Updated April 15, 2026
0:00 Welcome to Colaberry AI podcast brought to you by Colaberry AI Research Labs and Carl Foundation. Glad to be here for this one. It is, a really fascinating topic today. Yeah. I mean, think about it. 0:11 What if the next life saving medicine in your cabinet wasn't discovered by, you know, a scientist in a lab coat, but was actually computationally hallucinated by an artificial intelligence in milliseconds? Right. Which sounds like science fiction, but it is happening right now. Exactly. Today, we are unpacking a massive breaking financial and scientific report from CNBC. 0:32 This is from March 2026. Mhmm. And we're looking at a staggering $2,750,000,000 deal between The US pharmaceutical titan Eli Lilly and, a Hong Kong based AI startup called Insilico Medicine. A huge deal. Just a massive valuation. 0:49 Yeah. And whether you are tracking tech investments or maybe you're working in the biotech sector or just wondering how the compounds in your daily medication are actually synthesized, this deep dive is gonna reveal how generative AI is completely rewiring the pipeline of global drug discovery. It really is fundamentally changing the methods. Yeah. Okay. 1:07 Let's unpack this by starting right where the gravity is the heaviest, which is the financials. Oh, absolutely. The money always tells the story. Right. Because the architecture of this specific deal reveals exactly how much risk and, well, how much potential reward is currently tied up in AI drug discovery. 1:25 Yeah. The numbers here are highly instructive. I mean, they outline the industry's confidence in computational models while still, you know, acknowledging the immense unavoidable hurdles of the biological scientific method. So the headline number being broadcast everywhere is $2,750,000,000. Which is astronomical for a start up. 1:45 Right. But reading the fine print of the agreement, Insilico is actually getting $115,000,000 in liquid upfront cash. Exactly. The remainder of that multibillion dollar figure is entirely conditional. It is heavily subject to what the industry calls regulatory and commercial milestones, plus, royalties on future sales. 2:05 Right. So it's not a blank check. No. Not at all. So I have to ask. 2:09 I mean, is this $115,000,000 upfront actually a lot for a pharma giant like Eli Lilly, or is this just a calculated low risk down payment for access to cutting edge algorithms? Well, it's a brilliantly calculated hybrid of both, actually. Technically speaking, $115,000,000 in upfront capital is a very serious financial commitment for an unproven asset in a specific therapeutic area. Okay. So it's a lot of money. 2:37 Oh, yeah. However, you have to compare that to the baseline cost of taking a novel drug from concept to market. That typically exceeds $2,000,000,000 internally for a major pharma company. Right. So by structuring the deal around regulatory and commercial milestones, Eli Lilly is essentially paying strictly for empirical validation. 2:55 Yeah. It's like a high stakes video game progression system. A video game. Yeah. Like, they are holding the bulk of the money behind a series of incredibly difficult biological boss fights. 3:04 You don't unlock the next tier of billions until your therapeutic candidate survives the next level. That is actually a great way to put it. In the pharmaceutical industry, a regulatory milestone represents rigorous phased validation processes. The boss fights. Exactly. 3:21 It means the computationally generated molecule has successfully navigated phase one safety trials to prove it isn't toxic to humans. Right. Then it has to clear phase two efficacy trials to prove it actually interacts with the disease target. Then phase three, large scale population testing, and, ultimately, FDA or global equivalent regulatory approval. Wow. 3:43 Okay. So the bulk of that $2,750,000,000 is locked securely behind those specific biological hurdles. Eli Lilly mitigates the inherent risks of novel drug development, especially the notoriously high attrition rate in phase two trial. Right. Where everything usually failed. 3:58 Exactly. While Insilico gets an immediate substantial $115,000,000 runway to operate and scale their computational models. And it's worth noting that these two companies aren't just meeting for the first time either. The CNBC report highlights that they've been working together under an AI based software licensing agreement since 2023. Yeah. 4:16 That's a crucial detail. So Eli Lilly has had three years to essentially test drive Insilico's algorithms internally before making this multibillion dollar commitment. Which speaks volumes about the internal telemetry and data Eli Lilly must have seen during that probationary period. I mean, they must have seen something incredible. Right? 