From Google to bit tuples, every big company will eventually have a pharmaceutical factory

Five platforms: AnewFold does biomolecule structure prediction, and AnewSampling, AnewOmni, AnewDesign, and AnewMind cover other aspects of molecular design. International media reports mention Protenix and PXDesign among them. People familiar with the open source community are familiar with Protenix. Byte’s previously open source protein structure prediction model was once regarded as a benchmark for AlphaFold3. The company’s technical infrastructure is a continuation of Byte’s public performance in AI for Science in recent years.

Four pipelines: covering small molecules, antibodies and cyclic peptides. The one that attracts the most attention is what the company calls the world’s first: a small molecule inhibitor that can simultaneously target all three dimers of the IL-17 family.

This line needs explanation:

IL-17 is one of the most successful target families in the field of immunity. Novartis’s Cosentyx and Eli Lilly’s Tuzi, with annual sales of billions of dollars, are the main treatments for psoriasis and ankylosing spondylitis. There is only one huge regret in this market: all existing drugs are biological agents and must be injected; as for the three dimers of the IL-17 family (A/A, A/F, F/F), the strongest dual-target antibodies currently only cover two.

Using small molecules to capture three types of molecules at the same time, why has no one done it? Because the target itself is a restricted area. The binding surface of IL-17 to the receptor is large and flat, and it is a notoriously undruggable protein-interacting antibody. Such a big thing can open its arms to embrace an entire plane, but a small molecule has to be like a nail, wedged into a light wall.

If it is done, the situation will be different: patients will switch from lifelong injections to one pill a day, no cold chain is needed, and manufacturing costs will drop significantly. Anew’s head of biology publicly reported on the pipeline at an immunology conference in Boston earlier this year.

Another IL-4R is also worth talking about: the core target of atopic dermatitis is also a tens of billions of dollars market, and there is also room for oral administration.

Of course, to clarify this company, there is another sentence that must be remembered: none of the four pipelines has entered the clinic.

02The end of the big factory is the pharmaceutical factory

Anew is not an isolated case.

To roll out the global AI pharmaceutical landscape in 2026, several forces are in place almost at the same time.

NVIDIA provides computing power. In January, it and Eli Lilly announced a joint AI innovation laboratory, investing more than one billion US dollars in five years. The goal is to complete a round of dry and wet closed-loop iterations in two hours. Traditional medicinal chemists do a round of design, synthesis, and testing on a weekly basis.

OpenAI and Anthropic released models. OpenAI released the life science-specific model GPT-Rosalind in April. Amgen and Moderna have introduced the research and development process, and Novo Nordisk immediately announced a strategic cooperation. Anthropic CEO Amodei publicly stated that life science is one of the company’s highest strategic priority areas; a more radical move was also made in April, when it acquired Coefficient Bio, which was only eight months old, for about US$400 million, shifting from selling tools to directly holding pipelines. Bristol-Myers Squibb is promoting Claude to more than 30,000 employees.

Anew’s most direct benchmark is Google’s Isomorphic Labs. This company, a spin-off of DeepMind and run by Nobel Prize winner Hassabis, completed a US$2.1 billion Series B financing on May 12. Thrive Capital led the investment, and Alphabet, GV, Temasek, and the British Sovereign AI Fund all participated, with a total financing of approximately US$2.6 billion. Its engine IsoDDE covers protein-ligand modeling, affinity prediction, and pocket identification, and has signed multi-billion-dollar cooperation with Johnson & Johnson, Eli Lilly, and Novartis. It has set a goal for itself to allow AI-designed drugs to enter human trials by the end of 2026, but this timetable has been postponed once from the end of 2025.

Money is also flowing into transactions. The total global pharmaceutical transaction volume in the first quarter of this year was US$88 billion, a year-on-year increase of 30%, and institutions predict that the impact for the whole year will be US$150 billion. AstraZeneca and CSPC $18.5 billion, Eli Lilly and Innovent $8.85 billion, and Eli Lilly and Profluent up to $2.25 billion. Big pharmaceutical companies are using real money to lock AI and source innovation into their pipelines.

Look at Byte’s moves on this chessboard: the computing power is connected to its own Volcano engine, the model is connected to its own AI4S accumulation, one end of the capital is top VC, the other end is state-owned assets and pharmaceutical companies, and the pipeline is directly aligned with global targets.

