Within two days, players from home and abroad placed consecutive bets on the "AI+macromolecule" gambling table.
On June 15th, China. Platinum Pharmaceuticals and Baitu Biotechnology jointly announced the establishment of MegaStream TechBio, focusing on the research and development of complex macromolecular AI drugs.

On June 16th, in the United States. Protillion Biosciences, an AI protein design company, announced the signing of a multi-target discovery cooperation agreement with Merck, with a potential total value of up to 510 million US dollars. The core will utilize Protillion's lab in the loop AI platform to discover new generation drug candidates and focus on tackling complex protein engineering challenges.

Two things, two places, two modes of cooperation. But pointing in the same direction: AI+macromolecules.

In recent years, the application of AI in the pharmaceutical field has continued to expand, and AI has achieved deep penetration and practical application in core areas such as target discovery, molecular screening, and structural optimization. However, it should be clarified that AI pharmaceuticals have been focused on small molecule drugs for so many years, with relatively less research on large molecule drugs.

This round of joint bets at home and abroad may conceal the overall trend of the industry: the main battlefield of AI pharmaceuticals is shifting from the traditional small molecule track to the more difficult and valuable large molecule track.
 

01、Why is it a 'hard bone'?

For a long time, traditional innovative drugs have generally faced pain points such as high research and development costs, long cycles, and extremely low success rates.

With the penetration of AI technology into the entire pharmaceutical process, the efficiency of core processes such as target discovery, molecular screening, and structural optimization has been greatly improved. Empowering the pharmaceutical industry with AI has become a consensus in the global biopharmaceutical industry.

However, the industry has been implemented for many years, and mature applications have always been concentrated in the field of small molecules. The development of AI+large molecules has been slow and lacks benchmark results. The core bottleneck lies in the natural threshold of two dimensions.

One is the huge gap in data volume. Data is the 'fuel' for AI model training, and only with sufficient data can AI learn accurately and iterate efficiently.

CMB International mentioned in its research report that small molecule drugs have accumulated decades of public databases, with ChEMBL containing over 2 million compound activity data and ZINC containing billions of virtual compound structures.

In contrast, the International Protein Structure Database (PDB) only accumulates and stores over 250000 experimental structural data (including protein, nucleic acid, and other structural data), while SAbDab contains 10673 antibody structural data and 735 antibody affinity data, which is significantly different from the data size of small molecules.
The second is the ultra-high complexity of the molecule itself. Compared to small molecules with simple structures and fewer parameter dimensions, the difficulty of developing large molecules has increased exponentially, placing extreme demands on AI model capabilities.

Wei Yuan, the business leader of Baitu Shengke Molecular Discovery, further explained to Xieyi Jun that compared to small molecules, the difficulty of large molecules lies in their "sufficient complexity". Their sequence space is larger, their structure is more complex, and their development parameters are more multidimensional. They require higher model generation and multi-objective optimization capabilities.

Wei Yuan further gave an example that for a protein's amino acid sequence, even if there are only 1000 amino acid letters, its arrangement and combination are 20 to the power of 1000. It is impossible to rely on exhaustive experiments or traditional models to make point by point predictions. It is precisely this huge state space that provides a stage for generative AI to achieve "long slopes and thick snow".

The pain point of this industry has also been confirmed by Hebao Pharmaceuticals, who pointed out to Xieyi Jun that "macromolecular drugs not only have complex structures and almost unlimited design space, but also have extremely high requirements for core indicators such as in vivo activity, drug properties, and safety. Traditional research and development methods are inefficient

From this perspective, AI+macromolecules are not an industry with no one to lay out, but rather a recognized pharmaceutical backbone with high data barriers, technical difficulties, and research and development barriers.
According to the press release, MegaStream will focus on major unmet clinical needs such as cardiovascular, renal, anti-aging, and oncology, and develop macromolecular drugs.  

02、Same bet, two paths

Faced with the same opportunity, the two Chinese and American companies chose different ways to break through the situation.

Protillion chose to license its data platform to large pharmaceutical companies, which is a more traditional and easier to understand approach, exchanging BD authorization for down payment and milestone revenue.

Protillion was co founded by CEO Curtis Layton and Stanford University professor Will Greenleaf, dedicated to revolutionizing the discovery and development of next-generation therapeutic antibodies, with top tier venture capital firms ARCH Venture Partners and Illumina Ventures (a subsidiary of Innolux) as the backing investors.

Protillion's core technology, Prot MAP, is a large-scale protein data generation platform. Through fully automated instruments, millions of protein variants can be characterized in each run, providing a large-scale, real-time training set specifically designed for protein design AI - which is precisely what is most scarce in macromolecular AI.

The potential milestone total for this transaction with Merck is as high as $510 million. Juan Alvarez, Vice President of Discovery Biopharmaceuticals at Merck Research Laboratory, stated that Protillion's platform provides an attractive opportunity to characterize protein landscapes with unprecedented speed and accuracy.

Ultimately, the essence of the Protillion model is' data-as-a-service '. Solve the fundamental problem of scarce training sets for large molecule AI, and then empower this ability. This path is clear and quick to monetize. The heavy assets and long cycles of clinical development are entrusted to Merck.

Let's take a look at MegaStream, which has directly joined forces with Platinum Pharmaceuticals and Baitu Biotechnology to build a new entity, deeply binding and integrating the core resources of both parties.
As the "king of BD" in China's biopharmaceutical field, this time, Hebao Pharmaceutical did not choose the path it is familiar with.