4:35 Oh, absolutely. You do not transition from a standard lower tier software licensing agreement to a 2,750,000,000 codevelopment pact unless the algorithmic output from that software has demonstrated profound replicable value in preclinical environments. Right. They clearly saw hit rates and lead interesting. Now that we know the financial architecture of what Eli Lilly paid for, we need to examine the actual technical output of Insilico's technology. 5:11 Yes. Let's get into the results. Because the report states that Insilico has developed at least 28 drugs using generative AI tools. Which is a staggering number. Yeah. 5:20 Yeah. And crucially, nearly half of these are already at the clinical stage. Andrew Adams, the group vice president of molecule discovery at Lilly, explicitly noted that this technology allows the exploration of novel mechanisms Yeah. And accelerates the identification of promising therapeutic candidates. Right. 5:39 And that phrase, novel mechanisms, is the technical crux of this entire endeavor. It represents a complete paradigm shift from discovering drugs to engineering them. Okay. But I wanna push back on this concept of generative AI right here because we really need to clarify the mechanics. Sure. 5:54 Yeah. We've all seen generative AI hallucinate, you know, fake legal cases or generate images of humans with six fingers and stuff like that. The hallucination problem. Right. So if the underlying technology is that prone to hallucination, how can Eli Lilly confidently bet billions of dollars that this algorithm won't computationally hallucinate a molecular structure that looks mathematically perfect on a screen Mhmm. 6:18 But is, I don't know, functionally useless or highly dangerous in the real world. Well, what's fascinating here is how generative AI in chemistry differs fundamentally from a large language model predicting text. Okay. When we talk about traditional drug discovery, finding a viable drug relies heavily on high throughput screening. Human chemists physically test thousands of existing chemical compounds against the target, say, a protein that triggers an inflammatory response. 6:45 It's trial and error. Exactly. It takes years of physical iteration to find a molecule that demonstrates adequate binding affinity. So traditional discovery is essentially like trying to guess a complex 100 character password by typing randomly on a keyboard for decades until you get a hit. That is a highly accurate way to visualize it. 7:06 Now apply generative algorithms to that process. Instead of random physical screening, the AI maps the vast multidimensional chemical space. Like it sees the whole board. Yeah. It is analyzing the underlying code of that metaphorical password system. 7:20 The algorithm predicts the precise atomic arrangement required to interact with the disease target, synthesizing an entirely new molecule from scratch. This is the novel mechanism of action that a human chemist might never have conceptualized. Furthermore, specialized chemical AI doesn't just guess a shape. It does more than that. Oh, yes. 7:39 It is trained to predict binding affinity, toxicity, and pharmacokinetic properties. What does that mean in plain English? It means it natively simulates how the human liver will metabolize the drug or how quickly it will clear from the bloodstream all computationally. So the algorithm is simulating the physics, the biology, and the chemistry before anyone even picks up a physical test tube. Exactly. 8:03 But put those numbers into context for us. The report notes 28 drugs developed with 14 in the clinical stage. What is the baseline velocity for a traditional biotech firm to achieve those metrics? Getting 14 computationally designed drugs into human clinical trials within a few years of operation is an unprecedented velocity. Really? 8:23 Unprecedented. Entirely. For a traditional biotech firm, advancing a single lead compound from target identification to a phase one clinical trial often takes four to six years. Just for one. Just for one. 8:35 The clinical stage is the ultimate crucible. The fact that half of Insilico's portfolio has reached this stage means their generative models aren't just producing theoretical digital anomalies. Right. They aren't six fingered hands. Exactly. 