AI in Bio Biggest Questions

Dr. Shelby

May 13, 2026

With this piece, we strive to ask the biggest questions in biotech today. This is mainly due to incredibly divergent world views and world building we hear in biotech every day such as: “The Food and Drug Administration (FDA) is dead in 10 years”, “No one can beat Eli Lilly”, “Anthropic will make drugs and become a pharma co”, “HIMS will become the frontend distribution for biotech”, “Frontier labs will not expand CRO TAM”, “China has won biotech”, and others.

Each of these statements is implicitly destructive or entrenching for the biotech/pharma industry today. The Russian doll of layered questions and path dependencies is something that should be explored concretely to make the assumptions, undercurrents, and wagers, explicit.

To fully understand where we are, and what this means for company building opportunities, we created a decision tree of the most consequential boom or bust forks in the road over the next 10 years of biotech. As Ada Palmer astutely pinpoints on a recent Dwarkesh podcast, history normally doesn’t ask the right questions, due to focus on small vs. global changes. We try to not fall into that by asking the largest questions possible. Specifically, we make arguments on either side of:

Neolabs in bio will become very large, independent companies

Next-gen contract research organizations (CROs) that generate data for frontier and neolabs (along with biotechs) will become very big businesses

China will become the global hub of biotech

There will be more drug approvals per year in the US

There will be a collapse of incumbent drug distribution.

Biopharma becomes hyper-deflationary in mid-term (many assets for every target) and returns diminish drastically due to competitive pricing and higher distribution/marketing costs.

The asset-centric turn in biotech VC produce a wave of orphaned, single-asset companies.

We strive to have clear, defined bear and bull cases to help builders more accurately underwrite their company’s trajectory and implicit assumptions of it.

Fork #1: Neolabs in bio will become very large, independent companies.

Context: What we still don’t know is what a bio neolab that finds PMF with scientists, biotech, the world, looks like. What we do know is that forming the best bioML teams works for mid-size acquisitions (EvoscaleAI acquired by Chan-Zuckerberg Initiative and Coefficient Bio acquired by Anthropic) OR mid 8-figure deals with pharma (Chai Discovery and Eli Lilly and NOETIK and GSK). An inherent assumption in founders and VCs continuing to back these neolabs is that they will, with a technical step change, unlock new biological understanding and crucially be able to pair that with highly valuable product market fit (PMF). Yes or no?

Yes - Step changes in biological understanding through bio-AI labs will be meaningfully more valuable to pharma, biotech, and society than they were historically. Companies with the best ML teams which will prove themselves first in open benchmarks, increased predictive power, predicting disease progression, understanding toxicity, will be able to productize quickly after the science unlock to their ideal customers and broader industry once the technological unlock happens. Not only does there have to be excellent technology and compounding scientific speed/moats but continued model development, proprietary data acquisition, and productization will be required of the team.

No - the best ML-bio teams are acquihired by the frontier labs or neolabs further along. Expect to see Meta-level comp packages to thwart competition and upstarts before they even begin. Another version of neolabs not being successful is if initial benchmark steps can’t be compounded scientifically or quickly productized to key customers to justify their valuations and burn by the companies. Neolabs experience a valley of death transitioning from a technology-first company to a product-first company. Some of the complexity is the relative market size of biopharma vs. general LLMs and build vs. buy dynamics which biopharma is rife with.

Fork #2*: Next-gen contract research organizations (CROs) will become very big businesses.

Context: CROs today are massive businesses. Lonza makes $8bn+ per year, WuXi $6bn+, Charles River makes $4bn+. These companies have been dominant in CROs for decades and have relationships and incumbency dynamics with biotech and pharma companies which have made it hard for upstarts to unseat. With new potential customers in frontier and neolabs, patients vibecoding, and those using LLMs to do experiments as hobbies, there could be room for next-generation CROs that service net new customers. The question is whether next-generation CROs could become really large, durable businesses, given structural disadvantages. Yes or no?

Yes - Next-generation CROs will better serve the new range of customers of patients, biotech hobbyists, neolabs and frontier labs. Next-generation CROs will use state-of-the-art hardware, automate every possible step of the process and work quickly to ship to customers. On one end, self-serve experiments for small orders by patients and hobbyists with no relationships with incumbents could be possible. Next-generation CROs would likely have better UI/UX and could serve these customers from the earliest days of the CRO. On the other end, serving frontier labs and neolabs has similar opportunities of building relationships with customer new to biotech. CROs that generate libraries of training data for frontier labs and for bio neolabs becoming defensible versions of Mercor and ScaleAI for bio. Think of this as Plasmidsaurus to Genewiz, but for WuXi AppTec.) If you believe that all pharma companies become AI companies, this is another tailwind for next-generation CROs, but isn’t required for them to be successful.