Looking at the development of He Bo Pharmaceutical, in recent years, it has participated in the establishment of new companies more than once, such as Nona Biotech, É lanc é Therapeutics, etc., focusing on different industry directions.

Regarding this, the explanation from Platinum Pharmaceuticals stated that "internal self research is linear, while joint ventures can integrate top AI computing power and algorithm teams from outside, forming exponential synergies. We choose to quickly gather the world's best resources through joint ventures, rather than building a team from scratch, in order to seize the time window and the high ground of macromolecular AI design. ”

Simply put, a strong alliance is necessary to run faster and quickly seize the track advantage. And this joint venture to establish a new entity of "AI+macromolecule" also hides two major industry trends behind it.

Firstly, AI+macromolecules have become the core trend in both the capital market and the industrial sector. After the successful listing of "AI+macromolecules first stock" Yitai Technology, the track value has been verified by the capital market, and the model of independent new entities is more suitable for capitalization and industrialization development. Secondly, the strategic logic of large AI pharmaceutical companies is being fine tuned.

For Baitu Biotechnology, as a life science AI big model company that emerged from Baidu, up to now, this company does not have its own research pipeline. And this partnership with Platinum is its first participation in an AI pharmaceutical company with pipelines, which has also led to external speculation about the strategic adjustment of Mobike. After all, a few days ago, ByteDance, a large manufacturer, announced to split the AI pharmaceutical business and carry out independent financing.

Regarding cooperation, Wei Yuan explained that "Baitu Shengke's positioning is still as an AI technology platform enabler. In the collaboration with MegaStream, Baitu Biotech's role is to provide underlying AI technology empowerment, model engineering support, and intelligent research and development capabilities. The pipeline research and development of the new company is independently promoted by MegaStream, and Baitu Shengke does not participate in the direct research and operation of the pipeline. This is still an extension of platform empowerment, rather than a strategic shift. ” 

03、AI Pharmaceutical rushes forward

Two things happened one after another, not by chance in time.

They no longer only point towards the subdivision direction of AI+macromolecules. But the entire AI pharmaceutical industry is accelerating and moving forward.

In the past few years, cooperation in the field of AI pharmaceuticals has mostly remained relatively independent.
In January 2026, Eli Lilly and Nvidia reached a $1 billion partnership to jointly establish an AI Joint Innovation Laboratory. In March, they launched the LillyPod supercomputer, which is equipped with 1016 Blackwell GPUs and is currently the most powerful supercomputer in the pharmaceutical industry. This type of cooperation is essentially "borrowing computing power", where pharmaceutical companies control their own data and pipelines, while computing companies provide underlying computing capabilities.

There is also an algorithm authorization mode. Pharmaceutical companies entrust their targets or pipelines to AI companies for discovery and optimization, and pay according to milestones. Yingsi Intelligent and Schweya, Jingtai Technology and DoveTree are all following this path. This is the safest and easiest way to replicate.

The commonality of these modes is that the participants are relatively independent and the boundaries are clear, but now the forms of cooperation are becoming more complex and in-depth.

With Platinum Pharmaceuticals and Baitu Biotechnology, we have simply jumped out of the authorization framework and directly built a new company, locking data and models under the same roof, sharing risks and benefits in the long run.

Moreover, Wei Yuan bluntly stated to Xieyi Jun that "Baitu Biotech will explore and replicate similar cooperation models with other large pharmaceutical companies in the future, deeply integrating AI platform capabilities with industry resources of partners. But we will maintain high-quality selection standards - prioritizing cooperation with top enterprises with high-quality data assets, clear pipeline layout, and industrial landing capabilities, to ensure that cooperation can truly form a multiplication effect of 'data x model x pipeline'. ”

This means that this deeper collaboration may not necessarily be an isolated case. But the threshold will not be lowered because of this. Those who can replicate this model must already be top players in the industry.

From "borrowing computing power" to "authorizing algorithms", and then to "co building companies", the depth of cooperation is increasing layer by layer.

The deeper the cooperation, the finer the division of labor. Computing power providers, platform technology companies, data generation companies, traditional pharmaceutical companies, and AI native biotech are joining together in the same research and development chain in different ways. No one can run through all the links alone, and the degree of cooperation increases accordingly.

And all these explorations ultimately have to answer the same question: when will the first AI led drug, which the entire industry is waiting for, be approved? How will the results affect the value of AI pharmaceuticals?

In the face of this crucial milestone in the industry's expectations, we have admitted to Platinum Pharmaceuticals that we are indeed at the critical point of the industry's expectation for the approval of the first 'AI designed drug'. Whether the first outcome is a success or a setback, it will not change the overall trend of AI transforming the pharmaceutical industry. For MegaStream, we do not overly rely on a single landmark event, but instead rely on the wet experimental verification capability of platinum to build a high-frequency iterative loop of design synthesis verification, demonstrating value through continuous data output and milestone delivery. ” 

- Freehand summary - 

From small molecules to large molecules, AI pharmaceuticals have officially begun to tackle the more difficult and valuable industry challenges. Whoever can break through the barriers of this track first will hold the core entry ticket for industry competition in the next decade.

At present, during the window period of the track, industry players are developing their own strategies and seizing opportunities. Whether it is renting computing power, algorithm authorization, or deep joint ventures to jointly build new entities, each is relying on its own advantages to integrate resources, aggregate capabilities, and fully accelerate research and development.

Overall, the wave of AI pharmaceutical industry is surging, and the continuous iteration of technology and cooperation models will ultimately drive the dual innovation of global innovative drug completion efficiency and research and development paradigms.