8:49 They are outputting physical therapeutic candidates deemed safe and promising enough by regulatory bodies to be administered to human patients. They are condensing years of physical discovery into months of computational synthesis. That is incredible. Yeah. But the AI alone clearly isn't enough, which leads us to a fascinating dynamic in this deal. 9:09 We have to transition to why Insilico, despite creating this massive algorithmic engine, still heavily relies on Eli Lilly's physical infrastructure. Yeah. This is the bottleneck. Right. In the report, Insilico's founder and CEO, Alex Zaveronkoff, openly admits that, quote, Lilly is better than us in some areas of AI. 9:28 That is remarkably candid admission from a tech founder, honestly. Right. And it points directly to the most significant bottleneck in modern biotechnology. Yeah. He specifically noted that Lilly has one person who has brought biology, chemistry, and automation under one roof. 9:42 And as a core component of this pact, Insilico is gonna join Lilly's Gateway Labs community for biotech development. Which is a massive physical operation. Right. To me, this relationship feels like Insilico is an elite visionary architect. They had the supercomputing clusters. 9:59 You know, they're drawing up flawless digital blueprints for these novel molecules. Uh-huh. But Eli Lilly is the hyper advanced, massively automated construction company. And the problem with being an architect is that a digital blueprint can't cure a disease. You need the physical automation and the raw materials to actually deal the skyscraper. 10:19 The physical infrastructure bottleneck is where so many digital biology startups fail Really? All all the time. The friction point is no longer discovering the molecule. Yeah. The friction point is taking those digital AI outputs and physically automating their creation and testing in a wet lab environment. 10:35 So explain that integration. How exactly does automation bridge the gap between a digital algorithm and a wet lab? Well, you can have a neural network generate a brilliant novel mechanism with perfect predicted pharmacokinetics. But a physical chemist still has to synthesize that chemical compound. They still have to make it. 10:53 Right. They have to run the in vitro biological assays. What Zivorankov is acknowledging regarding Eli Lilly is their mastery of automated robotic integration. Okay. So we're talking robots here. 11:05 Exactly. We're talking about advanced liquid handling robots, automated high throughput mass spectrometry, and flow chemistry systems that operate continuously. Wow. Eli Lilly has the automated pipelines to take in silico's digital molecules and rapidly translate them into physical testable reality at a scale an AI startup simply cannot match. But they're bringing biology, chemistry, and robotics under one roof to literally physically manufacture the AI's hallucination. 11:32 It demonstrates that in modern Ferma, raw computational power is utterly insufficient in isolation. You need the hardware. True clinical success requires synthesizing pure digital intelligence with highly automated physical chemistry. Insilico provides the algorithmic acceleration to identify the target, and Eli Lilly provides the automated physical infrastructure to execute the validation. Man, that makes the pairing incredibly symbiotic. 11:59 But this collaboration doesn't just span digital and physical spaces. It spans the entire globe. It really does. Right? So with the technical and financial pieces in place, we really need to zoom out and look at the geopolitical map of how this pipeline operates across borders because the geographical breakdown of this workflow is highly optimized and incredibly specific. 12:18 The spatial distribution of this research and development highlights how multinational biotech organizations view the global supply chain of the scientific method. Yeah. So according to the CNBC report, Insilico develops its artificial intelligence outside of China specifically. The AI development and training happens in Canada and The Middle East. Right. 12:38 However, the early preclinical drug development that is based on that AI research is actually conducted in China. Meanwhile, Eli Lilly is executing a massive broader strategy in that exact region. Their CEO, David a. Ricks, recently attended a high level forum in Beijing. Lilly has announced plans to invest $3,000,000,000 in China over the next decade. 13:01 Which is a huge commitment. And here is the kicker. They're committing that $3,000,000,000, even though slightly less than 3% of Eli Lilly's revenue, actually came from China last year. Right. Add to that, Insilico just went public in Hong Kong in December with their shares up more than 50% year to date. 13:19 So why split the pipeline geographically like this? I mean, if the raw computational code and the AI models are being written in Canada or The Middle East, why move the entire operation to China for the preclinical lab work? Well, if we connect this to the bigger picture, multinational corporations are ruthlessly optimizing different stages of the technical life cycle based on regional infrastructure advantages. Okay. So it's about playing to regional strengths. 