AI and BD, the seesaw of pharmaceutical stocks and the little blue pill
Last night, the ADC deal between Hansoh and Roche went viral across WeChat Moments. The total agreement value stands at 15.5 billion USD, with a 400 million USD upfront payment, a 110 million USD first-anniversary payment, and total milestone payments exceeding 15 billion USD. Hansoh's share price once rose by more than 8% in early trading today.
The first reaction was a sense of familiarity, not excitement. The current pharmaceutical market is like a seesaw, with BD sitting on one end and AI on the other. When one side is pressed down, the other tilts up. They keep moving up and down, looking very busy, but neither of the two sides has actually stood up. What sustains the whole momentum is drug efficacy.
The advantage of the little blue pill is that it takes effect quickly, but that is also its flaw: it only solves the problem "for the moment", and if you rely on it for a long time, you have to increase the dosage.
This time last year, the market was propped up by BD.
In 2025, China's innovative drug overseas out-licensing deals reached a total value of 135.655 billion USD, with upfront payments hitting 7 billion USD, surpassing the United States for the first time and accounting for 49% of the global total License-out volume. This is a historical record, and the entire industry has been celebrating this achievement for a whole year.
This year, the baton has been passed to AI.
01 Hansoh's Progress
The 400 million USD upfront payment plus the 110 million USD first-anniversary payment totals 510 million USD in cash, accounting for 3.3% of the total agreement value. The remaining 96.7% comes from milestone payments, which will only be received step by step after the project completes development, registration and commercialization. And the two leading projects in this deal are still at the pre-clinical stage.
In October last year, the same Hansoh and the same Roche reached a deal on a CDH17 ADC, with an 80 million USD upfront payment and a total agreement value of 1.53 billion USD, where the upfront payment accounted for 5.2% of the total.
Within one year, the absolute value of the upfront payment has increased 5 times, while its proportion in the total agreement value has dropped from 5.2% to 3.3%.
This change is very meaningful. Roche is willing to pay more for the "platform", but not willing to pay more for "certainty", pushing the risks further to the later stages of the project.
The real progress lies in the transaction structure.
Last year, Roche came to buy molecules, to buy a product whose data was already visible. This year, Roche brought its own target, and let Hansoh use its own ADC platform to develop the project from scratch. Hansoh will be responsible for the work up to the global Phase Ia trial, and Roche will take over for the subsequent development. Hansoh has transformed from a "seller" to a "co-developer". This recognition of its capabilities is far more significant than the transaction amount.
Hansoh's series of operations this year is indeed systematic. In June, it out-licensed HS-20118 to Avere, with a 120 million USD upfront payment and 2.18 billion USD in milestone payments, and in return took 30% to 40% equity of the merged company. In December last year, it out-licensed almonertinib to Glenmark for more than 1 billion USD. Three deals in one year, each one more skillfully structured than the last.
02 Today's AI is Yesterday's BD
The statement of "history repeating itself" is backed by tangible data.
In the first half of 2025, a BD announcement was almost equivalent to a share price limit up. The deal between 3SBio and Pfizer, with a 1.25 billion USD upfront payment and 4.8 billion USD in milestone payments, immediately ignited the market. Later, there even appeared "preview-style BD" — as long as two pharmaceutical companies exchanged business cards at a conference, their share prices would move up.
In the second half of the year, the logic reversed.
On June 26, Rongchang out-licensed the overseas rights of telitacicept to Vor Bio, with a 45 million USD upfront payment and a total agreement value of more than 4 billion USD. On the day of the announcement, Rongchang's A-share price fell by 18.36%, its H-share price fell by 11.7%, while Vor Bio's share price rose by 60%.
On October 22, Innovent and Takeda reached a deal with a 1.2 billion USD upfront payment and a maximum of 10.2 billion USD in milestone payments. After the news was released, the share price did not rise but fell.
On December 21, after the pan-KRAS deal between Jaguar and AstraZeneca was officially announced, the share price fell by more than 13% on the first trading day.
For record-breaking transactions, the market sentiment shifted from "rising" to "falling" in only half a year. The market completed its first round of education: investors first look at the total amount, then the upfront payment, then check whether the upfront payment is enough to cover clinical expenses, and finally look at the vague areas of "separate agreements between the two parties" in the terms.
After analyzing all the terms, everyone found a very simple truth: the amount of money that can be confirmed is only so little.
Then the share prices returned to the starting point. This is the market fatigue.
The figures this year are even more staggering. As of October 17, there have been 209 out-licensing deals, with upfront payments exceeding 8 billion USD and a total transaction value of over 1400 billion USD, all surpassing the full-year figures of last year. Among the top 10 global BD deals in the first half of the year, Chinese pharmaceutical companies accounted for 8 seats. The deal between CSPC and AstraZeneca reached 18.5 billion USD, and the deal between Hengrui and BMS reached 15.2 billion USD.
The proportion of upfront payments is also rising, from just over 5% to around 6%. This is a sign of progress, indicating that Chinese assets no longer "take whatever they are given" at the negotiation table.
On the other hand, 94% of the total amount is still tied to conditional "if" clauses.
