Why VCs Are Investing In Cheaper "Generic" LLMs For Drug Discovery
By Andy Tang, Partner, Draper Associates
If your iPhone breaks, the Apple Store will look for the root cause — perhaps the wrong software release, a defective chip, an out-of-spec manufacturing process, etc. But if you are sick, you can’t really go back to your “manufacturer.” Our authorized repair dealers (modern medicinal practitioners) do not take this deterministic root cause approach. Yet, with the advent of AI, the industry is on the cusp of moving from art to science, science to engineering, and engineering to manufacturing. In other words, diagnosing and curing your body, or finding new drugs, could be as deterministic and cheap as fixing an iPhone one day. My regular challenge to the founder community is for them to create this: a set of diagnostic and therapeutic tools that can transform healthcare across all its areas, including drug discovery, to rival the convenience, accuracy, and price point of the consumer electronics industry.
A caveat: I know this is difficult, and especially difficult when it comes to drug discovery. Like many of my fellow deep-tech VCs, I trained as an engineer before entering the investing world. As an engineer, the bane, and foundation, of our existence was always physics — a subject I found difficult until I studied biology. The forces at work in physics problems are nothing compared to the volume of atoms and molecules of a biological system (not to mention its complexity). I admire, and now invest in, the scientists who have taken it upon themselves to investigate, model, and develop solutions for these systems. It makes the relatively simple circuitry of semiconductors look easy. The high cost of experimentation and simulation in biologics is, historically, what has held back many VCs, especially early-stage ones, from investing in drug and therapeutics discovery. But that changed with the advent of AI. And it is changing again as AI computation gets less expensive. From an investing perspective, the falling cost of AI computation is going to permanently change the landscape of funding for drug discovery.
Emerging Trends In Biotech Investment
As investors, we follow trends and distribute funding based on when we think they will (and will not) happen again. For example, the hottest investment ticket a few years ago was that of an “AI doctor.” But most of these startups went out of business — not because the technology wasn’t capable, but because these companies spent considerable resources on building expensive customized LLMs, which were shortly outpaced by the generic options. Imagine if medical schools had to teach toddlers how to speak before getting into basic science, organic chemistry, and then a medical school education. Today’s AI-biotech companies are standing on the shoulders of giants. From a funding perspective, we believe the same thing is happening now with drug discovery.
The first-generation computational-based biology companies tried to solve for very difficult targets: those considered undruggable. And there has been success here, especially in the first movers. Insilico Medicine, which has quite a few preclinical candidates, was founded in 2014. Other major players include Atomwise (2012) and Verge Genomics (2015).
But when you look at 2026, it is not this model of drug discovery that will get the bulk of new funding. In fact, early-stage investors are wary now of investing in these sorts of companies because we’ve seen such an astounding lowering of computational costs. The recent cohort of successful biotech startups have proven it is actually far more effective to build upon large generic LLMS for this work, because these platforms are cheaper, more powerful, and can already demonstrably solve chemistry problems. The sentiment in investing circles is that solving complex biological problems (i.e., complex targets) is therefore not far away.
This is not to say that investment is not flowing into drug and therapeutics discovery — it is. But I expect the breakdown of companies being funded to change significantly, especially from VCs (versus the venture/R&D arms of large pharmaceutical companies). Drug discovery companies leveraging customized LLMs will see a decrease, while LLM-based research tools and simulation/experimentation-oriented biotechs will see more support. New models need to not only suggest targets, but they also need to run simulations of clinical trials using state-of-the-art AI models.
Investing In Generic LLMs
The first area of increased investment is in research tooling. Investors are paying particular attention to drug discovery startups that can leverage generic LLM structures to discover new answers and approaches via the existing literature. One of these is Potato, which functions as a literature-based tool for designing and executing novel methodologies. Another is a generic LLM leveraging AI academic search engines to summarize new findings (Consensus NLP). From an investor's perspective, these emerging tools are far more resourceful for drug and therapeutic discovery than heavy investment in custom LLMs.
But the other area is arguably the more exciting one, given the novelty of its approach and its potential to find many more cures. And that is the investment in simulation/experimentation startups. I’ll bring us back briefly to physics, specifically a theory of quantum mechanics called perturbation theory. In short, it states: if there is a problem we cannot solve, we can instead look at the results of an experiment and make small changes to it. As the results change, we can estimate how the inputs change.
This is what excites VCs about generic LLMs now being powerful enough, and inexpensive enough, to tackle biological problems. We can run any number of experiments and simulations to find novel solutions, in that it is now economically feasible to do so. It is now possible to give the AI a problem statement and let it solve it for you (and with the advent of quantum computing, this will only get faster and cheaper).
Funding Advice For Drug Discovery Companies
To get a sense of the funding available for this kind of approach, just look at Periodic Labs, an AI materials science startup that secured $200 million for its systems that will not only learn from scientific literature but also will run its own physical experiments to generate new learnings (note: I am not an investor, just an admirer).
Outside of the physical sciences, though, there are companies that are using this theory to find more accurate protein and small molecule targets, including for cancer. That is the beauty of simulation; as the cost to run these digital experiments reduces, it becomes easier to predict which drug is the most durable.
As a cancer survivor, this is something I’ve followed closely for nearly two decades. The old approach was to look at a tumor and try different approaches until something worked, which, in the interim, can lead to the cancer becoming resistant. Now, we can have an AI evaluating the path forward for us, making thousands of incremental changes in their experiments until they find a more suitable target. Algen Biotechnologies and Ten63 Therapeutics are two examples of this. Algen recently signed a $555 million deal with AstraZeneca to co-develop novel therapeutic targets in immunology. These are the types of companies that investors are looking very closely at, and I expect they will get far more funding than any other older type of drug discovery play. Being a recovered quantum mechanics and cancer survivor, nothing would please me more than simulating a clinical trial to save money, time, and lives.
Conclusion
While the landscape of drug and therapeutics discovery investment is evolving, it remains a highly interesting and investable area for VCs. The shift in focus will be toward companies leveraging AI-assisted research tooling to drive novel discoveries, as well as toward quantum-based experimentation/simulation startups that are uniquely adept at posing problems to AI and utilizing the results to develop new targets. Both of these approaches — built on the foundation of increasingly powerful quantum-based simulators and affordable generic LLMs — are proving to be a more efficient and economically feasible path to success (and profitability) in the drug discovery sector, especially when it comes to attracting investment.
About The Author
Andy Tang is a venture capitalist and partner at Draper Associates and founding partner at Draper Dragon. With more than 20 years of venture investing, Andy has seeded more than 20 unicorns, with standouts in the biotech sector, including Colossal Biosciences, NewLimit, Algen Biotechnologies, Ten63 Therapeutics, and others.