Anew Labs, its ambitions can no longer be hidden.

03 US$1.5 billion, and 56% equity

Let’s first answer the most concerning question in the market: How can a company with 50 people and zero clinical trials be worth US$1.5 billion?

According to the traditional pipeline valuation method, each preclinical pipeline has a residual value of tens to 200 million U.S. dollars, with four pipelines worth as much as 500 million U.S. dollars. The extra billion is priced in something else by the market.

The first is the possibility of byte capacity overflow. Anew essentially transfers Byte’s model training, engineering and computing power scheduling capabilities accumulated in the recommendation system era to a new battlefield. This thing makes sense mathematically: the core of the recommendation algorithm is to quickly search for the optimal solution in a very large continuous solution space, and the mathematical nature of drug molecule design is similar to this. What investors are betting on is that the machine trained in the short video can also be used with protein corpus.

The second thing is the machine that produces the pipeline, not the pipeline itself. Cathay Haitong has a judgment that can serve as an anchor: the competitive unit of AI pharmaceuticals is being upgraded from a single molecule to a system capability of data × model × pipeline. Molecules will fail, pipelines will not. Isomorphic could raise $2.1 billion, while a biotech with a specific drug may raise less than a fraction. That is, the market has decided to price the factory, not the product.

The third thing is time. The last wave of Chinese AI pharmaceutical companies was established in batches from 2018 to 2021, and most of them are still stuck in the positioning of AI as a tool or a drug. The angel round only offers 1.5 billion. It is because the primary market is rushing ahead and waiting for the clinical data to come out. The valuation may go to a big level, or it may return to zero. Options are inherently two-way.

What is more worthy of consideration than US$1.5 billion is 56%.

Byte does not fully own it, nor does it sell it outright. As mentioned before about the issue of wholly owned shares, the conflict between the Double Ten Law and the rhythm of the Internet is almost unsolvable; selling out completely is equivalent to giving up a strategic card, and may also create future opponents. A holding split is a balance between the two: the parent company retains control and financial upside, and the new company gets independent capital channels, an independent compensation system, and the rhythm of the pharmaceutical industry.

This style of play is becoming a global paradigm. Alphabet split but was deeply involved in Isomorphic; Anthropic chose to acquire Coefficient Bio directly; Nvidia did not hold any shares and only formed alliances. The weight of the bet itself is a measure of each company’s confidence. It can be seen that Byte and Google are betting the most: providing money, computing power, and controlling shares.

For China’s major Internet companies, the spillover value of this matter may exceed the financing itself. The AI ​​teams within large factories that make materials, batteries, and weather systems now have a path to follow: separate them, connect them to the parent computing power, introduce industrial capital, and grow at the pace of the industry. It can be expected that the strategic departments of many major manufacturers will study this 56% equity structure.

04The bottleneck has changed its position

Where is the next bottleneck for AI pharmaceuticals? Industrial Securities has a judgment that illustrates the most counter-intuitive point in this industry: the decisive step is not how fast the model runs, but how quickly the wet experimental data is generated.

There is an arms race on this side of the model. AlphaFold has reduced the marginal cost of structure prediction to almost zero, and the generative model spits out millions of candidate molecules every day. But when the model output speed exceeds experimental verification by 10,000 times, the bottleneck completely shifts: you can design 100 million molecules, but the laboratory can only synthesize 50 in a week.

This also explains the switch of the narrative protagonist from an algorithm company to a company with wet experiment capabilities. The selling point of Nvidia and Eli Lilly’s laboratories is that iterations last two hours, which is not faster than the model, but faster in closed-loop rotation; the capital premium of Jingtai is based on 300 robots and 200,000 pieces of own experimental data per month.

For Anew, this is the most important piece of its territory that needs to be filled. According to public information, it is a company that places emphasis on calculations and light on experiments. All five platforms are dry experiments, and its wet experiment capabilities have not been disclosed. There are just a few options: self-build, cooperate, or outsource to a Chinese CRO.

In turn, this just explains that the experimental system, clinical resources and registration experience of the pharmaceutical company intended for this strategic investment by China Biopharmaceuticals are the half that Anew lacks. This is not a financial investment, it is complementary.

There is another layer of relationship that pushes this matter deeper.