The quality of upfront payments is also being tested. Rongchang's revenue in the first half of the year was 5.853 billion yuan, a year-on-year increase of 433%, with a net profit of 4.66 billion yuan, and the 650 million USD upfront payment contributed more than 70% of its revenue. Haisco confirmed 265 million USD in upfront payments from 4 deals, with its net profit increasing by 427%. CSPC turned losses into profits in half a year relying on the 420 million USD upfront payment from AstraZeneca.
The financial reports look good. But in the same half year, among the 508 listed pharmaceutical companies on the A-share market, more than 240 saw their performance decline year-on-year.
By October 14, the A-share innovative drug index fell by 5.43% within the month, and the Hong Kong stock innovative drug index fell by 7.60%. The index hit new highs in September, but in October the market started to calculate the pullback.
03 AI is Coming
The main line of this year has shifted to AI, and the shift is more rapid than that of BD.
Up to now in 2026, there have been 57 financing deals in the AI pharmaceutical sector, involving 44 companies, with a total financing amount of 157.65 billion yuan, exceeding the sum of 2023, 2024 and 2025. Isomorphic Labs completed a 2.1 billion USD financing round in May, with its valuation surging to 15-20 billion USD. It has no mature clinical pipelines, and what investors are betting on is AlphaFold and Google's computing power.
The primary market evaluates these assets with the valuation standard for tech stocks, while the secondary market evaluates them with the standard for biotech companies. The crack lies right here: Recursion has a market cap of 2 billion USD, Metis has a market cap of 2.7 billion USD, and Insilico Medicine has a market cap of 4.8 billion USD. For the same batch of assets, the two markets give several times of price difference.
The real boundary of AI application in the pharmaceutical industry has already been outlined by some people. The success rate of Phase I clinical trials has increased from the traditional 40%-65% to 80%-90%. After Phase II, there is no difference, and the success rate remains 40%.
Ren Feng from Insilico Medicine himself admitted at the Zhangjiang conference that there is currently no data proving that drugs developed by AI can improve the clinical success rate. The reason is very simple: the verification cycle is too long. Up to now, no drug discovered by AI has been approved for marketing.
Insilico's Rentosertib has entered Phase III clinical trials, which are carried out in 47 research centers, enrolling 320 patients with idiopathic pulmonary fibrosis, and the key data is expected to be read out in 2029.
In other words, the first hard evidence of AI pharmaceutical development will come as early as 2029. The current market rally is pricing for things that will happen after 2029. It is acceptable for the primary market to do so, as it is essentially betting on the next decade. But what can the secondary market use to take over these assets?
On October 19 alone, Eli Lilly and Metis signed an LNP cooperation deal with a maximum of 1.1 billion USD in milestone payments, and Hansoh and Roche signed the 15.5 billion USD deal. In September, HuaDeep and Sanofi reached a 2.56 billion USD out-licensing cooperation, Anew Labs spun off from ByteDance and completed its first round of 290 million USD financing with a valuation of 1.5 billion USD, while China Biopharmaceuticals also entered the game as a strategic investor.
The density of good news is indeed very high, and the feeling of a bubble is getting closer and closer.
04 Mr. Molecule's Concern
It took two years for BD to go from fanaticism to rationality. AI may go through the same process in just one year. The market's freshness for new stories is getting shorter and shorter, while it still takes the same number of years for a molecule to go from pre-clinical stage to marketing. This mismatch will become more and more intense, and the empty window period in between is where the share prices will be hit hard.
Converting the cash flow of the next ten years into a one-time income today, what is being given up is the possibility of becoming an independent pharmaceutical company.
The current pharmaceutical environment, or even a broader scope: the atmosphere of the entire industry is making people tired, because AI is becoming a kind of general knowledge tax.
Without mentioning AI, people can't hold meetings, write articles, talk about financing, and even have to mention "our AI layout" in earnings calls. Everyone uses the same batch of models, copies the same set of prompts, and tells the same set of words — end-to-end platform, dry-wet closed loop, AI-native. The outputs are getting more and more similar, to the point that you will doubt whether the differences between these companies lie in their molecules or in their PPTs.
The pharmaceutical industry was originally one of the industries where it is hardest to fake differentiation. The molecular structure cannot be faked, the clinical data cannot be faked, and the production quality cannot be faked. The number of new drugs that can be successfully launched can be counted on two hands. Now even this industry has begun to compete on who is better at telling stories.
There is a question that Mr. Molecule has always wanted to ask recently: before the emergence of AI, how did human civilization progress? Everyone only cares about when the next model will be released.
Then ask a more blunt question: without AI, how many people can sit down and write 2000 words of logically coherent content right now?
The stronger the tool is, the more valuable independent judgment becomes. AI can generate 2000 words in one second, but it cannot generate the answer to "whether this target is worth developing". It can give you 20 reasons to support your conclusion, and also 20 reasons to oppose it. Which one to choose is still a matter for humans to decide.
This article is from the WeChat official account "Molecular Homeostasis", written by Innovative Drug Team · Fenzi Jun, and authorized for release by 36Kr.