No - Frontier labs and bio neolabs will not pay for scaled, unique biological data and will stick to model development vs. buying proprietary data. Another version of this is that neolabs are acquihired by Frontier labs too early which concentrates customers to the point where there isn’t a thriving, cash-rich industry for next-generation CROs to sell directly into over long periods of time. This would require early pivots to biopharma before there’s a large cost and data scale differential in data generation. It will be hard to sell over Lonza, WuXi, and Charles River since incumbents have relationships and rapport with all of these folks.

**Fork #2 is Russian doll-nested inside the question of whether frontier labs (OpenAI and Anthropic) will move materially into bioAI and if neolabs become large, separate entities with 9-10 figure data spend ability.

Fork #3: China will become the global hub of biotech.

Context: The US is scared of losing global dominance in every domain. In biotech’s specific insecurity, there’s a big discussion on whether China surpassing us in clinical trials and innovation generally would lead to a collapse of the US biotech ecosystem. There’s a deeper point on whether geopolitical tensions could lead to supply chain fragility for active pharmaceutical ingredients (API), talent arbitrage forcing companies to use off-shore resources, all under the backdrop of IP theft. The Erasca and Revolution Medicines patent infringement debacle shows that there’s clear infringement for seemingly every molecule that’s made in China. The question is will China become the dominant biotech player or not?

Yes - China’s dominance on drug research and development, clinical trials, and manufacturing will only compound in the upcoming years. Their economies of scale, labor cost arbitrage, AI and robots excellence interface with scaled human studies and scaled manufacturing seamlessly. The best talent in the West increasingly studies at Chinese universities, fights for jobs to Chinese biotech companies and tenure positions, with China becoming an international magnet.

No - China’s aggressive action in the South Pacific and annexation of Taiwan reduces connectivity of China with the rest of the world, making it harder to do business there. Second to that, China has shown the rest of the world what is needed for a productive and scalable biotech industry (fast human evidence, fast innovation and manufacturing cycles, cheaper cost of goods sold (COGS)). The rest of the world aggressively copies the fast human experimentation laws, automation of bio R&D and innovation removes the relative cost savings of China vs. the West. In the intermediate term, China is used for fast human evidence but limited R&D happens there by international companies given likely risks of IP infringement.

Fork #4: There will be more drug approvals per year in the US

Context: Money is scarce to actually do clinical trials in the US (Series A+ biotech venture rounds have decreased since all-time-highs in 2021) and drug approval rates have plateaued over time. At the same time, people are now just making their own treatment pathways to cure “terminal” cancer (and their dogs’!). Adding to the backdrop is that we have more molecules to screen than ever with an ever-increasing docket of modalities and ML models broadening the search space of new compounds and libraries. Compound these increased therapeutic options, national biobank efforts have increased the amount of longitudinal, multiomic information significantly (at least 2M paired whole-genome sequence and medical records). The question is do these tailwinds translate to more drug approvals per year?

Yes - In a world with magnitudes of broadly accessible data, more tools to make drugs, better perspectives on screens, and where people are making their own drugs (Sid and Rosie), it’s inevitable that we will break out of the 50 drug approvals per year plateau. Competition from China enforces more capital efficiency in US biotech where companies pursue the path to human evidence aggressively and cost effectively. There’s a rise in Phase 0 which multiplexes lead series of compounds and biologics which improves adsorption, distribution, metabolism, excretion, and toxicity (ADMET) properties significantly while checking for strong target engagement. If the US doesn’t immediately change its regulatory processes, biotechs will do their clinical trials abroad to get efficacy readouts more cost effectively. A rise in human challenge trials for human steady state pushes the boundaries of human health to human enhancement. This means that incremental dollars spent could have higher output both on failing quickly and having more in vivo iterations. Increased amounts of drugs approved per year would likely spur more biotech investment leading to a compounding output.

No - US biotech companies don’t aggressively pursue capital efficient drug discovery. The number of drugs approved per year stays flat over the next years and decade. The slow movement dovetails with continued, prohibitive US regulation that ultimately prevents fast human evidence. Companies stay focused on high-throughput screens which have limited predictive validity. The cultural stuffiness of US biotech means they aren’t aggressive experimental modes such as Phase 0 nor capital efficient trials in other countries. The biotech oldguard (biopharmas and old biotech funds) continue to profit disproportionately and biotech continues to be a largely slow and risk-averse industry. Biotech isn’t translated to larger, aesthetic and consumer markets. In short, US swiftly and decisively loses the battle for biotech dominance to China. Any serious drug makers that are in the US have to move to China to be part of the dominant biotech ecosystem.*

*This mirrors Fork #3 significantly.

Fork #5: There will be a collapse of incumbent drug distribution.

Context: Pharmaceuticals and health insurance are some of the most hated industries in the US, to levels on par with the oil and gas industry. This disdain has had led to a lack of brand loyalt

[truncated for AI cost control]