I have written before that global payers are closing in (see “The End of Single Drug Economics” for details): the United States uses MFN to anchor drug prices to the lowest international prices, and China uses dual catalogs of basic medical insurance plus commercial insurance to divert high-value drugs. The ceiling of single drug unit price is institutionally lowered, and excess profits can only come from more patients multiplied by longer duration.

Put Anew’s pipeline into this framework, and you will see a perfect coincidence. There are tens of millions of patients with autoimmune diseases, which means more patients; life-long medication is required, which means it lasts longer; small molecules are taken orally, no cold chain is needed, and the manufacturing cost is much lower than antibodies, which is unit price-friendly. The same account is calculated in reverse, which is almost fatal for biological agents: once oral small molecules achieve equivalent efficacy to Cosentyx, those injections that cost tens of thousands of dollars per year will face patent cliffs and dosage form replacement at the same time.

AI reduces R&D costs, while payers suppress sales prices. The two forces are working from both ends of the industrial chain at the same time, sandwiching the old model of high unit price, small population, and biologics in the middle. Anew’s selected target flavors, IL-17 and IL-4R, are all oral. The large market envisioned shows that it knows the final rules better than anyone else.

The economic form of pharmaceuticals is approaching that of consumer goods: large market, low unit price, and long cycle. Whoever accepts this setting first will have his or her pipeline model updated first.

05The box has not been opened yet

Of course, there are still three questions for which there are currently no answers.

One is clinical. Anew’s four pipelines are all in pre-clinical development, and Isomorphic has postponed its human trial schedule. So far, no drug designed by AI from start to finish has completed Phase 3 in the industry. Cathay Haitong said that a wave of clinical verification of AI-designed drugs is coming. The word “imminent” currently carries a valuation of hundreds of billions of dollars. From 2027 to 2029, these companies will line up to submit their papers, which is the coming-of-age ceremony for this track.

The second is Byte’s patience. This company has been aggressive on multiple fronts in its history and has also contracted many times, including in education and games. The 56% holding could either be a long-term commitment or one of the first assets to be liquidated during the next strategic retrenchment. To judge the fate of a company, the ownership structure has the final say and how long the parent company is willing to give it has the final say. The answer is not in the announcement.

The third is the convergence speed of valuation. The market now pays according to the platform, but the value of the platform ultimately depends on the pipeline. If the platform cannot produce molecules that enter the clinic within three to five years, the platform valuation will collapse towards the pipeline valuation. Schrödinger’s market value has gone from US$8 billion to US$1.5 billion, which is a record of the platform narrative being modified by the reality of the pipeline.

Anew now has Schrödinger’s market value. No one wants it to follow Schrödinger’s curve.

These three questions can actually be combined into one: when will the box be opened?

In quantum mechanics, observation causes the superposition state to collapse. The cat is forced to make a choice between life and death at the moment he opens the box. The observation moment of AI pharmaceuticals is the readout of the first batch of clinical data. Before that, both valuations, $1.5 billion and zero, were legal; after that, each company could only take away its own one.

06結語

Back to the end of 2020, the winter when Byte AI Lab posted the first batch of AI pharmaceutical recruitment notices. Few people would have thought that the company that made a lot of money every day by relying on information flow recommendations would in five years hatch a pharmaceutical research company that models proteins.

The arithmetic of the capital market has always been rough: fifty people, four preclinical pipelines, US$1.5 billion. But behind this number, there are several things happening at the same time. Internet giants have found a second battlefield for AI capabilities, China’s innovative drugs have received a new capital narrative, payers have rewritten the economic form of drugs, and between large manufacturers and the Double Ten Law, there is finally an organizational form that can coexist.

There is a sentence on the home page of Anew’s official website that is so simple that it is almost naive: develop new drugs to bring tangible benefits to patients.

AI can write code, make PPT, and predict proteins. Only this sentence can be answered with real medicine.

From 2027 to 2029, the boxes will be opened one after another. By then, all companies that price based on platform will turn back to companies that price based on pipeline.

Related Reading

• Original BYD’s world has changed2026-09-22
• Wisdom spectrum, I was taught a lesson2026-09-22
• The tax paid in five years is about 53.6 billion! Jen-Hsun Huang: I’m not afraid of paying taxes, I’m just afraid of being poor2026-09-22
• AI is creating an era of app glut2026-09-22
• Ultraman failed! OpenAI CEO ranks among the top 50 richest people in the United States with $25.5 billion2026-09-21