13:43 Exactly. The computational development phase, the training of the generative AI tools, and the algorithmic refinement relies heavily on specific access to high performance computing clusters, specialized machine learning talent, and massive energy grids. Right. You need power and servers. Tons of it. 14:01 Regions in North America and The Middle East have aggressively positioned themselves as hubs for this dense digital infrastructure. So they handle the digital heavy lifting where the computing power is concentrated. But why the pivot to Asia for the preclinical phase? Because once your AI synthesizes a promising digital molecule, you enter the early preclinical drug development phase. This phase requires massive physical wet lab infrastructure. 14:27 Okay. Before a drug ever reaches a human phase one trial, it requires extensive toxicological assays running physical chemistry synthesis from milligram to kilogram scale and utilizing complex animal models. That sounds incredibly resource intensive. It is. Historically, China has developed a highly efficient, massively scaled infrastructure of contract research organizations or CROs. 14:50 The physical iteration required to validate a computationally generated molecule is incredibly demanding. So they have the facilities ready to go? Exactly. Moving that specific physical phase to a region where the laboratory infrastructure and specialized preclinical workforce are highly concentrated and scalable optimization. Wow. 15:11 So they are completely decoupling the creation of the algorithm from the physical validation of the science. The digital brain of the operation is crunching numbers in one hemisphere of the globe, while the automated physical hands conducting the mass scale lab work are in another. It is a globally optimized scientific supply chain. And regarding Eli Lilly's broader strategy, that $3,000,000,000 investment into China over the next decade, despite current revenue yields from that region being less than 3%, indicates long term strategic positioning. So it's not about selling pills there right now? 15:43 No. They're not investing $3,000,000,000 based on current consumer drug sales. They are investing to integrate deeply into that preclinical innovation and laboratory infrastructure ecosystem. They recognize that the future of drug development is a globally distributed network reliant on specialized regional nodes. And the financial markets are clearly responding to this distributed model. 16:06 Insilico's public listing in Hong Kong and their subsequent 50% stock surge year to date shows that institutional investors are highly confident in this hybrid pipeline. Very confident. Yeah. I mean, the convergence of Middle Eastern and Canadian computing, Chinese preclinical infrastructure, and American commercial and clinical automation is proving to be incredibly lucrative. It represents a highly coordinated borderless approach to overcoming the traditional limitations and, the traditional speed limits of pharmaceutical research. 16:36 So what does this all mean? If we bring this all the way back down to that medicine cabinet, we are looking at a fundamental rewiring of human health. The medicine you or your family might rely on in the next decade, the compound that might manage a chronic illness or target a previously untreatable biological condition will likely be the product of this exact global pipeline. That's the reality we're moving toward. It will be born as a digital concept in a Canadian or Middle Eastern algorithmic model. 17:06 It will be physically synthesized and preclinically tested in a massive Chinese laboratory. It will be automated and guided through clinical trials via an American Gateway Lab and funded by these complex multibillion dollar milestone agreements. We are no longer relying on a lone chemist doing trial and error. We are relying on a planetary scale computational machine. And that transition leaves us with a critical lingering technical challenge to consider. 17:32 Well, if generative AI successfully solves the target discovery bottleneck by mapping chemical space and advanced robotics solves the physical synthesis bottleneck by automating the wet lab, the ultimate bottleneck of drug discovery will shift. Shift to where? It will shift from machine driven discovery back to the unpredictable, highly complex reality of human biology. Wow. Right. 17:55 Because we aren't machines. Exactly. Computational models can simulate pharmacokinetics beautifully, but the human body is not a closed digital system. It is a chaotic biological environment. The next great frontier won't just be generating the molecule. 18:10 It will be generating computational models of human biology accurate enough to ensure those molecules models of human biology accurate enough to ensure those molecules succeed in real world clinical trials. Thank you for listening in. Subscribe and follow Colaberry on social media links in the description, and check out our website, www.colaberry.a I backslash podcast for more insights like this.
