this year covered six patients with heavily pretreated relapsed or refractory multiple myeloma. All achieved minimal residual disease negativity in bone marrow at one month and all had a clinical response. CAR-T cells expanded in the blood and remained detectable for months.
Those results are striking for a first-in-human study. They are also six patients in an uncontrolled dose-escalation trial, with the longest reported follow-up measured in months rather than years.
That is enough to demonstrate that in vivo generation of functional CAR-T cells in humans is possible and can produce antitumor activity. But it’s not enough to establish durability, comparative efficacy, reproducibility, rare safety risks or whether the platform can consistently match — or improve upon — established ex vivo CAR-T therapies.
The modality also isn’t a single technology. Companies are pursuing viral vectors, lipid nanoparticles and other delivery approaches, with different genetic payloads and distinct questions around targeting, persistence, control and off-target effects.
The strategic race is nonetheless accelerating. Johnson & Johnson struck a collaboration with Sail Biomedicines in July to develop in vivo CAR-T programs for immune-mediated disease and received an exclusive option to acquire the company for $2.58 billion.
That creates the sharpest gap of the three fields examined here between strategic conviction and definitive proof.
The early human data show that the central biological premise is possible. They do not yet show that in vivo CAR-T can reliably reproduce the efficacy and durability of established CAR-Ts while delivering the simpler manufacturing, broader access and safety advantages that make the approach commercially transformative.
Conviction is not validation
Betting before certainty is not the problem; drug development requires it. The risk comes when the size of the bet begins to masquerade as evidence that the underlying premise has already been proved.
“A deal is a price, not a data point,” Rea wrote.
Yet deals can begin to function like evidence anyway. Internally, the deal thesis becomes the development plan while externally, one company’s price becomes validation for the next bidder.
“Price stands in for proof,” he said, “and the more deals there are, the less anyone remembers that no new evidence has been added.”
